Information

Measuring meta-intelligence

Measuring meta-intelligence


We are searching data for your request:

Forums and discussions:
Manuals and reference books:
Data from registers:
Wait the end of the search in all databases.
Upon completion, a link will appear to access the found materials.

Does anyone know of any research in measuring metacognitive abilities (i.e. metacomprehension, metamemory, meta-emotions(?), etc) in people? In other words, I'm wondering if the same way we have IQ tests (which I know is a whole other topic of debate), if there would be a way to measure "meta-IQ"?


A good overview on the topic is Metacognition: A Literature Review (E. R. Lai), page 27 has a section assessing the measures of Metacognition.

Flavell (1979) describes assessment tasks that asked children to study a set of items until they were sure they could remember them completely. Children were then tested on their ability to recall all the items.

Another common task was to read a set of written instructions and indicate any omissions, mistakes, or areas of ambiguity. Schneider (2008) observes that the most studied type of procedural metamemory is self-monitoring. Assessments designed to capture this ability include ease of learning judgments, judgments of learning, and feelings of knowing. For example, ease of learning judgments typically ask students to study a set of test materials for a short amount of time and then assess their abilities to remember the material. After the students are tested on the material, their performances are compared to their initial predictions.

Generally assessments focus on one specific area of Metacognition like Metamemory.

See also Development and Validation of an Objective Measure of Metacognition that assess four studies evaluating metacognitive abilities.

Searching around on Google Scholar finds a great deal of studies using and proposing various measures for Metacognition. There's certainly no "standard" like an IQ test, as Metacoginiton: A Literature Review notes, part of the challenge in studying and teaching Metacognitive abilities is due to the lack of focus on them in traditional school curriculum.


The myth that the new SAT correlates less with IQ than the old SAT

For some reason many people believe that the old SAT (pre-April 1995) was a much better measure of IQ than the new SAT (post-April 1995). I started believing this too when I found research showing high SAT people regressed much more to the mean on the new SAT than on the old SAT. However this evening I read that the correlation between the old SAT and the new SAT is virtually identical to parallel forms of the old SAT, so the trend I noticed was probably just statistical noise.

The reason people think the new SAT is less like an IQ test than the old SAT is that originally the SAT was explicitly intended to be like an IQ test, the hope being to give opportunity to bright people from socially deprived homes who wouldn’t be able to attend a good college without a test of natural ability. However as IQ tests became more and more politically incorrect, the test makers wanted to distance themselves from IQ, so the test became increasingly about what you learned in school, and less about abstract reasoning.

However what made the SAT correlate with IQ was never the fact that anyone was trying to create an IQ test, it was the fact that the skills you need in college (reading and math) are closely linked to cognitive ability.

A similar case was when David Wechsler created the WAIS explicitly to measure intelligence, but created the WIAT, specifically to measure academic achievement. I doubt he was trying to make the WIAT a measure of intelligence, since he had already created an IQ test the point of the WIAT must have been to show clinically significant differences between the two constructs, allowing the diagnosis of learning disabilities. And yet a recent study found nearly a 0.9 correlation between the two tests.

I don’t know what the general U.S. correlation between the SAT and IQ is because there’s never been (to my knowledge) a study that correlated the SAT with IQ in a sample of ALL Americans (not just the college bound elite). All the studies I’ve seen involved students at the same school, sometimes with correction for range restriction (which can be misleading because students at the same school are range restricted on more than just test scores). I have tried to estimate the correlation in the general U.S. population indirectly, by seeing how much samples of high SAT folks regress to the mean of all Americans, but the results have been inconsistent.

Some here believe that the correlation between IQ and SAT is so high that the SAT should be called an IQ test. However the brilliant Chris Langan understood the value of verbal precision, and argued that not even the Mega Test, on which he earned the World record should be called an IQ test. In a landmark 1998 article, Langan wrote:

To avoid the problem of rendering a specific a priori definition of what any such test will measure, it suffices to create a generic alternative description covering all tests which differ in structure or protocol from ordinary IQ tests, and for which high positive correlation with IQ has not yet been established. This new term must refer to a measurable quantity that is specific to the tests it describes, and that may or may not equate to that which is measured by garden variety IQ tests.


Saturday, October 4, 2008

From meta ability to meta Intelligence

A Question: Is Play (playful ) a way of being, and intelligence something else?

Being Logical is ”a way of being” and an intelligence. ( they are one and the same)
Being Musical is “a way of being” and an intelligence. ( they are one and the same)
Being Natural is “a way of being” and an intelligence. ( they are one and the same)
Being Playful is a way of being and an intelligence. (They are one and the same)

The key to any Intelligence is in having the “ability”. The intelligence called “Logical” is based on a combination of several abilities if not dozens. Which eventually develops, through a playing with various abilities, (combining various abilities) until they fuse together into a seemingly single unit which we call Logical intelligence.

Due to the works of Howard Gardner and others, the classical definition of Intelligence has now broken free from its restraining prison. And The definition of intelligence, has become more flexible more fluid and varied. The 8 or 9 Intelligences so far being defined is only the beginning, I predict it will be discovered that there are as many intelligences as there are “ways of being”, And the role of abilities (Human Universals) will play a major role, for all these multitudes of intelligences “are” based on “Having the ability” Play develops naturally this ability, one learns this ability simply because play is all about “how to Learn” not what to Learn, (Which we tend to overlook when we are studying Children and play).

Such Intelligences open up new ways of understanding ourselves and they blossom up in our human mind. They are spreading outward and growing like branches in a tree, from this tree I call Play, the Meta Intelligence, behind all these branches, together with these branches and changing these branches as they grow and mature.
Just like the neurons in our brain we are Playing, and neurons when stimulated create new branches and link up in new ways to find the best and most optimal way to create a” New Ability” Our brain is designed to play, we are born with a Meta Intelligence. You could say we are playing automatically. Even when we think we are not - our brain is still playing.

As children we are more conscious of how important play is. It is a human need to learn “How to Learn “ We are born with a Meta Intelligence, our brain which is a super computer a meta computer is designed to “Learn How to Learn” Learning how to Learn, is without a doubt, an Intelligence worth having. Loosing this Meta intelligence, is not good for the brain.
Research into Altzeimers Disease would benefit from the importance of Play to stimulate the neurons to grow again. The more we play the more neurons we create. It is possible to reverse the loss of neurons in the brain as we get older.

From the beginning we use our brain, (or our brain is using us) to learn the ability to survive what is happening and to develop abilities as quickly as possible to help prolong our survival. This raw survival instinct matures into a “mastering” of what is happening. This is an ongoing process throughout life. Learning means mastering. (I have learned to cycle, I now have mastered a bicycle).

Maintaining the optimal level of the brain, in my opinion, is clearly linked to maintaining, and consciously being open to play. To play in such a way as children do - with everything! To play (open to building and breaking down and rebuilding)with ideas, thoughts, feelings, language, images, systems,principles,models, senses, objects, knowledge, sciences, relationships, The happier we will be. As Gregory Chaitin replied to my letter where I asked him to look at the possibility that Mathematics is playing. He replied, the worlds top meta- mathematician , “Yes! I am also playing with Ideas.”

Any society in the future which places Play as a priority will be a prosperous and contented society
This has been predicted by “future researchers “worldwide it is known as the 5th Society

What Odense Council is doing now- “Consciously” and what Denmark for years has been doing (supporting the value of play and children) already proves the point.
Denmark is one of the richest nations in the world, and has been voted by Unesco every year for the past so many years as “the happiest nation on earth”.
The more conscious we become of play as the underlying key to enrich our lives, intellectually and emotionally, the more we will enrich society and the world around us.


Systems Perspective

Seed Knowledge &mdash The Systems Scaffold

Our models of how the world, other people, and ourselves work are based on having a built-in intuition about how systems work in general. Indeed, our entire knowledge base is organized around systemness. And when we learn, we are incorporating our perceptions into a framework of systemness because that is how our brains are wired.

Our minds naturally look for things like boundaries, wholeness (Gestalt), cause-effect relations, and a myriad of characteristics of systemness. We automatically attempt to find patterns in noisy data, and categorize patterns in hierarchical structures. Our brains process incoming perceptions so as to see the systemic nature of nature. We can't help it.

This is not surprising since through science, which is supposed to be objective, we have discovered that the world, the universe, is indeed comprised of systems and systems of systems. We find causal relations among system components everywhere we look. In fact, the drive behind the scientific approach to knowing is that when we find phenomena that are not previously categorized, for which a pattern of organization and causal relations have not been identified, then we are essentially forced to look for these things. It is as if evolution predisposes us to see systemness because everywhere there are systems. We are systems. And we are subsystems of larger meta-systems.

This propensity to see systemness, or discover it if we don't immediately see it, is a fundamental organizing principle which our brains are constrained to use to learn about the world. The generic system is a kind of seed structure upon which we map percepts in order to have a means of organizing our knowledge.

Every knowledge construction requires some kind of template upon which to organize new knowledge. The mind is not a blank slate (Pinker, 2002). The brain itself is organized in such a way that we begin our construction of knowledge with the aid of built-in biases for key perceptions and organization of those into early conceptual structures, like categorization and hierarchies of types. Thus as we grow and develop our models of the world and ourselves, we start with a foundation of generic systemness and a scaffolding that provides a basic shape to how we understand the world. Literally, we can't see it any other way. To that structure we start fitting our experiences into place. It is probably more a matter of jostling the bits and pieces around until they 'fit' into the scaffolding and among other bits and pieces already integrated. It is a stochastic process. Some bits won't fit anywhere in the edifice and so get dropped even if they should legitimately be part of the knowledge base. Fortunately, these bits are likely to be encountered later again so they have more than one opportunity to get incorporated.

The point is that knowledge is built upon prior existing knowledge and the ultimate seed knowledge is provided by evolution in the form of an ability to model systems.

In the next part I will provide some ideas about how the brain actually accomplishes this feat. For now all you need recognize is that the generic system can be represented as a network or, in mathematics, a flow graph. Figure 1, above, is such a network and it represents the system I have been describing in words. The dashed line circumscribing sapience demarcates the system of interest and the other entities provide inputs and take outputs from that system. Figure 5 shows a generic system with the expected kinds of components. The system of interest has a boundary of some kind, it has component subsystems between which flows and associations occur in an internal network (not shown). It receives inputs of energy, material, and messages from environmental sources and it produces outputs of similar kinds that flow to environmental sinks. The arrows from and to environmental entities may also be reciprocal linkages with entities rather than explicit flows. This representation is kept simple for demonstration purposes.

Figure 5. A generic system has all of the features/attributes of a basic system in generalized form. Neural networks can encode the various elements and their generic interactions. The human brain has the ability to make copies of this generic model and then learn the particular features of each kind of component.

A generic system is encoded into the brain as a template for the learning of all real systems/objects that the brain will encounter in the future. Systems learning entails making a copy of the generic template somewhere in the cortex (probably in the frontal-parietal areas) and then begriming to link up specific perceptual and other conceptual features to the copy as it becomes particularized to the real system being learned. In Part 4 I will revisit this in terms of plausible neural circuits. The point here is that our brains are wired to look for subsystems and boundaries and connections, etc. as we construct a larger network of particulars. Figure 6 is meant to capture some of this. Starting with a fixed template copy, the brain learns the particulars of a system by identifying the features and attributes that should be attached to the model of the real system while also expanding and modifying some of the details. For example, the real system being modeled will have many more component subsystems with particular linkages. Characteristics, such as the nature of the boundary, may be modified as well.

Figure 6. A particular (real) system is learned by attaching perceptual and conceptual features to the template copy and expanding where needed, e.g. in the number of subsystems and their linkages. This is the basis for humans learning what is in the world and how things, and the world, work.

Since systems are subsystems of larger meta-systems, and are, themselves composed of subsystems, this copying-modifying procedure works in both the direction of the larger and the smaller. The brain can build a model of the meta-system by starting with an already built subsystem (now treated as a component) and situating it within the larger system. Note that the entities identified as sources and sinks can now be modeled in their own rights and their linkages constitute the more complete model of the meta-system.

Working from smaller systems to larger meta-systems is a synthesis/integration process. Working from a system inward to model the component subsystems as systems in their own right is analytical reduction. The brain automatically works at doing both of these. The former is driven by a need to understand the context of a particular system and leads to a grasp of a larger world. The latter is driven by the need to understand how a particular system works. Both of these processes are aimed at providing the brain with a basis for anticipating the future behavior of the systems it observes (see below).

Sapient Systems Thinking

As indicated above, one of the characteristics of judgment is in guiding what should be learned. We can now see that the systems bias is part of the basis for this. As our internal models of world systems improve over time and experience, our judgment derived from them can better guide the intelligence machinery in attending to perceptions that help improve the systems models. This is low-level judgment at work, the kind our biological ancestors had evolved. What makes for sapient systems thinking, and judgment so informed, is the role of strategic (long-term planning, see below) thinking, conscious reflection on knowledge being constructed and editing knowledge as needed (including editing plans for acquiring knowledge in the future). Such judgments guide which systems need to be learned.

This is a huge subject, of course, and will need much explication, beyond the scope of this work. One succinct way of looking at this is that sapience expands the role of judgment in guiding future learning and refines the systemic nature of what is attended to in that future time. As noted above, the drives that produce the learning of particular systems causes us to explore both inward (reductionist analysis) and outward (synthesis and integration). The more sapient mind is equally interested in both directions. But all too often most humans run into limitations on what they are able to do in terms of expanding their models and understandings both inward and outward. This is a scope issue relating to the same problem as mentioned above for judgment. Most humans have limited curiosity. They are not driven past a certain point, reached about middle age, I suspect. As children, while the brain is still in rapid development, curiosity directed at learning the smallest details and the largest relationships is at a maximum. It is hard to say for certain when in a person's life the drive to curiosity starts to diminish. It is hard to say why it does. One can imagine a storage limit, but as I have argued, this seems less likely given the way the brain encodes systems by reusing features that are common to many systems and simply organizing appropriate linkages (see Part 4). As an aside, I do posit that our modern education system may have a great deal to do with damping down children's enthusiasm as it attempts to force-feed knowledge, which is generally not systemic in nature, into the minds of young people. By the time they graduate from high school (if they graduate) they have been told, in so many words, that the world contains many different disparate bodies of knowledge and they must choose one such body to learn well so that they can do a good job in the marketplace. It is hard to imagine how this message can promote curiosity and a love for learning.

But I also suspect that a continuing life-long drive to curiosity depends on the level of sapience in the individual. With lower sapience comes a limited scope and time scale for thinking. People learn just what they need to know to get by in the world they are used to. They do not, in general, expect that world to change very much. They expect whatever trends exist to continue on into the future. So at some point they are no longer concerned with expanding their scope (learning the yet larger meta-system in which they are embedded) and they feel competent knowing &lsquoenough&rsquo about the daily systems they deal with that they do not need to know how they work inside. Lower sapience goes along with a limited world view.

Sapience involves intentional model building such that one becomes more effective in problem solving in an ever wider scope as experience grows. One attribute of a wise person is grasping the interconnections between elements of a complex system, especially a social organization. Applying systems thinking to such organizations increases the probability of finding solutions that will work. And wise people seem to continue learning their whole lives.


Very Computer

I am currently experimenting with some reinforcement learning (RL)
techniques for a small mobile robot moving about in a cluttered
environment. My aim is to have the 'bot moving happily about,
not bumping into obstacles, whilst navigating towards a goal.

In order to improve the robot's performance (in avoiding obstacles
and making progress towards a goal) I would like to be able to
have the reinforcement learning system try to predict the future
outcome of an action - in terms of the total reinforcement recieved
over a number of future time steps - and then incorperate this
prediction into its decision making process.

Ideally, I would like the learning of the prediction system to be
continuous. I want to avoid having to store the state/reinforcement
of the system over the number of prediction time steps required
(in order to retrospectively calculate the prediction error) if possible.

There is also the problem of how far ahead the system should attempt
to predict. I suspect that a good strategy might be to have this
'time window' adaptable, so that in some circumstances the 'bot
might only want to estimate a few steps ahead whereas in others a
more long-term vision might be beneficial.

I'm working with simulations only at the moment, but would like to
transfer the RL system to a mobile robot at a later stage.

I would be grateful to hear from anyone out there who has experience
with this kind of future predicting/modeling RL technique, or could
perhaps point me to some interesting references.

___________________________________________________________________________ _
Bob Mottram "Robots may move suddenly and without warning"

Reinforcement learning - predicting reinforcement

by Magnus Sandber » Sat, 23 Sep 1995 04:00:00

>I am currently experimenting with some reinforcement learning (RL)
>techniques for a small mobile robot moving about in a cluttered
>environment. My aim is to have the 'bot moving happily about,
>not bumping into obstacles, whilst navigating towards a goal.

>In order to improve the robot's performance (in avoiding obstacles
>and making progress towards a goal) I would like to be able to
>have the reinforcement learning system try to predict the future
>outcome of an action - in terms of the total reinforcement recieved
>over a number of future time steps - and then incorperate this
>prediction into its decision making process.

Quote: >Ideally, I would like the learning of the prediction system to be
>continuous. I want to avoid having to store the state/reinforcement
>of the system over the number of prediction time steps required
>(in order to retrospectively calculate the prediction error) if possible.

I hope my reasoning is understandable. And, I'd be delighted if anyone could
find a flaw in it because I don't like the implications. After all, humans
learn by reinforcement, and though the brain may be vastly more complex and
better designed from the start, the basic principle should still be the same.
We evaluate our successes and mistakes and learn from it, but the evaluation
process can also be modified if we find that it's not delivering what it
should. The evaluation of the evaluation process is, of course, also an
intelligent process. At first sight this seems to require a meta-intelligence
responsible for training our mind, and a meta-meta-intelligence to train the
meta-intelligence etc. In reality there has to be some kind of bootstrap effect
in that everything reinforces everything. But exactly how this works and
whether it can be copied to more modest AI scenarios is a big question mark to
me. Perhaps it requires a significant level of intelligence to start with.
There is, after all, a good deal of work behind a newborn baby's brain.

For the time being I've given up on finding a method of the kind you're asking
for. My current approach is to design a learning system that can be trained
with delay without too much overhead.

Quote: >I would be grateful to hear from anyone out there who has experience
>with this kind of future predicting/modeling RL technique, or could
>perhaps point me to some interesting references.

Reinforcement learning - predicting reinforcement

by Chris Connol » Tue, 26 Sep 1995 04:00:00

>I am currently experimenting with some reinforcement learning (RL)
>techniques for a small mobile robot moving about in a cluttered
>environment. My aim is to have the 'bot moving happily about,
>not bumping into obstacles, whilst navigating towards a goal.

Quote: >You're missing a piece of the puzzle. There is a whole gosh darn
>research area in "adaptive critics" that deals with this exact
>problem. Try looking up anything by Andy Barto, Richard Sutton, or
>Paul Werbos. Or pick up any recent NIPS proceedings.

o Peter Doyle and J. Laurie Snell (1984), "Random Walks and Electric
Networks", American Mathematical Society Carus Monographs in
Mathematics. (VERY clear, good reading)

o Kemeny, Snell and Knapp (1976) "Denumerable Markov Chains",
v. 40, Graduate Texts in Mathematics, Springer-Verlag.

Also, Jonathon Bachrach's Ph.D. thesis from UMass (CS Dept. Technical

with using RL for robot navigation -- some of this has also appeared
in NIPS 2 (Mozer and Bachrach?) and 3.

For an approach from the standpoint of harmonic potentials, there are
a few papers you can find in the UMass robotics lab:

In particular, the following tries to make a couple of connections:

o C. I. Connolly (1994), Harmonic functions and Collision
Probabilities, IEEE Conf. on Robotics and Automation, pp 3015.

SRI International Phone: (415) 859-5022
333 Ravenswood Ave. Fax: (415) 859-3735
Menlo Park, CA 94025 WWW: http://www.ai.sri.com/

SRI International Phone: (415) 859-5022
333 Ravenswood Ave. Fax: (415) 859-3735
Menlo Park, CA 94025 WWW: http://www.ai.sri.com/

Reinforcement learning - predicting reinforcement

by Shez » Fri, 29 Sep 1995 04:00:00

>. humans
>learn by reinforcement, and though the brain may be vastly more complex and
>better designed from the start, the basic principle should still be the same.
>We evaluate our successes and mistakes and learn from it, but the evaluation
>process can also be modified if we find that it's not delivering what it
>should. The evaluation of the evaluation process is, of course, also an
>intelligent process. At first sight this seems to require a meta-intelligence
>responsible for training our mind, and a meta-meta-intelligence to train the
>meta-intelligence etc.
> .

Regarding evaluation of high level (conscious) evaluation processes, where
they are unable to evaluate themselves I think we tend to get stuck - hence
the invention of the therapist. :-)

Obviously there are many areas where our learning mechanisms are
phenomenally successful, eg. being able to see things, acquiring language,
etc., but as you note yourself, these tend to be areas where we're born with
plenty of bootstrap code. As we ascend the levels of thinking, things get
progressively shakier. You mention meta & meta-meta-intelligences: for this
we tend to rely a great deal on interaction with external intelligent agents
- parents and teachers!

The equivalent to the parent, teacher, and, most particularly, the therapist
in AI is of course the programmer. And it is perhaps pertinent to note here
that some pretty successful real-life therapists* decided to call their
activities 'neuro-linguistic programming' (NLP). Their work has been done
entirely at a meta level - they help people generate new strategies to
resolve their problems, often without having to know anything at all about
the nature of the problem that the person has. *[R.Bandler & J.Grinder]

One interesting thing that the inventers of NLP noticed was that there are
plenty of people who not only know that they are failing to do something
right, but who also know exactly what they are doing wrong and why. Yet
despite having this knowledge they are often unable to revise their
behaviour without external help.

Consider for example someone with a phobia (be it spiders, enclosed spaces,
whatever). Here they have learned an inappropriate response to some stimulus,
and although they are fully aware of the problem and what behaviour would be
more appropriate, they need special coaching to implement the changes required.

We might compare this to a robot which keeps banging against a particular
shaped obstacle, and although it learns to recognise that obstacle, is unable
to generate the sequence of movements required to circumnavigate it.
Although we would hope to have it evaluate the failing in its strategy and
learn another way of tackling the problem, we cannot necessarily hold humans
up as proof that such mechanisms are possible.

Incidentally, people should not be misled, by the therapy/mental health
examples, into thinking that a special type of learning problem is involved.
The above comments apply equally well to pretty much any situation where a
failure to learn occurs, eg. in school subjects. People often think that
some facet of cognition they have come across applies only to the narrow
area where they found it, but in my experience cognitive procedures are
surprisingly generalisable.

[Disclaimer - GA & NN are not my field. I'm a cognitive scientist.]

Reinforcement learning - predicting reinforcement

by Steve Bu » Sun, 01 Oct 1995 04:00:00

>>I am currently experimenting with some reinforcement learning (RL)
>>techniques for a small mobile robot moving about in a cluttered
>>environment. My aim is to have the 'bot moving happily about,
>>not bumping into obstacles, whilst navigating towards a goal.

Quote: >>Ideally, I would like the learning of the prediction system to be
>>continuous. I want to avoid having to store the state/reinforcement
>>of the system over the number of prediction time steps required
>>(in order to retrospectively calculate the prediction error) if possible.
>I'm not sure if this can be done. In order to make a decent prediction of the
>future without knowing anything else than the present state and past success of
>the robot, the evaluation system would have to be far more intelligent than the
>decision making process itself. No simple function/extrapolation would do. So
>let's design another learning system whose task is to determine (without delay)
>the success of each of the robot's moves. Let's say we pick some kind of neural
>network. Then the network, too, needs to be trained, but it can't be trained
>continously (since that would require us to know at once how successful the
>latest move was, which is what we are looking for, in order to evaluate the
>net's performance). We thus choose to train the network with a delay of ten
>(say) steps. Knowing the states of the robot between steps [t,t+10] will let us
>make a decent evaluation of the robot's move at step t and thus permit us to
>evaluate the net's output at step t. But in order to do this we have to have
>saved the complete state of the network at step t, and so nothing is gained.

A human baby comes with certain "burned in" routines ("rooting" for a
* and suckling, is an example of what consitutes a more complex
behaiviour, blinking is an example of a rather simple reflex and
digestion is an example of a bio-chemical reaction. With the addition
of a rather "blank" but potentially utile (LOTS of potential
connections) neural network we have a baby as it comes "from the
manufacturer"

This baby will shortly be able to make learned decisions that involve
a very small number of steps. Example: If I need/want something and if
I make a loud crying sound, my parent will appear, figure out what I
need/want and give it to me. Learned decisions that require many steps
or abstract concepts are beyond the baby at this stage no matter how
beneficial they might be for the baby. Example: If I take the money
grandpa and grandpa sent when I was born and invest it in a mutual
fund that buys hi-tech stocks, I won't need to worry about food when I
retire. This type of behaviour is only exhibited as the organism
matures (if ever).

Sorry this is so long, but here's where I'm going with this.

The little robots, when they start out, could use a similar structure:

An electro-chemical system - they don't need to "know" how to get
electricity from their batteries and use it, it just happens.

Simple reflexes - blink, jerk back from hazards etc.

simple behaviours that ensure survival - hunger/food: how "hungry" am
I? what is the most direct route to the nearst food source? I have to
go to the nearest food source while I still have enough energy. This
could be coded as ROM and may or may not use a dedicated processor (a
sub-conscious/background task if you will)

A neural network, possibly seeded with simple actions - obstacle
behaviours: go around it, push it, pick it up, go over it. At this
point, the number of steps from the action to the evaluation of the
"goodness" of the action should be very small, 1-2 steps. As the robot
becomes more proficient it could try to mix actions together and take
more steps to evaluate "goodness". Example of an intermediate stage -
pick up object A, place it next to object B as a ramp, go up ramp,
check out the top of object B. Example of an advanced stage: push
object B next to a much taller object you can't push (a tabletop or
counter) pick up object A, build a ramp, get on top of object B, pick
up A, build a ramp get on top of the counter, find a new food source
(power outlet)

I hope this made sense. I'm just a layman blessed with internet access
so if I've misused any terms please forgive me. I'll go back to
lurking now

Reinforcement learning - predicting reinforcement

by Marty Stonem » Mon, 02 Oct 1995 04:00:00

[Most of Shez article snipped]

: The above comments apply equally well to pretty much any situation where a
: failure to learn occurs, eg. in school subjects. People often think that
: some facet of cognition they have come across applies only to the narrow
: area where they found it, but in my experience cognitive procedures are
: surprisingly generalisable.

I fail to see what procedures are generalizable, e.g., for learning in
school among the following:
1. learning to play a sport, e.g., baseball
2. learning to quote a Shakespeare sonnet
3. learning to tell the difference between species of birds
4. learning to avoid stubbing your toe on your desk legs.
5. learning to do calculus
etc.

I would appreciate a clue.
Thanks.

Reinforcement learning - predicting reinforcement

by Michael Green » Tue, 03 Oct 1995 04:00:00

>I fail to see what procedures are generalizable, e.g., for learning in
>school among the following:
>1. learning to play a sport, e.g., baseball
>2. learning to quote a Shakespeare sonnet
>3. learning to tell the difference between species of birds
>4. learning to avoid stubbing your toe on your desk legs.
>5. learning to do calculus
> etc.

* a bit of observing or thinking,
* attempting the task
* examining where you erred
* compensating for the error
* Repeat as necessary.

All of the 5 of the tasks you've cited call for those steps. Learning
*anything* exercises those functions and reinforces the brains ability
to learn something else. My son's 2nd grade teacher had the kids
memorizing a lot of poetry principally to exercise their memories.

Reinforcement learning - predicting reinforcement

by Shez » Wed, 04 Oct 1995 04:00:00

> Shez (S. @sv.span.com) wrote:
> : . People often think that some facet of cognition they have come
> : across applies only to the narrow area where they found it, but in my
> : experience cognitive procedures are surprisingly generalisable.

> I fail to see what procedures are generalizable, e.g., for learning in
> school among the following:
> 1. learning to play a sport, e.g., baseball
> 2. learning to quote a Shakespeare sonnet
> 3. learning to tell the difference between species of birds
> 4. learning to avoid stubbing your toe on your desk legs.
> 5. learning to do calculus

One thing I had particularly in mind was Kuhn's theory that knowledge
acquisition (in other words, learning) often progresses in 'revolutionary'
paradigm shifts.

He believed that it applied only to scientific knowledge, whereas I think it
applies to all knowledge, including perception. His book has an example
which purports to show that perceptual paradigm shifts are reversible whilst
scientific ones are not, but his example is misconceived. (I'm not up to
date on the literature so it's likely that others have said this too by now
(I first noted it in 1987)).

So we might say that something nearly all learning processes have in common
is the formation and testing of successive hypotheses about the data. Under
good conditions where data is of high quality and contains little that is
novel, we might only need to form one hypothesis. Newton's characterisation
of the universe served for hundreds of years. In contrast, when trying to
recognise a shape in a darkened room our initial hypothesis might last for
less than one second before we change our mind and decide that it's
something else.

But in each case the process involves forming an model which seems to fit
the data, and then as additional data accumulates which cannot be fitted to
the original model, we first ignore the discrepancy as noise, then recognise
that something is wrong, but stay with it, then build new models, and when
one offers a better fit a paradigm shift occurs as we discard the old one
and use the new one.

The striking thing here is that a single process model serves both something
that happens for one person at one time, and something that happens socially
for a community comprising successive generations of people. To me this
constitutes a surprising generalisation. (Well it was an exciting discovery
for me!)

However, let's try and find some processes which your list of tasks have in
common. (I expect others will have a go at this too - it's an interesting
challenge!)

Looking at the list, some obvious similarities spring to mind between 1 and
4 (physical coordination), also between 2,3,5 (a degree of rote learning),
and especially between 3 & 5 (classification) but I suspect this is not what
you are interested in. (I also note that item 1 actually includes at least
two separate learning tasks - learning the physical skills, and learning the
rules of the game.)

But let's try and find a process that all five have in common. The most
basic task would appear to be the sonnet, which involves learning a sequence
of words. If all the tasks actually do have something in common we might
expect it to be most easily identifiable here.

The general hypothesis building approach already described almost certainly
applies. In learning a list of words we associate them together in some way.
This is easiest if the words make sense, as we are offered an easily
identifiable superstructure which serves to glue the sequence of words
together in memory. (In this context I mean "easy" in the sense that we
already have the skill of making sense of sentences). In contrast, learning
an arbitrary sequence of words requires us to invent a linking structure -
I'm sure we are all familiar with the memorising tricks that are employed in
such cases, such as attaching the words to a meaningful structure of
imagery, or finding semantic or phonetic relationships between the words,
etc etc. (Computers obviously have an enormous advantage to us here as they
are specifically designed to store and process sequential material directly).

When learning a sonnet, we form some unconscious semantic structure from
which we can "read back" the words. Repetition allows this to be refined
until it has a form suited to accurate recall of the exact words as well as
the meaning. Indeed, the bad thing about rote learning is that the structures
we form tend to be devised to recall the words at the expense of an accurate
grasp of the meaning.

There is unlikely to be any wholesale paradigm shift in learning a sonnet,
but the structure will certainly undergo many revisions before we become
word perfect. It is a commonplace observation that in learning their lines
actors often get a mental block over specific sentences and I expect we have
all seen those out-takes on TV where they get a sentence wrong several times
in a row. Each repetition of the error reinforces it. Overcoming it can be
done be creating a new fragment of the associative structure, linking it in
at the appropriate point, and detaching the erroneous fragment. (The final
stage tends to happen spontaneously through preferential reinforcement of
the correct form.) This is exactly how phobias are cured in NLP. The main
difference in something like learning a sonnet is that the revision of the
cognitive structure tends to be muddled through at a subconscious level,
although there is no reason why we shouldn't consciously examine those
phrases we find hard to learn and discover some new meaning in them from
which the correct words can be extracted more easily.

At this stage I am tempted to leave applying the above learning mechanism
to the other four tasks as an exercise for the reader, as this posting is
already two pages long! Certainly I don't feel the need to demonstrate it
applies to recognition of a species of bird or differential equation. The
learning of physical skills is worth a brief look though. If you stub your
toe a lot, your mental image of the size of your foot and/or the amount of
physical movement resulting from a given instruction to your muscles is
obviously an underestimate and needs revising upwards. Either that or your
mental model of the position of your desk's legs is wrong, or perhaps you
have developed a bad habit of swinging your legs too much as you sit,
a habit presumably formed whilst sitting at a different desk. In all these
cases we would normally muddle through and overcome the problem, but again
we can model it explicitly.

There are many ways that we might cease stubbing our toes on our desk, not
all of which might necessarily be described as "learning". For instance, if
it comes from a bad habit of swinging ones legs whilst sitting at the desk,
it might be a conditioned reflex that goes away because the triggering
stimulus is removed or we get into the habit of working instead of looking
out the window during class. (In such a case no link to learning a sonnet
springs to mind.)

If we're talking about stubbing them whilst walking about though, we might
cure it by paying more attention to our feet until the problem goes away.
This could involve a mechanism similar to the sonnet example: we form a new
and more accurate fragment of the mental model we have for our body's size
and position and how it moves, and after attending to it consciously for a
while the fragment becomes incorporated in our unconscious body model.

I see no reason why these mechanisms should be the only possible way of
learning something, but they are *a* way. Going back to the original posting
that I responded to, the issue was the evaluation of a learning mechanism
with a view to improving it when it failed. It seems to me that the
procedure at the meta-level is similar to that of the learning examples
given here - when one thing doesn't work, try something different! The
manner of producing permuatations or alternative models will depend on the
task, but that doesn't mean to say the process is necessarily different,
it's just that the data types being manipulated have different modalities.

We might draw a parallel with object oriented programming. Different types
of Object require their own implentations of the Methods used to manipulate
them, but the method names and arguments often stay the same, and a calling
program needn't know anything about the difference between say, the display
method for a video object and the display method for a text object, but can
manipulate both objects using the same algorithm. Similarly in cognition, we
needn't start from scratch in approaching a new domain, provided of course
that our algorithms are abstracted to suitable levels and we have not
jumbled high level tasks in amongst low level implementation of routines
that apply only to one domain.

This of course is a primary drawback of neural nets, where the model is not
explicit. Unless we partition the problem into various levels before
presenting it, the solution net is unlikely to be of any use to another
problem. For myself I find neural nets of limited interest, as although they
may solve problems, and show how our intelligence may be implemented, they
cast little or no light on the problem domains themselves, so we are none
the wiser as to how applicable a particular neural net solution might be to
another problem domain. (I am not an expert on neural nets, but it seems to
me that this is inherent in their nature - comments please!)

Anyway, this is all I have time to write today! I welcome discussion on any
of these issues, especially as I'm just finding my feet again in cognitive
science after several years of having .


The Positive Manifold: Reactive Control in Fluid Intelligence?

What neural mechanisms underlie "fluid intelligence," the ability to reason and solve novel problems? This is the question addressed by Gray et al. in Nature Neuroscience. The authors begin by suggesting that fluid intelligence (aka, gF) is related to both attentional control and active maintenance of information in the face of ongoing processing (i.e., working memory). Each of these concepts, in turn, has been associated with the functioning of the lateral prefrontal cortex - a region that has been massively expanded in humans compared to even our closest evolutionary relatives.

To confirm that individual differences in gF are related to prefrontal functioning, Gray et al. measured performance both on a standard gF task (Raven's matrices) as well as on a standard test of prefrontal function from cognitive neuroscience: the 3-back task. In Raven's matrices, subjects are required to pick which of several stimuli "fits" as the final item in a matrix of abstract patterns (see an example.) In contrast, the 3-back task provides subjects with a series of stimuli, presented sequentially, and requires that they respond if the current stimulus matches the one presented 3 items previously (i.e., to respond yes to the second "B" in a sequence like "A X B Y X B X A"). This task is performed in an ongoing fashion, such that subjects must constantly displace the third item in memory with the second, and update memory with the current item. (If you can't tell from my description, this is an extremely difficult task).

Intuitively, one might not expect a strong relationship between these tasks: 3-back relies heavily on memory, whereas all the relevant stimuli are simultaneously present in Raven's. Conversely, Raven's requires abstract and somewhat "analogical" reasoning, but 3-back requires only rote memorization. So these tasks seem to require very different computations - an individual's performance might be expected vary substantially between them.

On the other hand, there's the concept of the "positive manifold": performance on any two reliably-measured tasks is positively correlated (indeed, this is part of the basis for the concept of "general intelligence"). Surprisingly, the positive manifold may apply to neuroscience data as well: despite the possibility that different neural regions would underlie performance on these two very different tasks, certain regions in prefrontal cortex reliably mediate the behavioral correlations between these tasks.

To demonstrate this surprising fact, the authors distinguished between 3-back performance on lure trials (where the target item had occurred on perhaps the 2nd or 4th previous trial, but not the 3rd back) and those on non-lure trials (where target items occured on 1 trial ago, or more than 5 trials ago). Lure trials actually seemed more sensitive to performance than target trials (in which an item was actually presented 3 trials ago) insofar as accuracy was just as bad as target trials, but RTs were even longer.

Estimates of gF were positively correlated with accuracy on all trials types, but was most strongly related with lure trial performance: taking into account accuracy on non-lure trials or accuracy on target trials, gF still showed a significant relationship with lure trials. Activity in lateral PFC, anterior cingulate, and lateral cerebellum all predicted accuracy, and activity in these regions during lure trials overlapped with up to 92% of the shared variance between gF and 3-back performance. In contrast, this pattern was much more subtle on both target and non-lure trials.

Interestingly, the magnitude of sustained activation (thought to subserve active maintenance) was correlated with 3-back accuracy but not with gF ability. This finding is somewhat at odds with accounts that put "vanilla" active maintenance at the center of intelligence and executive control - other processes (such as those recall and discrimination processes involved in lure trials) appear to more strongly manifest the variance shared with gF. This would seem to have applications to the notion of "reactive control" and "secondary memory" as discussed recently in the literature - future work will need to clarify the relationships between these constructs.

The authors note that grey matter volume in lateral prefrontal cortex is under "significant" genetic control, suggesting that perhaps gF is itself largely heritable. Word has it that a new (but still under review) publication is showing the heritability of gF as being close to 1. In contrast, the authors here suggest that gF is probably not entirely heritable, and that a better understanding of individual differences in the neural correlates of gF could contribute to future attempts at enhancing fluid intelligence.

More like this

You can't see the wood for the trees..smile
Intelligence, intelligent, from Latin intelligere understand (INTER+legere gather, pick out).
Clever, Adroit, dexterous, skilful, talented.
What do you mean by intelligence? What is the difference between being intelligent and being clever? Can a person be intelligent without being clever and visa versa? Hear are some definitions of intelligence.
the ability to comprehend to understand and profit from experience
����
Intelligence is a general mental capability that involves the ability to reason, plan, solve problems, think abstractly, comprehend ideas and language, and learn. In psychology, the study of intelligence is related to the study of personality but is not the same as creativity, personality, character, or wisdom.
����..
Adaptability to a new environment or to changes in the current environment
Capacity for knowledge and the ability to acquire it
Capacity for reason and abstract thought
Ability to comprehend relationships
Ability to evaluate and judge
Capacity for original and productive thought

Of the preceding, the first definition with some addition comes closest to a good definition. That is the continuous desire and ability to comprehend and understand by observation and logical reasoning. Who is intelligent by this definition? The current IQ test has been used (with modifications) for over 100 years. The average IQ score is 100. An IQ score of 160 places you into the genius category and a score of >200 is categorised as unmeasurable genius. Computers are becoming ever more powerful and sophisticated. Is a computer intelligent? No, it will never be able to comprehend and understand. It might appear to do so but that will be an illusion. It will only ever be a programmed machine. Even if it is programmed to generate its own coding it will do so as an uncomprehending programmed manner. It will never think (I think, therefore I am). It will only appear to be as clever as the men or women that programmed it. If you do not comprehend and understand this then you are not intelligent (having understanding). There are young children (seven to ten years old) who have genius IQs of 160-170. They have above normal learning abilities and talents. However, like an autistic savant they are clever not intelligent. They see the world in a simplistic child like way. One of them may write music and play the violin to a professional standard. Another might be able to do complex mathematical problems. However, they do not have understanding. You would not expect complex philosophical insight and understanding from any one of them. The IQ test should be called the CQ test (cleverness quotient) for it has everything to do with measuring cleverness and nothing to do with measuring intelligence.
We live on the thin crust of a sphere of molten rock and iron, which is nearly 8,000 miles in diameter. It moves at 18 miles each second through the black vacuum of space circling a star (the sun) which is 93 million miles away. The sphere has a thin layer of breathable atmosphere (less than 7 miles) and if it were not for the magnetic field generated by its iron core the radiation from the sun would be lethal to life. Only an intelligent person will comprehend and understand how strange, grotesque and bizarre this is.
Many people use homeopathic medicine. Substances are diluted down to such an extreme level that nothing of the original remains. It is obvious that if nothing of the original source remains its only action on the body can be that of a placebo. An intelligent person will understand this fact but a clever person with the IQ of a genius might not. It is more than a 50% certainty that your death will not be pleasant. The percentage chance of you dieing peacefully in a bed surrounded by loved ones is not good. The reality is fear of dying, cancer, Alzheimer�s disease, painful infirmity, and all the indignities that come with old age. What intelligent person would want to bring another into this world with the near certainty of that happening?
You cannot be mad and intelligent but you can be mad and clever.
Many millions of people have a religious belief. That is the belief that there is an unseen intelligent all knowing, all understanding, all powerful, perfect in love entity who is the originator of everything here. It is obvious to an intelligent person that this is not true. The facts are not hidden. This world is extremely violent, dangerous and ugly and always has been. If there is a hidden unseen entity, it is obvious that it is malicious, evil and not intelligent. If you had a young child, you would not say �see how intelligent I am, I can do thousand of things that you cannot, love me, bow down, worship and adore me�. If you did so, you would be both mad and unintelligent. Yet that is how religious people view there God.
An intelligent person would not think that a person who supposedly lived two thousand years ago could be his/her saviour by giving his life for them. The intelligent person would know that sin is subjective and that he/she does not need saving. An intelligent person would not believe that by blowing himself to pieces with the men, women and children in his vicinity he will be transported to paradise (by a just and loving God) and have beautiful serving females granting him his every sexual desire.
By the same measure, it is unintelligent to accept a theory called evolution. The theory proposes that all mammals (including humans) evolved from a small mouse/shrew like creature that lived at the time the dinosaurs became extinct 65 million years ago. An intelligent person would want to know how and why he was here and would examine the theory, looking at all the evidence and paying particular attention to the time lines involved in that short period of 65 million years. The person who did not could not call himself intelligent. That person might be very clever and accomplished in many ways but would not be intelligent. If you are one of those you are not alone, you are legion.
It is estimated that the human brain has approximately 100 billion neurons (100,000,000,000). You might think that this a very large number? If a neuron was equated to a computer byte that would be equivalent to 100 gigabytes. My PC has a storage area of 150 gigabytes. I take photos with my digital camera, which are 7 megapixels (7,000,000) in size. If you equated one neuron to one pixel, the total capacity of 7 megapixel photos my brain could hold would be 14,000 to 15,000. A neuron is only an organic non-thinking unaware cell that connects with many other neurons via chemical and electrical changes in its synapses. As you can see, 100 billion is not a lot for the equivalent of a simple on-off switch. How do these unthinking cells combine and make us sentient, conscious beings? And why is it that although there are no difference in everyone�s neurons some people are very clever and others are not? And why is there so little intelligence (understanding)?
Imagine an intelligent visitor to this planet for the first time. What would he think after looking at its history and current situation? Elected governing representatives from different factions shout at each other like children. Countries ruled by unstable aggressive people. Millions dieing of starvation while billions are spent on unimportant pursuits. Men woman and children killed for reasons of race or religion in wars and internal conflicts. Countless women becoming pregnant and having abortions. A divide between a minority that are wealthy and the majority who are not. Pollution of the seas and exploitation of the land. The list would go on and on and on. The only correct conclusion possible is that the human race per se has no intelligence.
Probably the most profound words in any language are �I think, therefore I am�. The man who said that also said �If you would be a real seeker after truth, it is necessary that at least once in your life you doubt, as far as possible, all things�. The truly intelligent person examines himself/herself first. He then has the benchmark to question and examine others.
Robert


General versus domain intelligence

Our brains come with hard-wired algorithms. Cats can catch birds or mice without thinking about it. I can grab and eat a strawberry without thinking. The Savanna-IQ Interaction Hypothesis says that general intelligence may originally have evolved as a domain-specific adaptation to deal with evolutionarily novel, nonrecurrent problems. We can derive from this hypothesis that people with better general intelligence won’t be better at routine tasks. In fact, they may fare worse at it! They may only have an edge for novel tasks. Thus, general and domain intelligence may be somewhat separate entities.

How do you recognize people with better general intelligence? They are better at adapting to new settings. They are the first to adopt new strategies. But they may not be very good at baseball or boxing, and they may be socially inept.

Modern Artificial Intelligence (and Machine Learning) is typically domain-specific. My spam filter can detect spam, but it won’t ever do anything else. Our software has evolved to cope with specific problems. Yet, we still lack software with general intelligence. Trying to build better spam filters may be orthogonal to achieving general intelligence in software. In fact, software with good general intelligence may not do so well at spam filtering.

Reference: Satoshi Kanazawa, Kaja Perina, Why night owls are more intelligent, Personality and Individual Differences 47 (2009) 685–690

Further reading: Language, Cognition, and Evolution: Modularity versus Unity by Peter Turney


Truth, Lies and Enlightenment: how AI can help us to build knowledge and understanding in the echo chambers of life

AI is both a cause and a solution to the problem of a world where there is far more information than any one person can possibly effectively process to construct their own understanding about what they believe and what they don’t. AI can amplify the echo chamber by promoting the most believed over the most evidenced. BUT it can also help us to recognize valid information from noise, IF we know the right questions to ask and IF WE KNOW HOW TO WORK WITH OUR AI we can develop deep understanding and escape from the maze of invention…

Early in my career I was advised that if I wanted to get a point across when teaching, during an interview, as part of a presentation or when debating, I must repeat the point I wanted to make three times. There is an empirical basis for this advice: something eloquently explained my Malcolm Gladwell and the motivation for my blog identity: The Knowledge Illusion. Put simply, when people are provided with more information about X, they believe that they know more about X, when in fact they often know less about X. I wrote about this many blogs ago (transcribed below for ease of reference) to draw attention to the essential need to help people decipher the huge volume of information that comes their way so that they can discern what is genuine from what is fake.

I still follow the “say things three times” advice in my endeavour to communicate what I consider to be valid, some might say truthful, information. My objective is to persuade people that my perspective, opinion, or information presentation is the stuff to be believed. However, I accept that it is entirely up to my audience to decide whether or not they are won over. The importance of this subjective experience and the belief that an audience are actively analysing the information that comes their way is ever more important. In a world of echo-chambers and deluge of social media, we need people to be able to look at a stream of data and information and make intelligent decisions about what they believe to be the stuff of knowledge.

The problem is not new. It was JFK who once observed that “No matter how big the lie repeat it often enough and the masses will regard it as the truth.” This is an enormous insult to the intelligence of the “masses”, but unless we pay attention to helping these “masses” to navigate through the morass of mediocracy that social media precipitates, proliferates and perpetuates then we will return to the pre-enlightenment era when the world was flat and knowledge was the privilege of those who knew how to decipher the written word and who acted as the mouth-piece for and the collective intellect of their communities: the “masses”.

The word “masses” is no longer widely used so let’s just refer to the “masses” as the people: the global human race whom education is intended to equip with the skills and abilities to think and make sense of the world and the information others produce about it. To consider what it is we need to do to help people to make sense of the world it is worth travelling even further back in time to the views of Roman Emperor Marcus Aurelius that: “Everything we hear is an opinion, not a fact. Everything we see is a perspective, not the truth.” We need to encourage a nuanced belief system where people are provided with the skills, confidence and resources to construct their own understanding from the tidal wave of data and information that threatens to engulf them.

Again, history can help to inform us. The scientific revolution set the stage for the age of enlightenment that transformed the human race and promoted the importance of reason. Influential thinkers like Bacon, Locke and Descartes paved the way for the likes of Voltaire, Kant and Smith. Life was so much simpler then of course, but the huge increase in what it is possible for an individual to try to understand and know does not discount the important role that influential thinkers can play.

The birth of the www and social media represent a new generation of publications that play the role of the encyclopedias and dictionaries in the age of enlightenment. BUT who are the key philosophers and scientists who can catalyze the popular debates in the way that the philosophers of the enlightenment did? Stephen Hawking would probably be high on the list of influential thinkers who many people (the “masses”) might be able to name. Who else?

Whilst the volume of information and data about the world has ballooned, the number of influential thinkers who can help people find their way to knowledge and understanding has may not have kept pace. Technologies that harvest the ‘wisdom’ of the crowd often promote the loudest shouters and the most-followed, rather than the considered and grounded reasoning of the real intellectuals. The demise of expertise has exacerbated the problem as professional predictions have failed to materialize…. Let’s just stop there for a moment.

Could the real problem be that we, the people, don’t know how to interpret expertise? We want simple answers when there are none to be had. In schools we still encourage the belief that rote learning and subject specific information of the type that can be reproduced by a single person when challenged with a standardized test sufficient. This outdated approach gives the impression that knowledge and understanding are way more simple than they really are. They encourage people to believe that there is a body of stuff that they need to learn and reproduce, and that if they can do this they will be knowledgeable. However, what we should be doing is ALSO encouraging people to constantly probe, prod, compare and conclude for themselves their understanding of the world so that they can apply this knowledge to solve the problems they encounter every day.

The surge of tweets that give the impression that meaningful things can be said in 140 characters is not always helpful either. There is certainly something to be said for trying to distil understanding into a short text — it is difficult and can test how much we really understand. However, the believe that a tweet can be the whole story in and of itself is misguiding. Knowledge and wisdom need to be worked at, by questioning, analyzing, aggregating and synthesizing to reach our own evidence-based beliefs about what we know and what we understand. Someone else’s tweet might start this process, but we have to finish it for ourselves.

Ai can help us to do the work here. AI can analyze and visualize complex data and information in order to literally help us see the ‘wood from the trees’. AI can be built to model human understanding and to justify the decisions and predictions that it makes. AI can explain to us how to complete complex activities, such as solving mathematical equations or managing a complex power plant. BUT Artificial and Human Intelligence must work together to help people extract the truth from the lies. We as humans must ensure that we know enough about what AI is capable of doing to ensure that we ask the right questions. We must learn to be discerning enough to challenge the AI when we are not convinced by what it is telling us.

This means that now more than ever we must educate the educators. Because educators must instill in us, the people, the investigative skills that we need to ask the right questions so that we can differentiate evidence from falsehood. Educators must encourage the confidence and self-efficacy in us that will help us believe our own minds. Educators must engender the perspective taking and integrative thinking that will enable us to work together to solve problems and to develop the influential thinkers we need now more than ever to enlighten us.


Very Computer

I am currently experimenting with some reinforcement learning (RL)
techniques for a small mobile robot moving about in a cluttered
environment. My aim is to have the 'bot moving happily about,
not bumping into obstacles, whilst navigating towards a goal.

In order to improve the robot's performance (in avoiding obstacles
and making progress towards a goal) I would like to be able to
have the reinforcement learning system try to predict the future
outcome of an action - in terms of the total reinforcement recieved
over a number of future time steps - and then incorperate this
prediction into its decision making process.

Ideally, I would like the learning of the prediction system to be
continuous. I want to avoid having to store the state/reinforcement
of the system over the number of prediction time steps required
(in order to retrospectively calculate the prediction error) if possible.

There is also the problem of how far ahead the system should attempt
to predict. I suspect that a good strategy might be to have this
'time window' adaptable, so that in some circumstances the 'bot
might only want to estimate a few steps ahead whereas in others a
more long-term vision might be beneficial.

I'm working with simulations only at the moment, but would like to
transfer the RL system to a mobile robot at a later stage.

I would be grateful to hear from anyone out there who has experience
with this kind of future predicting/modeling RL technique, or could
perhaps point me to some interesting references.

___________________________________________________________________________ _
Bob Mottram "Robots may move suddenly and without warning"

Reinforcement learning - predicting reinforcement

by Magnus Sandber » Sat, 23 Sep 1995 04:00:00

>I am currently experimenting with some reinforcement learning (RL)
>techniques for a small mobile robot moving about in a cluttered
>environment. My aim is to have the 'bot moving happily about,
>not bumping into obstacles, whilst navigating towards a goal.

>In order to improve the robot's performance (in avoiding obstacles
>and making progress towards a goal) I would like to be able to
>have the reinforcement learning system try to predict the future
>outcome of an action - in terms of the total reinforcement recieved
>over a number of future time steps - and then incorperate this
>prediction into its decision making process.

Quote: >Ideally, I would like the learning of the prediction system to be
>continuous. I want to avoid having to store the state/reinforcement
>of the system over the number of prediction time steps required
>(in order to retrospectively calculate the prediction error) if possible.

I hope my reasoning is understandable. And, I'd be delighted if anyone could
find a flaw in it because I don't like the implications. After all, humans
learn by reinforcement, and though the brain may be vastly more complex and
better designed from the start, the basic principle should still be the same.
We evaluate our successes and mistakes and learn from it, but the evaluation
process can also be modified if we find that it's not delivering what it
should. The evaluation of the evaluation process is, of course, also an
intelligent process. At first sight this seems to require a meta-intelligence
responsible for training our mind, and a meta-meta-intelligence to train the
meta-intelligence etc. In reality there has to be some kind of bootstrap effect
in that everything reinforces everything. But exactly how this works and
whether it can be copied to more modest AI scenarios is a big question mark to
me. Perhaps it requires a significant level of intelligence to start with.
There is, after all, a good deal of work behind a newborn baby's brain.

For the time being I've given up on finding a method of the kind you're asking
for. My current approach is to design a learning system that can be trained
with delay without too much overhead.

Quote: >I would be grateful to hear from anyone out there who has experience
>with this kind of future predicting/modeling RL technique, or could
>perhaps point me to some interesting references.

Reinforcement learning - predicting reinforcement

by Chris Connol » Tue, 26 Sep 1995 04:00:00

>I am currently experimenting with some reinforcement learning (RL)
>techniques for a small mobile robot moving about in a cluttered
>environment. My aim is to have the 'bot moving happily about,
>not bumping into obstacles, whilst navigating towards a goal.

Quote: >You're missing a piece of the puzzle. There is a whole gosh darn
>research area in "adaptive critics" that deals with this exact
>problem. Try looking up anything by Andy Barto, Richard Sutton, or
>Paul Werbos. Or pick up any recent NIPS proceedings.

o Peter Doyle and J. Laurie Snell (1984), "Random Walks and Electric
Networks", American Mathematical Society Carus Monographs in
Mathematics. (VERY clear, good reading)

o Kemeny, Snell and Knapp (1976) "Denumerable Markov Chains",
v. 40, Graduate Texts in Mathematics, Springer-Verlag.

Also, Jonathon Bachrach's Ph.D. thesis from UMass (CS Dept. Technical

with using RL for robot navigation -- some of this has also appeared
in NIPS 2 (Mozer and Bachrach?) and 3.

For an approach from the standpoint of harmonic potentials, there are
a few papers you can find in the UMass robotics lab:

In particular, the following tries to make a couple of connections:

o C. I. Connolly (1994), Harmonic functions and Collision
Probabilities, IEEE Conf. on Robotics and Automation, pp 3015.

SRI International Phone: (415) 859-5022
333 Ravenswood Ave. Fax: (415) 859-3735
Menlo Park, CA 94025 WWW: http://www.ai.sri.com/

SRI International Phone: (415) 859-5022
333 Ravenswood Ave. Fax: (415) 859-3735
Menlo Park, CA 94025 WWW: http://www.ai.sri.com/

Reinforcement learning - predicting reinforcement

by Shez » Fri, 29 Sep 1995 04:00:00

>. humans
>learn by reinforcement, and though the brain may be vastly more complex and
>better designed from the start, the basic principle should still be the same.
>We evaluate our successes and mistakes and learn from it, but the evaluation
>process can also be modified if we find that it's not delivering what it
>should. The evaluation of the evaluation process is, of course, also an
>intelligent process. At first sight this seems to require a meta-intelligence
>responsible for training our mind, and a meta-meta-intelligence to train the
>meta-intelligence etc.
> .

Regarding evaluation of high level (conscious) evaluation processes, where
they are unable to evaluate themselves I think we tend to get stuck - hence
the invention of the therapist. :-)

Obviously there are many areas where our learning mechanisms are
phenomenally successful, eg. being able to see things, acquiring language,
etc., but as you note yourself, these tend to be areas where we're born with
plenty of bootstrap code. As we ascend the levels of thinking, things get
progressively shakier. You mention meta & meta-meta-intelligences: for this
we tend to rely a great deal on interaction with external intelligent agents
- parents and teachers!

The equivalent to the parent, teacher, and, most particularly, the therapist
in AI is of course the programmer. And it is perhaps pertinent to note here
that some pretty successful real-life therapists* decided to call their
activities 'neuro-linguistic programming' (NLP). Their work has been done
entirely at a meta level - they help people generate new strategies to
resolve their problems, often without having to know anything at all about
the nature of the problem that the person has. *[R.Bandler & J.Grinder]

One interesting thing that the inventers of NLP noticed was that there are
plenty of people who not only know that they are failing to do something
right, but who also know exactly what they are doing wrong and why. Yet
despite having this knowledge they are often unable to revise their
behaviour without external help.

Consider for example someone with a phobia (be it spiders, enclosed spaces,
whatever). Here they have learned an inappropriate response to some stimulus,
and although they are fully aware of the problem and what behaviour would be
more appropriate, they need special coaching to implement the changes required.

We might compare this to a robot which keeps banging against a particular
shaped obstacle, and although it learns to recognise that obstacle, is unable
to generate the sequence of movements required to circumnavigate it.
Although we would hope to have it evaluate the failing in its strategy and
learn another way of tackling the problem, we cannot necessarily hold humans
up as proof that such mechanisms are possible.

Incidentally, people should not be misled, by the therapy/mental health
examples, into thinking that a special type of learning problem is involved.
The above comments apply equally well to pretty much any situation where a
failure to learn occurs, eg. in school subjects. People often think that
some facet of cognition they have come across applies only to the narrow
area where they found it, but in my experience cognitive procedures are
surprisingly generalisable.

[Disclaimer - GA & NN are not my field. I'm a cognitive scientist.]

Reinforcement learning - predicting reinforcement

by Steve Bu » Sun, 01 Oct 1995 04:00:00

>>I am currently experimenting with some reinforcement learning (RL)
>>techniques for a small mobile robot moving about in a cluttered
>>environment. My aim is to have the 'bot moving happily about,
>>not bumping into obstacles, whilst navigating towards a goal.

Quote: >>Ideally, I would like the learning of the prediction system to be
>>continuous. I want to avoid having to store the state/reinforcement
>>of the system over the number of prediction time steps required
>>(in order to retrospectively calculate the prediction error) if possible.
>I'm not sure if this can be done. In order to make a decent prediction of the
>future without knowing anything else than the present state and past success of
>the robot, the evaluation system would have to be far more intelligent than the
>decision making process itself. No simple function/extrapolation would do. So
>let's design another learning system whose task is to determine (without delay)
>the success of each of the robot's moves. Let's say we pick some kind of neural
>network. Then the network, too, needs to be trained, but it can't be trained
>continously (since that would require us to know at once how successful the
>latest move was, which is what we are looking for, in order to evaluate the
>net's performance). We thus choose to train the network with a delay of ten
>(say) steps. Knowing the states of the robot between steps [t,t+10] will let us
>make a decent evaluation of the robot's move at step t and thus permit us to
>evaluate the net's output at step t. But in order to do this we have to have
>saved the complete state of the network at step t, and so nothing is gained.

A human baby comes with certain "burned in" routines ("rooting" for a
* and suckling, is an example of what consitutes a more complex
behaiviour, blinking is an example of a rather simple reflex and
digestion is an example of a bio-chemical reaction. With the addition
of a rather "blank" but potentially utile (LOTS of potential
connections) neural network we have a baby as it comes "from the
manufacturer"

This baby will shortly be able to make learned decisions that involve
a very small number of steps. Example: If I need/want something and if
I make a loud crying sound, my parent will appear, figure out what I
need/want and give it to me. Learned decisions that require many steps
or abstract concepts are beyond the baby at this stage no matter how
beneficial they might be for the baby. Example: If I take the money
grandpa and grandpa sent when I was born and invest it in a mutual
fund that buys hi-tech stocks, I won't need to worry about food when I
retire. This type of behaviour is only exhibited as the organism
matures (if ever).

Sorry this is so long, but here's where I'm going with this.

The little robots, when they start out, could use a similar structure:

An electro-chemical system - they don't need to "know" how to get
electricity from their batteries and use it, it just happens.

Simple reflexes - blink, jerk back from hazards etc.

simple behaviours that ensure survival - hunger/food: how "hungry" am
I? what is the most direct route to the nearst food source? I have to
go to the nearest food source while I still have enough energy. This
could be coded as ROM and may or may not use a dedicated processor (a
sub-conscious/background task if you will)

A neural network, possibly seeded with simple actions - obstacle
behaviours: go around it, push it, pick it up, go over it. At this
point, the number of steps from the action to the evaluation of the
"goodness" of the action should be very small, 1-2 steps. As the robot
becomes more proficient it could try to mix actions together and take
more steps to evaluate "goodness". Example of an intermediate stage -
pick up object A, place it next to object B as a ramp, go up ramp,
check out the top of object B. Example of an advanced stage: push
object B next to a much taller object you can't push (a tabletop or
counter) pick up object A, build a ramp, get on top of object B, pick
up A, build a ramp get on top of the counter, find a new food source
(power outlet)

I hope this made sense. I'm just a layman blessed with internet access
so if I've misused any terms please forgive me. I'll go back to
lurking now

Reinforcement learning - predicting reinforcement

by Marty Stonem » Mon, 02 Oct 1995 04:00:00

[Most of Shez article snipped]

: The above comments apply equally well to pretty much any situation where a
: failure to learn occurs, eg. in school subjects. People often think that
: some facet of cognition they have come across applies only to the narrow
: area where they found it, but in my experience cognitive procedures are
: surprisingly generalisable.

I fail to see what procedures are generalizable, e.g., for learning in
school among the following:
1. learning to play a sport, e.g., baseball
2. learning to quote a Shakespeare sonnet
3. learning to tell the difference between species of birds
4. learning to avoid stubbing your toe on your desk legs.
5. learning to do calculus
etc.

I would appreciate a clue.
Thanks.

Reinforcement learning - predicting reinforcement

by Michael Green » Tue, 03 Oct 1995 04:00:00

>I fail to see what procedures are generalizable, e.g., for learning in
>school among the following:
>1. learning to play a sport, e.g., baseball
>2. learning to quote a Shakespeare sonnet
>3. learning to tell the difference between species of birds
>4. learning to avoid stubbing your toe on your desk legs.
>5. learning to do calculus
> etc.

* a bit of observing or thinking,
* attempting the task
* examining where you erred
* compensating for the error
* Repeat as necessary.

All of the 5 of the tasks you've cited call for those steps. Learning
*anything* exercises those functions and reinforces the brains ability
to learn something else. My son's 2nd grade teacher had the kids
memorizing a lot of poetry principally to exercise their memories.

Reinforcement learning - predicting reinforcement

by Shez » Wed, 04 Oct 1995 04:00:00

> Shez (S. @sv.span.com) wrote:
> : . People often think that some facet of cognition they have come
> : across applies only to the narrow area where they found it, but in my
> : experience cognitive procedures are surprisingly generalisable.

> I fail to see what procedures are generalizable, e.g., for learning in
> school among the following:
> 1. learning to play a sport, e.g., baseball
> 2. learning to quote a Shakespeare sonnet
> 3. learning to tell the difference between species of birds
> 4. learning to avoid stubbing your toe on your desk legs.
> 5. learning to do calculus

One thing I had particularly in mind was Kuhn's theory that knowledge
acquisition (in other words, learning) often progresses in 'revolutionary'
paradigm shifts.

He believed that it applied only to scientific knowledge, whereas I think it
applies to all knowledge, including perception. His book has an example
which purports to show that perceptual paradigm shifts are reversible whilst
scientific ones are not, but his example is misconceived. (I'm not up to
date on the literature so it's likely that others have said this too by now
(I first noted it in 1987)).

So we might say that something nearly all learning processes have in common
is the formation and testing of successive hypotheses about the data. Under
good conditions where data is of high quality and contains little that is
novel, we might only need to form one hypothesis. Newton's characterisation
of the universe served for hundreds of years. In contrast, when trying to
recognise a shape in a darkened room our initial hypothesis might last for
less than one second before we change our mind and decide that it's
something else.

But in each case the process involves forming an model which seems to fit
the data, and then as additional data accumulates which cannot be fitted to
the original model, we first ignore the discrepancy as noise, then recognise
that something is wrong, but stay with it, then build new models, and when
one offers a better fit a paradigm shift occurs as we discard the old one
and use the new one.

The striking thing here is that a single process model serves both something
that happens for one person at one time, and something that happens socially
for a community comprising successive generations of people. To me this
constitutes a surprising generalisation. (Well it was an exciting discovery
for me!)

However, let's try and find some processes which your list of tasks have in
common. (I expect others will have a go at this too - it's an interesting
challenge!)

Looking at the list, some obvious similarities spring to mind between 1 and
4 (physical coordination), also between 2,3,5 (a degree of rote learning),
and especially between 3 & 5 (classification) but I suspect this is not what
you are interested in. (I also note that item 1 actually includes at least
two separate learning tasks - learning the physical skills, and learning the
rules of the game.)

But let's try and find a process that all five have in common. The most
basic task would appear to be the sonnet, which involves learning a sequence
of words. If all the tasks actually do have something in common we might
expect it to be most easily identifiable here.

The general hypothesis building approach already described almost certainly
applies. In learning a list of words we associate them together in some way.
This is easiest if the words make sense, as we are offered an easily
identifiable superstructure which serves to glue the sequence of words
together in memory. (In this context I mean "easy" in the sense that we
already have the skill of making sense of sentences). In contrast, learning
an arbitrary sequence of words requires us to invent a linking structure -
I'm sure we are all familiar with the memorising tricks that are employed in
such cases, such as attaching the words to a meaningful structure of
imagery, or finding semantic or phonetic relationships between the words,
etc etc. (Computers obviously have an enormous advantage to us here as they
are specifically designed to store and process sequential material directly).

When learning a sonnet, we form some unconscious semantic structure from
which we can "read back" the words. Repetition allows this to be refined
until it has a form suited to accurate recall of the exact words as well as
the meaning. Indeed, the bad thing about rote learning is that the structures
we form tend to be devised to recall the words at the expense of an accurate
grasp of the meaning.

There is unlikely to be any wholesale paradigm shift in learning a sonnet,
but the structure will certainly undergo many revisions before we become
word perfect. It is a commonplace observation that in learning their lines
actors often get a mental block over specific sentences and I expect we have
all seen those out-takes on TV where they get a sentence wrong several times
in a row. Each repetition of the error reinforces it. Overcoming it can be
done be creating a new fragment of the associative structure, linking it in
at the appropriate point, and detaching the erroneous fragment. (The final
stage tends to happen spontaneously through preferential reinforcement of
the correct form.) This is exactly how phobias are cured in NLP. The main
difference in something like learning a sonnet is that the revision of the
cognitive structure tends to be muddled through at a subconscious level,
although there is no reason why we shouldn't consciously examine those
phrases we find hard to learn and discover some new meaning in them from
which the correct words can be extracted more easily.

At this stage I am tempted to leave applying the above learning mechanism
to the other four tasks as an exercise for the reader, as this posting is
already two pages long! Certainly I don't feel the need to demonstrate it
applies to recognition of a species of bird or differential equation. The
learning of physical skills is worth a brief look though. If you stub your
toe a lot, your mental image of the size of your foot and/or the amount of
physical movement resulting from a given instruction to your muscles is
obviously an underestimate and needs revising upwards. Either that or your
mental model of the position of your desk's legs is wrong, or perhaps you
have developed a bad habit of swinging your legs too much as you sit,
a habit presumably formed whilst sitting at a different desk. In all these
cases we would normally muddle through and overcome the problem, but again
we can model it explicitly.

There are many ways that we might cease stubbing our toes on our desk, not
all of which might necessarily be described as "learning". For instance, if
it comes from a bad habit of swinging ones legs whilst sitting at the desk,
it might be a conditioned reflex that goes away because the triggering
stimulus is removed or we get into the habit of working instead of looking
out the window during class. (In such a case no link to learning a sonnet
springs to mind.)

If we're talking about stubbing them whilst walking about though, we might
cure it by paying more attention to our feet until the problem goes away.
This could involve a mechanism similar to the sonnet example: we form a new
and more accurate fragment of the mental model we have for our body's size
and position and how it moves, and after attending to it consciously for a
while the fragment becomes incorporated in our unconscious body model.

I see no reason why these mechanisms should be the only possible way of
learning something, but they are *a* way. Going back to the original posting
that I responded to, the issue was the evaluation of a learning mechanism
with a view to improving it when it failed. It seems to me that the
procedure at the meta-level is similar to that of the learning examples
given here - when one thing doesn't work, try something different! The
manner of producing permuatations or alternative models will depend on the
task, but that doesn't mean to say the process is necessarily different,
it's just that the data types being manipulated have different modalities.

We might draw a parallel with object oriented programming. Different types
of Object require their own implentations of the Methods used to manipulate
them, but the method names and arguments often stay the same, and a calling
program needn't know anything about the difference between say, the display
method for a video object and the display method for a text object, but can
manipulate both objects using the same algorithm. Similarly in cognition, we
needn't start from scratch in approaching a new domain, provided of course
that our algorithms are abstracted to suitable levels and we have not
jumbled high level tasks in amongst low level implementation of routines
that apply only to one domain.

This of course is a primary drawback of neural nets, where the model is not
explicit. Unless we partition the problem into various levels before
presenting it, the solution net is unlikely to be of any use to another
problem. For myself I find neural nets of limited interest, as although they
may solve problems, and show how our intelligence may be implemented, they
cast little or no light on the problem domains themselves, so we are none
the wiser as to how applicable a particular neural net solution might be to
another problem domain. (I am not an expert on neural nets, but it seems to
me that this is inherent in their nature - comments please!)

Anyway, this is all I have time to write today! I welcome discussion on any
of these issues, especially as I'm just finding my feet again in cognitive
science after several years of having .


Systems Perspective

Seed Knowledge &mdash The Systems Scaffold

Our models of how the world, other people, and ourselves work are based on having a built-in intuition about how systems work in general. Indeed, our entire knowledge base is organized around systemness. And when we learn, we are incorporating our perceptions into a framework of systemness because that is how our brains are wired.

Our minds naturally look for things like boundaries, wholeness (Gestalt), cause-effect relations, and a myriad of characteristics of systemness. We automatically attempt to find patterns in noisy data, and categorize patterns in hierarchical structures. Our brains process incoming perceptions so as to see the systemic nature of nature. We can't help it.

This is not surprising since through science, which is supposed to be objective, we have discovered that the world, the universe, is indeed comprised of systems and systems of systems. We find causal relations among system components everywhere we look. In fact, the drive behind the scientific approach to knowing is that when we find phenomena that are not previously categorized, for which a pattern of organization and causal relations have not been identified, then we are essentially forced to look for these things. It is as if evolution predisposes us to see systemness because everywhere there are systems. We are systems. And we are subsystems of larger meta-systems.

This propensity to see systemness, or discover it if we don't immediately see it, is a fundamental organizing principle which our brains are constrained to use to learn about the world. The generic system is a kind of seed structure upon which we map percepts in order to have a means of organizing our knowledge.

Every knowledge construction requires some kind of template upon which to organize new knowledge. The mind is not a blank slate (Pinker, 2002). The brain itself is organized in such a way that we begin our construction of knowledge with the aid of built-in biases for key perceptions and organization of those into early conceptual structures, like categorization and hierarchies of types. Thus as we grow and develop our models of the world and ourselves, we start with a foundation of generic systemness and a scaffolding that provides a basic shape to how we understand the world. Literally, we can't see it any other way. To that structure we start fitting our experiences into place. It is probably more a matter of jostling the bits and pieces around until they 'fit' into the scaffolding and among other bits and pieces already integrated. It is a stochastic process. Some bits won't fit anywhere in the edifice and so get dropped even if they should legitimately be part of the knowledge base. Fortunately, these bits are likely to be encountered later again so they have more than one opportunity to get incorporated.

The point is that knowledge is built upon prior existing knowledge and the ultimate seed knowledge is provided by evolution in the form of an ability to model systems.

In the next part I will provide some ideas about how the brain actually accomplishes this feat. For now all you need recognize is that the generic system can be represented as a network or, in mathematics, a flow graph. Figure 1, above, is such a network and it represents the system I have been describing in words. The dashed line circumscribing sapience demarcates the system of interest and the other entities provide inputs and take outputs from that system. Figure 5 shows a generic system with the expected kinds of components. The system of interest has a boundary of some kind, it has component subsystems between which flows and associations occur in an internal network (not shown). It receives inputs of energy, material, and messages from environmental sources and it produces outputs of similar kinds that flow to environmental sinks. The arrows from and to environmental entities may also be reciprocal linkages with entities rather than explicit flows. This representation is kept simple for demonstration purposes.

Figure 5. A generic system has all of the features/attributes of a basic system in generalized form. Neural networks can encode the various elements and their generic interactions. The human brain has the ability to make copies of this generic model and then learn the particular features of each kind of component.

A generic system is encoded into the brain as a template for the learning of all real systems/objects that the brain will encounter in the future. Systems learning entails making a copy of the generic template somewhere in the cortex (probably in the frontal-parietal areas) and then begriming to link up specific perceptual and other conceptual features to the copy as it becomes particularized to the real system being learned. In Part 4 I will revisit this in terms of plausible neural circuits. The point here is that our brains are wired to look for subsystems and boundaries and connections, etc. as we construct a larger network of particulars. Figure 6 is meant to capture some of this. Starting with a fixed template copy, the brain learns the particulars of a system by identifying the features and attributes that should be attached to the model of the real system while also expanding and modifying some of the details. For example, the real system being modeled will have many more component subsystems with particular linkages. Characteristics, such as the nature of the boundary, may be modified as well.

Figure 6. A particular (real) system is learned by attaching perceptual and conceptual features to the template copy and expanding where needed, e.g. in the number of subsystems and their linkages. This is the basis for humans learning what is in the world and how things, and the world, work.

Since systems are subsystems of larger meta-systems, and are, themselves composed of subsystems, this copying-modifying procedure works in both the direction of the larger and the smaller. The brain can build a model of the meta-system by starting with an already built subsystem (now treated as a component) and situating it within the larger system. Note that the entities identified as sources and sinks can now be modeled in their own rights and their linkages constitute the more complete model of the meta-system.

Working from smaller systems to larger meta-systems is a synthesis/integration process. Working from a system inward to model the component subsystems as systems in their own right is analytical reduction. The brain automatically works at doing both of these. The former is driven by a need to understand the context of a particular system and leads to a grasp of a larger world. The latter is driven by the need to understand how a particular system works. Both of these processes are aimed at providing the brain with a basis for anticipating the future behavior of the systems it observes (see below).

Sapient Systems Thinking

As indicated above, one of the characteristics of judgment is in guiding what should be learned. We can now see that the systems bias is part of the basis for this. As our internal models of world systems improve over time and experience, our judgment derived from them can better guide the intelligence machinery in attending to perceptions that help improve the systems models. This is low-level judgment at work, the kind our biological ancestors had evolved. What makes for sapient systems thinking, and judgment so informed, is the role of strategic (long-term planning, see below) thinking, conscious reflection on knowledge being constructed and editing knowledge as needed (including editing plans for acquiring knowledge in the future). Such judgments guide which systems need to be learned.

This is a huge subject, of course, and will need much explication, beyond the scope of this work. One succinct way of looking at this is that sapience expands the role of judgment in guiding future learning and refines the systemic nature of what is attended to in that future time. As noted above, the drives that produce the learning of particular systems causes us to explore both inward (reductionist analysis) and outward (synthesis and integration). The more sapient mind is equally interested in both directions. But all too often most humans run into limitations on what they are able to do in terms of expanding their models and understandings both inward and outward. This is a scope issue relating to the same problem as mentioned above for judgment. Most humans have limited curiosity. They are not driven past a certain point, reached about middle age, I suspect. As children, while the brain is still in rapid development, curiosity directed at learning the smallest details and the largest relationships is at a maximum. It is hard to say for certain when in a person's life the drive to curiosity starts to diminish. It is hard to say why it does. One can imagine a storage limit, but as I have argued, this seems less likely given the way the brain encodes systems by reusing features that are common to many systems and simply organizing appropriate linkages (see Part 4). As an aside, I do posit that our modern education system may have a great deal to do with damping down children's enthusiasm as it attempts to force-feed knowledge, which is generally not systemic in nature, into the minds of young people. By the time they graduate from high school (if they graduate) they have been told, in so many words, that the world contains many different disparate bodies of knowledge and they must choose one such body to learn well so that they can do a good job in the marketplace. It is hard to imagine how this message can promote curiosity and a love for learning.

But I also suspect that a continuing life-long drive to curiosity depends on the level of sapience in the individual. With lower sapience comes a limited scope and time scale for thinking. People learn just what they need to know to get by in the world they are used to. They do not, in general, expect that world to change very much. They expect whatever trends exist to continue on into the future. So at some point they are no longer concerned with expanding their scope (learning the yet larger meta-system in which they are embedded) and they feel competent knowing &lsquoenough&rsquo about the daily systems they deal with that they do not need to know how they work inside. Lower sapience goes along with a limited world view.

Sapience involves intentional model building such that one becomes more effective in problem solving in an ever wider scope as experience grows. One attribute of a wise person is grasping the interconnections between elements of a complex system, especially a social organization. Applying systems thinking to such organizations increases the probability of finding solutions that will work. And wise people seem to continue learning their whole lives.


Measuring meta-intelligence - Psychology

Last edit: 2014 Apr

THIS might take a lifetime to explain better to just experience it. [btw: I have different experience & understanding of this to Dr Dave]

. you want something to read? Ah well, here goes. (we are getting very philosophical here, mind you) into a very Yellow [& beyond] view. I hope you've read the earlier pages.

The 20th Century has been a time of self-examination for humanity.

Katherine Benziger [and now the 21st, ed]

Now if you are starting your journey as a nerdy type like me (a solely head-space who THINKS he knows all -make that past tense now for me) - come back in a few decades - just remember that you did read about this sort of stuff in your youth. enjoy & see you later after you find your feelings - oh, just have a glance first before you move on. For the rest of you, read on and chew well and digest slowly for good health.

We now delve into a very holistic perspective. the Green paradigm started this part of the journey into holism the Yellow paradigm is another huge leap (sorry, I don't want to scare you off - but it's really really worth the trouble to get to).


Saturday, October 4, 2008

From meta ability to meta Intelligence

A Question: Is Play (playful ) a way of being, and intelligence something else?

Being Logical is ”a way of being” and an intelligence. ( they are one and the same)
Being Musical is “a way of being” and an intelligence. ( they are one and the same)
Being Natural is “a way of being” and an intelligence. ( they are one and the same)
Being Playful is a way of being and an intelligence. (They are one and the same)

The key to any Intelligence is in having the “ability”. The intelligence called “Logical” is based on a combination of several abilities if not dozens. Which eventually develops, through a playing with various abilities, (combining various abilities) until they fuse together into a seemingly single unit which we call Logical intelligence.

Due to the works of Howard Gardner and others, the classical definition of Intelligence has now broken free from its restraining prison. And The definition of intelligence, has become more flexible more fluid and varied. The 8 or 9 Intelligences so far being defined is only the beginning, I predict it will be discovered that there are as many intelligences as there are “ways of being”, And the role of abilities (Human Universals) will play a major role, for all these multitudes of intelligences “are” based on “Having the ability” Play develops naturally this ability, one learns this ability simply because play is all about “how to Learn” not what to Learn, (Which we tend to overlook when we are studying Children and play).

Such Intelligences open up new ways of understanding ourselves and they blossom up in our human mind. They are spreading outward and growing like branches in a tree, from this tree I call Play, the Meta Intelligence, behind all these branches, together with these branches and changing these branches as they grow and mature.
Just like the neurons in our brain we are Playing, and neurons when stimulated create new branches and link up in new ways to find the best and most optimal way to create a” New Ability” Our brain is designed to play, we are born with a Meta Intelligence. You could say we are playing automatically. Even when we think we are not - our brain is still playing.

As children we are more conscious of how important play is. It is a human need to learn “How to Learn “ We are born with a Meta Intelligence, our brain which is a super computer a meta computer is designed to “Learn How to Learn” Learning how to Learn, is without a doubt, an Intelligence worth having. Loosing this Meta intelligence, is not good for the brain.
Research into Altzeimers Disease would benefit from the importance of Play to stimulate the neurons to grow again. The more we play the more neurons we create. It is possible to reverse the loss of neurons in the brain as we get older.

From the beginning we use our brain, (or our brain is using us) to learn the ability to survive what is happening and to develop abilities as quickly as possible to help prolong our survival. This raw survival instinct matures into a “mastering” of what is happening. This is an ongoing process throughout life. Learning means mastering. (I have learned to cycle, I now have mastered a bicycle).

Maintaining the optimal level of the brain, in my opinion, is clearly linked to maintaining, and consciously being open to play. To play in such a way as children do - with everything! To play (open to building and breaking down and rebuilding)with ideas, thoughts, feelings, language, images, systems,principles,models, senses, objects, knowledge, sciences, relationships, The happier we will be. As Gregory Chaitin replied to my letter where I asked him to look at the possibility that Mathematics is playing. He replied, the worlds top meta- mathematician , “Yes! I am also playing with Ideas.”

Any society in the future which places Play as a priority will be a prosperous and contented society
This has been predicted by “future researchers “worldwide it is known as the 5th Society

What Odense Council is doing now- “Consciously” and what Denmark for years has been doing (supporting the value of play and children) already proves the point.
Denmark is one of the richest nations in the world, and has been voted by Unesco every year for the past so many years as “the happiest nation on earth”.
The more conscious we become of play as the underlying key to enrich our lives, intellectually and emotionally, the more we will enrich society and the world around us.


Truth, Lies and Enlightenment: how AI can help us to build knowledge and understanding in the echo chambers of life

AI is both a cause and a solution to the problem of a world where there is far more information than any one person can possibly effectively process to construct their own understanding about what they believe and what they don’t. AI can amplify the echo chamber by promoting the most believed over the most evidenced. BUT it can also help us to recognize valid information from noise, IF we know the right questions to ask and IF WE KNOW HOW TO WORK WITH OUR AI we can develop deep understanding and escape from the maze of invention…

Early in my career I was advised that if I wanted to get a point across when teaching, during an interview, as part of a presentation or when debating, I must repeat the point I wanted to make three times. There is an empirical basis for this advice: something eloquently explained my Malcolm Gladwell and the motivation for my blog identity: The Knowledge Illusion. Put simply, when people are provided with more information about X, they believe that they know more about X, when in fact they often know less about X. I wrote about this many blogs ago (transcribed below for ease of reference) to draw attention to the essential need to help people decipher the huge volume of information that comes their way so that they can discern what is genuine from what is fake.

I still follow the “say things three times” advice in my endeavour to communicate what I consider to be valid, some might say truthful, information. My objective is to persuade people that my perspective, opinion, or information presentation is the stuff to be believed. However, I accept that it is entirely up to my audience to decide whether or not they are won over. The importance of this subjective experience and the belief that an audience are actively analysing the information that comes their way is ever more important. In a world of echo-chambers and deluge of social media, we need people to be able to look at a stream of data and information and make intelligent decisions about what they believe to be the stuff of knowledge.

The problem is not new. It was JFK who once observed that “No matter how big the lie repeat it often enough and the masses will regard it as the truth.” This is an enormous insult to the intelligence of the “masses”, but unless we pay attention to helping these “masses” to navigate through the morass of mediocracy that social media precipitates, proliferates and perpetuates then we will return to the pre-enlightenment era when the world was flat and knowledge was the privilege of those who knew how to decipher the written word and who acted as the mouth-piece for and the collective intellect of their communities: the “masses”.

The word “masses” is no longer widely used so let’s just refer to the “masses” as the people: the global human race whom education is intended to equip with the skills and abilities to think and make sense of the world and the information others produce about it. To consider what it is we need to do to help people to make sense of the world it is worth travelling even further back in time to the views of Roman Emperor Marcus Aurelius that: “Everything we hear is an opinion, not a fact. Everything we see is a perspective, not the truth.” We need to encourage a nuanced belief system where people are provided with the skills, confidence and resources to construct their own understanding from the tidal wave of data and information that threatens to engulf them.

Again, history can help to inform us. The scientific revolution set the stage for the age of enlightenment that transformed the human race and promoted the importance of reason. Influential thinkers like Bacon, Locke and Descartes paved the way for the likes of Voltaire, Kant and Smith. Life was so much simpler then of course, but the huge increase in what it is possible for an individual to try to understand and know does not discount the important role that influential thinkers can play.

The birth of the www and social media represent a new generation of publications that play the role of the encyclopedias and dictionaries in the age of enlightenment. BUT who are the key philosophers and scientists who can catalyze the popular debates in the way that the philosophers of the enlightenment did? Stephen Hawking would probably be high on the list of influential thinkers who many people (the “masses”) might be able to name. Who else?

Whilst the volume of information and data about the world has ballooned, the number of influential thinkers who can help people find their way to knowledge and understanding has may not have kept pace. Technologies that harvest the ‘wisdom’ of the crowd often promote the loudest shouters and the most-followed, rather than the considered and grounded reasoning of the real intellectuals. The demise of expertise has exacerbated the problem as professional predictions have failed to materialize…. Let’s just stop there for a moment.

Could the real problem be that we, the people, don’t know how to interpret expertise? We want simple answers when there are none to be had. In schools we still encourage the belief that rote learning and subject specific information of the type that can be reproduced by a single person when challenged with a standardized test sufficient. This outdated approach gives the impression that knowledge and understanding are way more simple than they really are. They encourage people to believe that there is a body of stuff that they need to learn and reproduce, and that if they can do this they will be knowledgeable. However, what we should be doing is ALSO encouraging people to constantly probe, prod, compare and conclude for themselves their understanding of the world so that they can apply this knowledge to solve the problems they encounter every day.

The surge of tweets that give the impression that meaningful things can be said in 140 characters is not always helpful either. There is certainly something to be said for trying to distil understanding into a short text — it is difficult and can test how much we really understand. However, the believe that a tweet can be the whole story in and of itself is misguiding. Knowledge and wisdom need to be worked at, by questioning, analyzing, aggregating and synthesizing to reach our own evidence-based beliefs about what we know and what we understand. Someone else’s tweet might start this process, but we have to finish it for ourselves.

Ai can help us to do the work here. AI can analyze and visualize complex data and information in order to literally help us see the ‘wood from the trees’. AI can be built to model human understanding and to justify the decisions and predictions that it makes. AI can explain to us how to complete complex activities, such as solving mathematical equations or managing a complex power plant. BUT Artificial and Human Intelligence must work together to help people extract the truth from the lies. We as humans must ensure that we know enough about what AI is capable of doing to ensure that we ask the right questions. We must learn to be discerning enough to challenge the AI when we are not convinced by what it is telling us.

This means that now more than ever we must educate the educators. Because educators must instill in us, the people, the investigative skills that we need to ask the right questions so that we can differentiate evidence from falsehood. Educators must encourage the confidence and self-efficacy in us that will help us believe our own minds. Educators must engender the perspective taking and integrative thinking that will enable us to work together to solve problems and to develop the influential thinkers we need now more than ever to enlighten us.


According to the theory, an intelligence 'modality' must fulfill eight criteria: [3]

  1. potential for brain isolation by brain damage
  2. place in evolutionary history
  3. presence of core operations
  4. susceptibility to encoding (symbolic expression)
  5. a distinct developmental progression
  6. the existence of savants, prodigies and other exceptional people
  7. support from experimental psychology
  8. support from psychometric findings

In Frames of Mind: The Theory of Multiple Intelligences (1983) and its sequels, Howard Gardner proposed eight abilities that manifest multiple intelligences. [4]

Musical-rhythmic and harmonic Edit

This area of intelligence with sensitivity to the sounds, rhythms, and tones of music. People with musical intelligence normally have good pitch or might possess absolute pitch, and are able to sing, play musical instruments, and compose music. They have sensitivity to rhythm, pitch, meter, tone, melody or timbre. [5] [6]

Visual-spatial Edit

This area deals with spatial judgment and the ability to visualize with the mind's eye. Spatial ability is one of the three factors beneath g in the hierarchical model of intelligence. [6]

Linguistic-verbal Edit

People with high verbal-linguistic intelligence display a facility with words and languages. They are typically good at reading, writing, telling stories and memorizing words along with dates. [6] Verbal ability is one of the most g-loaded abilities. [7] This type of intelligence is measured with the Verbal IQ in WAIS-IV.

Logical-mathematical Edit

This area has to do with logic, abstractions, reasoning, numbers and critical thinking. [6] This also has to do with having the capacity to understand the underlying principles of some kind of causal system. [5] Logical reasoning is closely linked to fluid intelligence and to general intelligence (g factor). [8]

Bodily-kinesthetic Edit

The core elements of the bodily-kinesthetic intelligence are control of one's bodily motions and the capacity to handle objects skillfully. [6] Gardner elaborates to say that this also includes a sense of timing, a clear sense of the goal of a physical action, along with the ability to train responses.

People who have high bodily-kinesthetic intelligence should be generally good at physical activities such as sports, dance and making things.

Gardner believes that careers that suit those with high bodily-kinesthetic intelligence include: athletes, dancers, musicians, actors, builders, police officers, and soldiers. Although these careers can be duplicated through virtual simulation, they will not produce the actual physical learning that is needed in this intelligence. [9]

Interpersonal Edit

In theory, individuals who have high interpersonal intelligence are characterized by their sensitivity to others' moods, feelings, temperaments, motivations, and their ability to cooperate to work as part of a group. According to Gardner in How Are Kids Smart: Multiple Intelligences in the Classroom, "Inter- and Intra- personal intelligence is often misunderstood with being extroverted or liking other people. " [10] Those with high interpersonal intelligence communicate effectively and empathize easily with others, and may be either leaders or followers. They often enjoy discussion and debate." Gardner has equated this with emotional intelligence of Goleman. [11]

Gardner believes that careers that suit those with high interpersonal intelligence include sales persons, politicians, managers, teachers, lecturers, counselors and social workers. [12]

Intrapersonal Edit

This area has to do with introspective and self-reflective capacities. This refers to having a deep understanding of the self what one's strengths or weaknesses are, what makes one unique, being able to predict one's own reactions or emotions.

Naturalistic Edit

Not part of Gardner's original seven, naturalistic intelligence was proposed by him in 1995. "If I were to rewrite Frames of Mind today, I would probably add an eighth intelligence – the intelligence of the naturalist. It seems to me that the individual who is readily able to recognize flora and fauna, to make other consequential distinctions in the natural world, and to use this ability productively (in hunting, in farming, in biological science) is exercising an important intelligence and one that is not adequately encompassed in the current list." [13] This area has to do with nurturing and relating information to one's natural surroundings. [6] Examples include classifying natural forms such as animal and plant species and rocks and mountain types. This ability was clearly of value in our evolutionary past as hunters, gatherers, and farmers it continues to be central in such roles as botanist or chef. [5]

This sort of ecological receptiveness is deeply rooted in a "sensitive, ethical, and holistic understanding" of the world and its complexities – including the role of humanity within the greater ecosphere. [14]

Existential Edit

Gardner did not want to commit to a spiritual intelligence, but suggested that an "existential" intelligence may be a useful construct, also proposed after the original eight in his 1999 book. [15] The hypothesis of an existential intelligence has been further explored by educational researchers. [16]

Additional intelligences Edit

In January 2016, Gardner mentioned in an interview with BigThink that he is considering adding the teaching-pedagogical intelligence "which allows us to be able to teach successfully to other people". [17] In the same interview, he explicitly refused some other suggested intelligences like humour, cooking and sexual intelligence. [17] Professor Nan B. Adams argues that based on Gardner's definition of multiple intelligences, digital intelligence – a meta-intelligence composed of many other identified intelligences and stemmed from human interactions with digital computers – now exists. [18]

Physical intelligence, also known as bodily-kinesthetic intelligence, is any intelligence derived through physical and practiced learning such as sports, dance, or craftsmanship. It may refer to the ability to use one's hands to create, to express oneself with one's body, a reliance on tactile mechanisms and movement, and accuracy in controlling body movement. An individual with high physical intelligence is someone who is adept at using their physical body to solve problems and express ideas and emotions. [19] The ability to control the physical body and the mind-body connection is part of a much broader range of human potential as set out in Howard Gardner’s Theory of multiple intelligences. [20]

Characteristics Edit

Exhibiting well developed bodily kinesthetic intelligence will be reflected in a person's movements and how they use their physical body. Often people with high physical intelligence will have excellent hand-eye coordination and be very agile they are precise and accurate in movement and can express themselves using their body. Gardner referred to the idea of natural skill and innate physical intelligence within his discussion of the autobiographical story of Babe Ruth – a legendary baseball player who, at 15, felt that he has been ‘born’ on the pitcher's mound. Individuals with a high body-kinesthetic, or physical intelligence, are likely to be successful in physical careers, including athletes, dancers, musicians, police officers, and soldiers.

Theory Edit

A professor of Education at Harvard University, developmental psychologist Howard Gardner, outlined nine types of intelligence, including spatial intelligence and linguistic intelligence among others. His seminal work, Frame of Mind, was published in 1983 and was influenced by the works of Alfred Binet and the German psychologist William Stern, who originally coined the term 'Intelligence quotient' (IQ). Within his paradigm of intelligence, Gardner defines it as being "the ability to learn" or "to solve problems," referring to intelligence as a "bio-psychological potential to process information". [21]

Gardner suggested that each individual may possess all of the various forms of intelligence to some extent, but that there is always a dominant, or primary, form. Gardner granted each of the different forms of intelligence equal importance, and he proposed that they have the potential to be nurtured and so strengthened, or ignored and weakened. There have been various critiques of Gardner's work, however, predominantly due to the lack of empirical evidence used to support his thinking. Furthermore, some have suggested that the 'intelligences' refer to talents, personality, or ability rather than a distinct form of intelligence. [22]

Impact on education Edit

Within his Theory of Multiple Intelligences, Gardner stated that our "educational system is heavily biased towards linguistic modes of intersection and assessment and, to a somewhat lesser degree, toward logical quantities modes as well". His work went on to shape educational pedagogy and influence relevant policy and legislation across the world with particular reference to how teachers must assess students’ progress to establish the most effective teaching methods for the individual learner. Gardner's research into the field of learning regarding bodily kinesthetic intelligence has resulted in the use of activities that require physical movement and exertion, with students exhibiting a high level of physical intelligence reporting to benefit from 'learning through movement' in the classroom environment. [23]

Although the distinction between intelligences has been set out in great detail, Gardner opposes the idea of labelling learners to a specific intelligence. Gardner maintains that his theory should "empower learners", not restrict them to one modality of learning. [24] According to Gardner, an intelligence is "a biopsychological potential to process information that can be activated in a cultural setting to solve problems or create products that are of value in a culture." [25] According to a 2006 study, each of the domains proposed by Gardner involves a blend of the general g factor, cognitive abilities other than g, and, in some cases, non-cognitive abilities or personality characteristics. [26]

Gardner argues that there is a wide range of cognitive abilities, but that there are only very weak correlations among them. For example, the theory postulates that a child who learns to multiply easily is not necessarily more intelligent than a child who has more difficulty on this task. The child who takes more time to master multiplication may best learn to multiply through a different approach, may excel in a field outside mathematics, or may be looking at and understanding the multiplication process at a fundamentally deeper level.

Intelligence tests and psychometrics have generally found high correlations between different aspects of intelligence, rather than the low correlations which Gardner's theory predicts, supporting the prevailing theory of general intelligence rather than multiple intelligences (MI). [27] The theory has been criticized by mainstream psychology for its lack of empirical evidence, and its dependence on subjective judgement. [2]

Definition of intelligence Edit

One major criticism of the theory is that it is ad hoc: that Gardner is not expanding the definition of the word "intelligence", but rather denies the existence of intelligence as traditionally understood, and instead uses the word "intelligence" where other people have traditionally used words like "ability" and "aptitude". This practice has been criticized by Robert J. Sternberg, [28] [29] Eysenck, [30] and Scarr. [31] White (2006) points out that Gardner's selection and application of criteria for his "intelligences" is subjective and arbitrary, and that a different researcher would likely have come up with different criteria. [32]

Defenders of MI theory argue that the traditional definition of intelligence is too narrow, and thus a broader definition more accurately reflects the differing ways in which humans think and learn. [33]

Some criticisms arise from the fact that Gardner has not provided a test of his multiple intelligences. He originally defined it as the ability to solve problems that have value in at least one culture, or as something that a student is interested in. He then added a disclaimer that he has no fixed definition, and his classification is more of an artistic judgment than fact:

Ultimately, it would certainly be desirable to have an algorithm for the selection of intelligence, such that any trained researcher could determine whether a candidate's intelligence met the appropriate criteria. At present, however, it must be admitted that the selection (or rejection) of a candidate's intelligence is reminiscent more of an artistic judgment than of a scientific assessment. [34]

Generally, linguistic and logical-mathematical abilities are called intelligence, but artistic, musical, athletic, etc. abilities are not. Gardner argues this causes the former to be needlessly aggrandized. Certain critics are wary of this widening of the definition, saying that it ignores "the connotation of intelligence . [which] has always connoted the kind of thinking skills that makes one successful in school." [35]

Gardner writes "I balk at the unwarranted assumption that certain human abilities can be arbitrarily singled out as intelligence while others cannot." [36] Critics hold that given this statement, any interest or ability can be redefined as "intelligence". Thus, studying intelligence becomes difficult, because it diffuses into the broader concept of ability or talent. Gardner's edition of the naturalistic intelligence and conceptions of the existential and moral intelligence are seen as the fruits of this diffusion. Defenders of the MI theory would argue that this is simply a recognition of the broad scope of inherent mental abilities and that such an exhaustive scope by nature defies a one-dimensional classification such as an IQ value.

The theory and definitions have been critiqued by Perry D. Klein as being so unclear as to be tautologous and thus unfalsifiable. Having a high musical ability means being good at music while at the same time being good at music is explained by having high musical ability. [37]

Henri Wallon argues that "We can not distinguish intelligence from its operations". [38] Yves Richez distinguishes 10 Natural Operating Modes (Modes Opératoires Naturels – MoON). [39] Richez's studies are premised on a gap between Chinese thought and Western thought. In China, the notion of "being" (self) and the notion of "intelligence" don't exist. These are claimed to be Graeco-Roman inventions derived from Plato. Instead of intelligence, Chinese refers to "operating modes", which is why Yves Richez does not speak of "intelligence" but of "natural operating modes" (MoON).

Neo-Piagetian criticism Edit

Andreas Demetriou suggests that theories which overemphasize the autonomy of the domains are as simplistic as the theories that overemphasize the role of general intelligence and ignore the domains. He agrees with Gardner that there are indeed domains of intelligence that are relevantly autonomous of each other. [40] Some of the domains, such as verbal, spatial, mathematical, and social intelligence are identified by most lines of research in psychology. In Demetriou's theory, one of the neo-Piagetian theories of cognitive development, Gardner is criticized for underestimating the effects exerted on the various domains of intelligences by the various subprocesses that define overall processing efficiency, such as speed of processing, executive functions, working memory, and meta-cognitive processes underlying self-awareness and self-regulation. All of these processes are integral components of general intelligence that regulate the functioning and development of different domains of intelligence. [41]

The domains are to a large extent expressions of the condition of the general processes, and may vary because of their constitutional differences but also differences in individual preferences and inclinations. Their functioning both channels and influences the operation of the general processes. [42] [43] Thus, one cannot satisfactorily specify the intelligence of an individual or design effective intervention programs unless both the general processes and the domains of interest are evaluated. [44] [45]

Human adaptation to multiple environments Edit

The premise of the multiple intelligences hypothesis, that human intelligence is a collection of specialist abilities, have been criticized for not being able to explain human adaptation to most if not all environments in the world. In this context, humans are contrasted to social insects that indeed have a distributed "intelligence" of specialists, and such insects may spread to climates resembling that of their origin but the same species never adapt to a wide range of climates from tropical to temperate by building different types of nests and learning what is edible and what is poisonous. While some such as the leafcutter ant grow fungi on leaves, they do not cultivate different species in different environments with different farming techniques as human agriculture does. It is therefore argued that human adaptability stems from a general ability to falsify hypotheses and make more generally accurate predictions and adapt behavior thereafter, and not a set of specialized abilities which would only work under specific environmental conditions. [46] [47]

IQ tests Edit

Gardner argues that IQ tests only measure linguistic and logical-mathematical abilities. He argues the importance of assessing in an "intelligence-fair" manner. While traditional paper-and-pen examinations favor linguistic and logical skills, there is a need for intelligence-fair measures that value the distinct modalities of thinking and learning that uniquely define each intelligence. [6]

Psychologist Alan S. Kaufman points out that IQ tests have measured spatial abilities for 70 years. [48] Modern IQ tests are greatly influenced by the Cattell-Horn-Carroll theory which incorporates a general intelligence but also many more narrow abilities. While IQ tests do give an overall IQ score, they now also give scores for many more narrow abilities. [48]

According to a 2006 study, many of Gardner's "intelligences" correlate with the g factor, supporting the idea of a single dominant type of intelligence. According to the study, each of the domains proposed by Gardner involved a blend of g, of cognitive abilities other than g, and, in some cases, of non-cognitive abilities or of personality characteristics. [26]

The Johnson O'Connor Research Foundation has tested hundreds of thousands of people [49] to determine their "aptitudes" ("intelligences"), such as manual dexterity, musical ability, spatial visualization, and memory for numbers. [50] There is correlation of these aptitudes with the g factor, but not all are strongly correlated correlation between the g factor and "inductive speed" ("quickness in seeing relationships among separate facts, ideas, or observations") is only 0.5, [51] considered a moderate correlation. [52]

Linda Gottfredson (2006) has argued that thousands of studies support the importance of intelligence quotient (IQ) in predicting school and job performance, and numerous other life outcomes. In contrast, empirical support for non-g intelligences is either lacking or very poor. She argued that despite this, the ideas of multiple non-g intelligences are very attractive to many due to the suggestion that everyone can be smart in some way. [53]

A critical review of MI theory argues that there is little empirical evidence to support it:

To date, there have been no published studies that offer evidence of the validity of the multiple intelligences. In 1994 Sternberg reported finding no empirical studies. In 2000 Allix reported finding no empirical validating studies, and at that time Gardner and Connell conceded that there was "little hard evidence for MI theory" (2000, p. 292). In 2004 Sternberg and Grigerenko stated that there were no validating studies for multiple intelligences, and in 2004 Gardner asserted that he would be "delighted were such evidence to accrue", [54] and admitted that "MI theory has few enthusiasts among psychometricians or others of a traditional psychological background" because they require "psychometric or experimental evidence that allows one to prove the existence of the several intelligences." [54] [55]

The same review presents evidence to demonstrate that cognitive neuroscience research does not support the theory of multiple intelligences:

. the human brain is unlikely to function via Gardner's multiple intelligences. Taken together the evidence for the intercorrelations of subskills of IQ measures, the evidence for a shared set of genes associated with mathematics, reading, and g, and the evidence for shared and overlapping "what is it?" and "where is it?" neural processing pathways, and shared neural pathways for language, music, motor skills, and emotions suggest that it is unlikely that each of Gardner's intelligences could operate "via a different set of neural mechanisms" (1999, p. 99). Equally important, the evidence for the "what is it?" and "where is it?" processing pathways, for Kahneman's two decision-making systems, and for adapted cognition modules suggests that these cognitive brain specializations have evolved to address very specific problems in our environment. Because Gardner claimed that the intelligences are innate potentialities related to a general content area, MI theory lacks a rationale for the phylogenetic emergence of the intelligences. [55]

The theory of multiple intelligences is sometimes cited as an example of pseudoscience because it lacks empirical evidence or falsifiability, [56] though Gardner has argued otherwise. [57]

Gardner defines intelligence as "bio-psychological potential to process information that can be activated in a cultural setting to solve problems or create products that are of value in a culture." [58] According to Gardner, there are more ways to do this than just through logical and linguistic intelligence. Gardner believes that the purpose of schooling "should be to develop intelligence and to help people reach vocational and avocational goals that are appropriate to their particular spectrum of intelligence. People who are helped to do so, [he] believe[s], feel more engaged and competent and therefore more inclined to serve the society in a constructive way." [a]

Gardner contends that IQ tests focus mostly on logical and linguistic intelligence. Upon doing well on these tests, the chances of attending a prestigious college or university increase, which in turn creates contributing members of society. [59] While many students function well in this environment, there are those who do not. Gardner's theory argues that students will be better served by a broader vision of education, wherein teachers use different methodologies, exercises and activities to reach all students, not just those who excel at linguistic and logical intelligence. It challenges educators to find "ways that will work for this student learning this topic". [60]

James Traub's article in The New Republic notes that Gardner's system has not been accepted by most academics in intelligence or teaching. [61] Gardner states that "while Multiple Intelligences theory is consistent with much empirical evidence, it has not been subjected to strong experimental tests . Within the area of education, the applications of the theory are currently being examined in many projects. Our hunches will have to be revised many times in light of actual classroom experience." [62]

Jerome Bruner agreed with Gardner that the intelligence was "useful fictions," and went on to state that "his approach is so far beyond the data-crunching of mental testers that it deserves to be cheered." [63]

George Miller, a prominent cognitive psychologist, wrote in The New York Times Book Review that Gardner's argument consisted of "hunch and opinion" and Charles Murray and Richard J. Herrnstein in The Bell Curve (1994) called Gardner's theory "uniquely devoid of psychometric or other quantitative evidence." [64]

In spite of its lack of general acceptance in the psychological community, Gardner's theory has been adopted by many schools, where it is often conflated with learning styles, [65] and hundreds of books have been written about its applications in education. [66] Some of the applications of Gardner's theory have been described as "simplistic" and Gardner himself has said he is "uneasy" with the way his theory has been used in schools. [67] Gardner has denied that multiple intelligences are learning styles and agrees that the idea of learning styles is incoherent and lacking in empirical evidence. [68] Gardner summarizes his approach with three recommendations for educators: individualize the teaching style (to suit the most effective method for each student), pluralize the teaching (teach important materials in multiple ways), and avoid the term "styles" as being confusing. [69]


The myth that the new SAT correlates less with IQ than the old SAT

For some reason many people believe that the old SAT (pre-April 1995) was a much better measure of IQ than the new SAT (post-April 1995). I started believing this too when I found research showing high SAT people regressed much more to the mean on the new SAT than on the old SAT. However this evening I read that the correlation between the old SAT and the new SAT is virtually identical to parallel forms of the old SAT, so the trend I noticed was probably just statistical noise.

The reason people think the new SAT is less like an IQ test than the old SAT is that originally the SAT was explicitly intended to be like an IQ test, the hope being to give opportunity to bright people from socially deprived homes who wouldn’t be able to attend a good college without a test of natural ability. However as IQ tests became more and more politically incorrect, the test makers wanted to distance themselves from IQ, so the test became increasingly about what you learned in school, and less about abstract reasoning.

However what made the SAT correlate with IQ was never the fact that anyone was trying to create an IQ test, it was the fact that the skills you need in college (reading and math) are closely linked to cognitive ability.

A similar case was when David Wechsler created the WAIS explicitly to measure intelligence, but created the WIAT, specifically to measure academic achievement. I doubt he was trying to make the WIAT a measure of intelligence, since he had already created an IQ test the point of the WIAT must have been to show clinically significant differences between the two constructs, allowing the diagnosis of learning disabilities. And yet a recent study found nearly a 0.9 correlation between the two tests.

I don’t know what the general U.S. correlation between the SAT and IQ is because there’s never been (to my knowledge) a study that correlated the SAT with IQ in a sample of ALL Americans (not just the college bound elite). All the studies I’ve seen involved students at the same school, sometimes with correction for range restriction (which can be misleading because students at the same school are range restricted on more than just test scores). I have tried to estimate the correlation in the general U.S. population indirectly, by seeing how much samples of high SAT folks regress to the mean of all Americans, but the results have been inconsistent.

Some here believe that the correlation between IQ and SAT is so high that the SAT should be called an IQ test. However the brilliant Chris Langan understood the value of verbal precision, and argued that not even the Mega Test, on which he earned the World record should be called an IQ test. In a landmark 1998 article, Langan wrote:

To avoid the problem of rendering a specific a priori definition of what any such test will measure, it suffices to create a generic alternative description covering all tests which differ in structure or protocol from ordinary IQ tests, and for which high positive correlation with IQ has not yet been established. This new term must refer to a measurable quantity that is specific to the tests it describes, and that may or may not equate to that which is measured by garden variety IQ tests.


General versus domain intelligence

Our brains come with hard-wired algorithms. Cats can catch birds or mice without thinking about it. I can grab and eat a strawberry without thinking. The Savanna-IQ Interaction Hypothesis says that general intelligence may originally have evolved as a domain-specific adaptation to deal with evolutionarily novel, nonrecurrent problems. We can derive from this hypothesis that people with better general intelligence won’t be better at routine tasks. In fact, they may fare worse at it! They may only have an edge for novel tasks. Thus, general and domain intelligence may be somewhat separate entities.

How do you recognize people with better general intelligence? They are better at adapting to new settings. They are the first to adopt new strategies. But they may not be very good at baseball or boxing, and they may be socially inept.

Modern Artificial Intelligence (and Machine Learning) is typically domain-specific. My spam filter can detect spam, but it won’t ever do anything else. Our software has evolved to cope with specific problems. Yet, we still lack software with general intelligence. Trying to build better spam filters may be orthogonal to achieving general intelligence in software. In fact, software with good general intelligence may not do so well at spam filtering.

Reference: Satoshi Kanazawa, Kaja Perina, Why night owls are more intelligent, Personality and Individual Differences 47 (2009) 685–690

Further reading: Language, Cognition, and Evolution: Modularity versus Unity by Peter Turney


The Positive Manifold: Reactive Control in Fluid Intelligence?

What neural mechanisms underlie "fluid intelligence," the ability to reason and solve novel problems? This is the question addressed by Gray et al. in Nature Neuroscience. The authors begin by suggesting that fluid intelligence (aka, gF) is related to both attentional control and active maintenance of information in the face of ongoing processing (i.e., working memory). Each of these concepts, in turn, has been associated with the functioning of the lateral prefrontal cortex - a region that has been massively expanded in humans compared to even our closest evolutionary relatives.

To confirm that individual differences in gF are related to prefrontal functioning, Gray et al. measured performance both on a standard gF task (Raven's matrices) as well as on a standard test of prefrontal function from cognitive neuroscience: the 3-back task. In Raven's matrices, subjects are required to pick which of several stimuli "fits" as the final item in a matrix of abstract patterns (see an example.) In contrast, the 3-back task provides subjects with a series of stimuli, presented sequentially, and requires that they respond if the current stimulus matches the one presented 3 items previously (i.e., to respond yes to the second "B" in a sequence like "A X B Y X B X A"). This task is performed in an ongoing fashion, such that subjects must constantly displace the third item in memory with the second, and update memory with the current item. (If you can't tell from my description, this is an extremely difficult task).

Intuitively, one might not expect a strong relationship between these tasks: 3-back relies heavily on memory, whereas all the relevant stimuli are simultaneously present in Raven's. Conversely, Raven's requires abstract and somewhat "analogical" reasoning, but 3-back requires only rote memorization. So these tasks seem to require very different computations - an individual's performance might be expected vary substantially between them.

On the other hand, there's the concept of the "positive manifold": performance on any two reliably-measured tasks is positively correlated (indeed, this is part of the basis for the concept of "general intelligence"). Surprisingly, the positive manifold may apply to neuroscience data as well: despite the possibility that different neural regions would underlie performance on these two very different tasks, certain regions in prefrontal cortex reliably mediate the behavioral correlations between these tasks.

To demonstrate this surprising fact, the authors distinguished between 3-back performance on lure trials (where the target item had occurred on perhaps the 2nd or 4th previous trial, but not the 3rd back) and those on non-lure trials (where target items occured on 1 trial ago, or more than 5 trials ago). Lure trials actually seemed more sensitive to performance than target trials (in which an item was actually presented 3 trials ago) insofar as accuracy was just as bad as target trials, but RTs were even longer.

Estimates of gF were positively correlated with accuracy on all trials types, but was most strongly related with lure trial performance: taking into account accuracy on non-lure trials or accuracy on target trials, gF still showed a significant relationship with lure trials. Activity in lateral PFC, anterior cingulate, and lateral cerebellum all predicted accuracy, and activity in these regions during lure trials overlapped with up to 92% of the shared variance between gF and 3-back performance. In contrast, this pattern was much more subtle on both target and non-lure trials.

Interestingly, the magnitude of sustained activation (thought to subserve active maintenance) was correlated with 3-back accuracy but not with gF ability. This finding is somewhat at odds with accounts that put "vanilla" active maintenance at the center of intelligence and executive control - other processes (such as those recall and discrimination processes involved in lure trials) appear to more strongly manifest the variance shared with gF. This would seem to have applications to the notion of "reactive control" and "secondary memory" as discussed recently in the literature - future work will need to clarify the relationships between these constructs.

The authors note that grey matter volume in lateral prefrontal cortex is under "significant" genetic control, suggesting that perhaps gF is itself largely heritable. Word has it that a new (but still under review) publication is showing the heritability of gF as being close to 1. In contrast, the authors here suggest that gF is probably not entirely heritable, and that a better understanding of individual differences in the neural correlates of gF could contribute to future attempts at enhancing fluid intelligence.

More like this

You can't see the wood for the trees..smile
Intelligence, intelligent, from Latin intelligere understand (INTER+legere gather, pick out).
Clever, Adroit, dexterous, skilful, talented.
What do you mean by intelligence? What is the difference between being intelligent and being clever? Can a person be intelligent without being clever and visa versa? Hear are some definitions of intelligence.
the ability to comprehend to understand and profit from experience
����
Intelligence is a general mental capability that involves the ability to reason, plan, solve problems, think abstractly, comprehend ideas and language, and learn. In psychology, the study of intelligence is related to the study of personality but is not the same as creativity, personality, character, or wisdom.
����..
Adaptability to a new environment or to changes in the current environment
Capacity for knowledge and the ability to acquire it
Capacity for reason and abstract thought
Ability to comprehend relationships
Ability to evaluate and judge
Capacity for original and productive thought

Of the preceding, the first definition with some addition comes closest to a good definition. That is the continuous desire and ability to comprehend and understand by observation and logical reasoning. Who is intelligent by this definition? The current IQ test has been used (with modifications) for over 100 years. The average IQ score is 100. An IQ score of 160 places you into the genius category and a score of >200 is categorised as unmeasurable genius. Computers are becoming ever more powerful and sophisticated. Is a computer intelligent? No, it will never be able to comprehend and understand. It might appear to do so but that will be an illusion. It will only ever be a programmed machine. Even if it is programmed to generate its own coding it will do so as an uncomprehending programmed manner. It will never think (I think, therefore I am). It will only appear to be as clever as the men or women that programmed it. If you do not comprehend and understand this then you are not intelligent (having understanding). There are young children (seven to ten years old) who have genius IQs of 160-170. They have above normal learning abilities and talents. However, like an autistic savant they are clever not intelligent. They see the world in a simplistic child like way. One of them may write music and play the violin to a professional standard. Another might be able to do complex mathematical problems. However, they do not have understanding. You would not expect complex philosophical insight and understanding from any one of them. The IQ test should be called the CQ test (cleverness quotient) for it has everything to do with measuring cleverness and nothing to do with measuring intelligence.
We live on the thin crust of a sphere of molten rock and iron, which is nearly 8,000 miles in diameter. It moves at 18 miles each second through the black vacuum of space circling a star (the sun) which is 93 million miles away. The sphere has a thin layer of breathable atmosphere (less than 7 miles) and if it were not for the magnetic field generated by its iron core the radiation from the sun would be lethal to life. Only an intelligent person will comprehend and understand how strange, grotesque and bizarre this is.
Many people use homeopathic medicine. Substances are diluted down to such an extreme level that nothing of the original remains. It is obvious that if nothing of the original source remains its only action on the body can be that of a placebo. An intelligent person will understand this fact but a clever person with the IQ of a genius might not. It is more than a 50% certainty that your death will not be pleasant. The percentage chance of you dieing peacefully in a bed surrounded by loved ones is not good. The reality is fear of dying, cancer, Alzheimer�s disease, painful infirmity, and all the indignities that come with old age. What intelligent person would want to bring another into this world with the near certainty of that happening?
You cannot be mad and intelligent but you can be mad and clever.
Many millions of people have a religious belief. That is the belief that there is an unseen intelligent all knowing, all understanding, all powerful, perfect in love entity who is the originator of everything here. It is obvious to an intelligent person that this is not true. The facts are not hidden. This world is extremely violent, dangerous and ugly and always has been. If there is a hidden unseen entity, it is obvious that it is malicious, evil and not intelligent. If you had a young child, you would not say �see how intelligent I am, I can do thousand of things that you cannot, love me, bow down, worship and adore me�. If you did so, you would be both mad and unintelligent. Yet that is how religious people view there God.
An intelligent person would not think that a person who supposedly lived two thousand years ago could be his/her saviour by giving his life for them. The intelligent person would know that sin is subjective and that he/she does not need saving. An intelligent person would not believe that by blowing himself to pieces with the men, women and children in his vicinity he will be transported to paradise (by a just and loving God) and have beautiful serving females granting him his every sexual desire.
By the same measure, it is unintelligent to accept a theory called evolution. The theory proposes that all mammals (including humans) evolved from a small mouse/shrew like creature that lived at the time the dinosaurs became extinct 65 million years ago. An intelligent person would want to know how and why he was here and would examine the theory, looking at all the evidence and paying particular attention to the time lines involved in that short period of 65 million years. The person who did not could not call himself intelligent. That person might be very clever and accomplished in many ways but would not be intelligent. If you are one of those you are not alone, you are legion.
It is estimated that the human brain has approximately 100 billion neurons (100,000,000,000). You might think that this a very large number? If a neuron was equated to a computer byte that would be equivalent to 100 gigabytes. My PC has a storage area of 150 gigabytes. I take photos with my digital camera, which are 7 megapixels (7,000,000) in size. If you equated one neuron to one pixel, the total capacity of 7 megapixel photos my brain could hold would be 14,000 to 15,000. A neuron is only an organic non-thinking unaware cell that connects with many other neurons via chemical and electrical changes in its synapses. As you can see, 100 billion is not a lot for the equivalent of a simple on-off switch. How do these unthinking cells combine and make us sentient, conscious beings? And why is it that although there are no difference in everyone�s neurons some people are very clever and others are not? And why is there so little intelligence (understanding)?
Imagine an intelligent visitor to this planet for the first time. What would he think after looking at its history and current situation? Elected governing representatives from different factions shout at each other like children. Countries ruled by unstable aggressive people. Millions dieing of starvation while billions are spent on unimportant pursuits. Men woman and children killed for reasons of race or religion in wars and internal conflicts. Countless women becoming pregnant and having abortions. A divide between a minority that are wealthy and the majority who are not. Pollution of the seas and exploitation of the land. The list would go on and on and on. The only correct conclusion possible is that the human race per se has no intelligence.
Probably the most profound words in any language are �I think, therefore I am�. The man who said that also said �If you would be a real seeker after truth, it is necessary that at least once in your life you doubt, as far as possible, all things�. The truly intelligent person examines himself/herself first. He then has the benchmark to question and examine others.
Robert


Watch the video: Measuring! Mini Math Movies. Scratch Garden (May 2022).