Friday, January 25, 2008

minsky's attempt to explain hard problem solving

Some thoughts about a recent artichoke blog ("I'd love to kiss you, but I just washed my hair...") which I thought oversimplified learning theory. This includes an excursion to a Stephen Downes article which was linked to by arti. Some thoughts about the virtues and deficiencies of Skinner's behaviourism are revisited. By degrees this leads me into a borrowed review of Marvin Minsky's book, The Emotion Machine, not necessarily as a solution to the problem but as an illustration of the complexity of learning ...

We learn things by practice and repetition. Very true. This is simple behaviourism. The law of effect. A response to a stimulus which is rewarded tends to be repeated. (behaviourism and the inner environment)

Stephen Downes says he is not a behaviourist but when he talks about learning he sounds behaviourist to me:
When you learn, you are trying to create patterns of connectivity in your brain. You are trying to connect neurons together, and to strengthen that connection. This is accomplished by repeating sets of behaviours or experiences. Learning is a matter of practice and repetition.

Thus, when learning anything - from '2+2=4' to the principles of quantuum mechanics - you need to repeat it over and over, in order to grow this neural connection. Sometimes people learn by repeating the words aloud - this form of rote learning was popular not so long ago. Taking notes when someone talks is also good, because you hear it once, and then repeat it when you write it down.
I agree with that, as far as it goes:
Learning can be viewed as self design. There doesn't appear to be a more powerful way to think about design than thinking of it as an evolution wrought by generate and test!
- behaviourism and the inner environment
Arti links to Stephen's article and then goes onto suggest that:
Enhancing student “engagement”, “authenticity of task”, and or “belonging” does not cause improved student learning outcomes anymore than “just washed hair“ prevents you kissing someone you desire
- "I'd love to kiss you, but I just washed my hair..."

My position: Behaviourism works. We are all behaviourists even though it has been unfashionable to admit it because Skinner took it too far and annoyed everyone by making us feel non creative.

But that is only the beginning of the story

It is way premature to take “engagement”, “authenticity of task”, and or “belonging” out of the learning loop just because their connection to learning is not as obvious as the simple behaviourist case.

I believe that when we begin to tease apart the inputs that arti rejects that problems arise. "Engagement" and "belonging" could be described as being more in the "emotional" category and “authenticity of task” might be more in the "cognitive" category. If you understand a topic more deeply then you can teach the important things first - or setup a task which assists this. I would argue that those with the superior concept map of a problem domain could be capable of teaching that domain better, provided they know how to teach, of course. How to take a learner from point A to point B. They would be capable of developing more authentic tasks because they understand the more important (and less important) parts of the problem domain. When I studied quadratics deeply - by writing a logo program to teach quadratics behaviouristically - I was able to teach them more effectively. See quadratics and behaviourism.

I still like Papert's idea of "objects to think with" (isdp) and that some materials (such as the logo programming language) are better with regard to the following criteria:
  • appropriability (some things lend themselves better than others to being made one's own)
  • evocativeness (some materials are more apt than others to precipitate personal thought)
  • integration (some materials are better carriers of multiple meaning and multiple concepts)
For these sorts of reasons I believe it is premature to throw away "engagement", "belonging" and “authenticity of task” even though the connection of these things to learning is not simple and still poorly understood.

Skinnerian behaviourism does not explain complicated human learning. It only explains simple animal learning, including some human learning.

I discussed this in Dennett's creatures:
  • Darwinian creatures - random mutation and selection by environment
  • Skinnerian creatures - favourably actions are reinforced and then tend to be repeated
  • Popperian creatures - have an inner environment that can preview and select amongst possible actions
  • Gregorian creatures - import mind-tools (words) from the outer cultural environment to create an inner environment which improve both the generators and testers
  • Scientific creatures - an organised process of making and learning from mistakes in public, of getting others to assist in the recognition and correction of mistakes
So, how do we explain more complex human behaviour such as recognising that a problem is hard, that we can't solve it and then selecting alternative appropriate strategies to try in different ways?

This is my context for introducing Marvin Minsky's book, The Emotion Machine. A draft is also available online. Not because Minsky is correct but because he looks at the harder questions. It's preferable to be wrong about the harder questions than to oversimplify human learning (which I think perhaps arti and stephen are doing)

I've read parts of Minsky's book and would recommend this summary from amazon readers reviews, which I reproduce in full:
1. We don't recognize a problem as hard until we've spent some time on it without making any significant progress. For if you can diagnose the particular type of problem you face, then you can use that knowledge to switch to a more appropriate way to think.

2. Critic-selector model of thinking: Each critic object can recognize a certain species of problem type. When a critic sees enough evidence, the critic will activate a "selector", which tries to start up a set of resources that it has learned is likely too act as a way to think that may help in this situation.

3. If a problem seems familiar, use reasoning by analogy. If it seems unfamiliar, change the way you're describing it. If it seems too difficult, divide it into several parts. If it still seems difficult, replace it by a simpler problem. If none of these work, ask someone for help.

4. If too many critics are aroused, then describe the problem in more detail. If too few critics are aroused, then make the description more abstract. If important resources conflict then you should try to discover a cause. If there has been a series of failures, then switch to a different set of critics.

5. Emotional reactions: cautious vs. reckless, unfriendly vs. amicable, visionary vs. practical, inattentive vs. vigilant, reclusive vs. sociable, and courageous vs. cowardly; each such emotional way to think can lead to different ways to deal with things-either by making you see things from new points of view or by increasing your courage or doggedness. If too many critics are active then your emotions would keep changing too quickly. And if those critics stopped working at all, then you'd get stuck in just one of states.

6. The best way to solve a problem is to already know a way to solve it. Searching extensively. When one has no better alternative, one could try to search through all possible chains of actions. But that method is not often practical because such searches grow exponentially.

7. Reasoning by analogy: when a problem reminds you of one that you solved in the past, you may be able to adapt that case to the present case situation.

8. Divide and conquer: if you can't solve a problem all at once, then break it down into smaller parts.

9. Reformulation: find a different representation that highlights more relevant information. Understand in a different way.

10. Planning: consider the set of subgoals and examine how they affect each other.

11. Techniques for problem solving: simplifying, elevating, and changing the subject.

12. More reflective ways to think: wishful thinking, self-reflection, impersonation.

13. Other modes of thinking: 1) logical contradiction: try to prove that your problem cannot be solved, and then look for a flaw in that argument. 2) Logical reasoning. We often try to make chains of deduction. 3) External representation. Drawing suitable diagrams 4) Imagination. What would happen if by simulating possible actions inside the mental models that one has built.

14. Creating higher level selectors and critics help to reduce the sizes of the searches we make.

15. Modes of thought: preparation, incubation, revelation, and evaluation.

16. Creative ideas must be combined with the knowledge and skills already possess-so it must not be too different from ideas with which we're already familiar.

17. If too may critics are active then you notice flaws to correct and spend much time repairing them and never get at the important things and people perceive us as depressed. If too many critics are turned off then you ignore alarms and concerns that would help you concentrate allowing errors and flaws. The fewer the critics active, then the fewer goals pursued, making one intellectually dull.


artichoke said...

Thanks for all this new thinking Bill ...

We are starting a new ict_pd cluster this year, the schools involved have chosen to base the new learning on environmental “sustainability”.

Means I am currently trying to unpack “What so worthy and requiring of understanding in Sustainability (The balance between human needs and those of the natural environment), and Kaitiakitanga (Managing the modern day environment based on a Maori world view and notions of guardianship, all life is connected, manu, tapu and mauri).

The mind map I have created of curriculum connections and classifications for Sustainability is so extensive that it is increasingly hard to read it on one screen.

It seems sustainability is a little like one of those Russian matryoshka dolls, every time I determine something that is important to understand I realise that it also has important prior understandings inside.

It occurs to me that many of the “concepts” we expect students to grapple with are so complex, abstract and evolving that most students and their teachers struggle with a lack of the background facts and information necessary if they are even to begin to understand. And it helps me understand why the learning outcomes in these sort of concept curriculum topics often seem to limit themselves to a stream study or a project on reducing school litter– not unlike those environmental studies learning outcomes in the 1980’s.

Learning is an even bigger concept than sustainability – so thanks for adding in some extra connections to Artichoke - my mind map equivalent on Learning. I’ve put Minsky on my “What is worthy and requiring of understanding in Learning” list

Unknown said...

More on problem solving:
Principles for Teaching Problem
Technical Paper #4
Jamie Kirkley
Indiana University