Intro Mind Notes

Week 3: Computational Theories of the Mind

(HMW, Ch. 2, pp. 59-111)

1. Common Sense Psychology (more commonly called Folk Psychology) is an everyday method for explaining human behavior. It explains people's actions by referring to their mental events such as their perceptions, sensations, beliefs, desires, plans and goals.
2. For example, one way to explain why I went to the fridge and took out a can of beer is the following: I had a perception of a cold beer in a TV ad that made me aware of my thirst sensation, which caused me to desire a beer. I also had a belief that there was a beer in the fridge, so this together with my desire cause me to plan to go to the fridge and get it, which caused and action: my muscles acted in the right way to get the beer.
3. Explanations like this are commonplace and extremely useful. They are basic to our abilities to explain and predict the behavior of others. It is hard to imagine how we could function as social creatures without the help of Folk Psychology.
4. Note that explanations in Folk Psychology typically involve interactions between natural events (images on a TV screen, body motions, etc.) and mental events (a sensation of thirst, a belief that beer was in the fridge, a plan to get the beer, etc.)
5. A problem for Cognitive Science is to give an account of Folk Psychology within the vocabulary of the natural sciences. In particular, the problem is to explain what mental events are so as to allow an "interaction" with physical events.

1. One explanation for why Folk Psychology works is the Dualist answer: Mental events are the activities of Souls - things that are not part of the physical world. Pinker objects, that Souls do not explain anything. They leave the interaction a mystery. Furthermore, there is massive evidence that human actions are the result of the activities of the Brain, not something unnatural like a Soul.
2. Another possible response is the Wonder Tissue answer: Mental events are the result of the activity of a special kind of neuro-chemical stuff that produces mental features like sensations, beliefs, and goals. Pinker objects that on this hypothesis, brain tumors, which have all the neuro-chemical stuff brains have, would be expected to think, but they do not. Besides even if wonder tissue is the answer, we would still not have an explanation for why it works.
3. Another response is the Eliminativist answer: There are no minds, so there are no mental events. Folk Psychology is scientifically mistaken. Behaviorists will hold that the only features worth talking about in science are ones that can be defined in terms of observable behavior. Since sensations, desires, beliefs, etc. are not scientifically observable, they are not part of science, and should be treated as invalid concepts. Although Pinker does not mention it here, another eliminativist position is called Eliminative Materialism. This view holds that science allows the use of concepts that are not defined by observation. Nevertheless, Folk Psychology is a bad theory of what is going on in the brain - one that will eventually be replaced by a better theory that uses psychological concepts unlike those of Folk Psychology. Pinker rejects eliminativism as too harsh. We see from our own experience that sensations, beliefs, goals, etc. exist. The problem is to explain what they are with the help of natural science.

1. Pinker believes that our best bet is the computational answer: The mental events described by folk psychology can be identified with information processing states in the brain.
2. Information is generated when one event leaves a trace of another (for example the footprint left by an animal on the beach).
3. Information processing is the physical manipulation of such traces to obtain more complex and useful information about the world.
4. The fundamental idea is that the brain contains symbolic representations (like the data stored in a computer's memory). New input data can are "entered" into the brain in the process of perception. The symbolic representations constitute what some have called a Language of Thought or Mentalese - the symbolic language which the brain uses for information processing - the computer language of the brain so to speak.
5. Mentalese is not English because it must represent multiple meanings of English worlds.
6. The brain contains rules (akin to program steps) that operate over the representations (data) in mentalese, thereby creating new representations and eventually new outputs in the form of muscle activity. So some people call this the rules and representations account or the data and procedure account of the mind
7. By operating over data in this way, the brain is capable of being creative. It can deduce conclusions that it never explicitly stored before.
8. On this picture, we can provide a computer science account of Folk Psychological states: Beliefs are symbolic data stored in memory. Desires are symbolic data that control goal setting during a computation. Sensations are data produced directly by our senses. Trying is the execution of operations encoded in goal data.

 

1. One reason for believing in the computational account of the mind has to do with the work of Alan Turing. Turing defined the simplest sort of computer - a Turing Machine, and argued (in Turing's Thesis ) that any rule based set of operations over symbolic data can be carried out by one of his machines. The digital computers we have today are all (more complex) descendents of Turing's basic machine.
2. So this guarantees that no matter how complex a set of rules over symbolic data might be, there is always a computer that can carry out that process. If what the brain does is to apply rules to data, we can be guaranteed that there will be a machine that can match exactly what it does. (Turing did not show that the computer could do the job with the same speed or using the same amount of memory as the brain does. But he did show that the process could be carried out in principle, leaving aside practical matters such as speed and memory size.) So it would seem that any coherent thinking is something that a digital computer can do, at least in principle.
3. While the brain may not actually be a digital computer, but Turing's Thesis suggests a good strategy for Mother Nature to create an intelligence: rig up the brain so that it can do what a digital computer can do.
4. We also have considerable evidence from successes in computer science that computers can be as intelligent as we are.

 

1. Metaphysical Problem: What is the Mind?

The mind is a digital computer installed in the brain.
2. Interaction Problem: How do mental events interact with natural events?

(Review the discussion on pp. 71-77 where Pinker explains the point with an example.)
There is no mystery about how computational events (such as adding the sum 7+5) can interact with natural events such as key presses (typing in "PRINT 7+5") at a keyboard and the printing of output (printing the answer '12'). Adding a sum is just a very complex physical event in the computer.
For example, in today's computers it is a complex sequence of states of the computer's circuits controlled by the wiring and the program stored in its memory. States of a computer can be described at two levels: The physical level , which explains the electrical events in the computer's circuits, and the computational level , which redescribes these physical events at the information processing level.
It is not as if there are two different kinds of events, the physical events and the computational events with a mysterious interaction linking them together. Since any computational event is already of physical event, the interaction between physical and computational event is really an interaction between events all of which are physical.
On the computational picture, computational event (such as the calculation of 7+5) can be carried out in a wide variety of different physical systems: electric circuits, rotating drums, even tinkertoys. So from the point of view of the computational level, how the computation is implemented in physical machinery is irrelevant. On the other hand, every calculation is carried out by some physical process which mirrors the laws of mathematics. Similarly every mental event is carried out by some physical process that mirrors the laws of thought: the rules over representations that make up a mind.
3. Homunculus Problem: How can we explain perception without bringing in a homunculus?

We are allowed homunculi in the computational picture as long as we break the story of intelligence as a whole into an account of the interaction of homunculi which are dumber that the whole. Then each of these homuncui can be analysed in to dumber ones, until eventually the analysis "bottoms out" at homunculi that are so dumb that it is easy to see how the neural circuits of the brain could carry out the tasks they accomplish.
4. The Problem of Intentionality: How can the Brain contain Representations with Meaning?

Garson explained in class how one could work out the meanings of the states of an Alien Calculator. In this case, one determines the meaning of each state according to the way in which that state interacts with all the others. This is an inferential role account of how the sates of a calculator get meaning: namely it is their complex relationships with each other than fixes the meaning.
This explanation might work for defining the meaning of states of a calculator, but in the case of the brain, the inferential role answer is thought by some to be inadequate. (See Pinker's account (p. 81 top) of the chess program that mirrors the Six Day War.) Another answer is the causal account of meaning, which states that mental states obtain their meanings from their connections to the outside world via our senses. For example, a mental state which is caused exactly when we see a dog, would mean dog.
There is hot debate among philosophers and psychologists as to whether one or both of these views can give an adequate account of meaning for states in the brain. But at least the computational theory provides what appears to be a testable hypothesis about meaning that can be verified by natural science.

1. Searle's Objection: The Chinese Room

Searle claims that the computational account of the mind is wrong because of the following thought experiment. Imagine a man who knows only English who operates on symbols on bits of paper put in boxes according to rules. Imagine that he obeys all the rules that simulate a human who is conversing with someone in Chinese. According to the computational picture, the man/room understands Chinese. For on that view, understanding simply means running the right program with the right connections to the outside world, and the man/room system is doing that. But Searle points out that the man does not understand Chinese and adding boxes and symbols on paper into consideration makes it no more reasonable to say that the man/room system understands Chinese. So the computational story leaves something out: namely what it takes for humans to understand meaning (something that blind rule following can simulate but never duplicate ).
One of Pinker's answers is that if the man/room system could converse in a normal way and simulate a Chinese conversant, it would understand Chinese. Our intuitions that it would not are unscientific.
2. The Alien's Objection

Aliens cannot believe that humans think with meat. In fact it is hard for us to imagine that humans think with meat. Pinker's answer is that while our intuitions have trouble with this idea, worries about the material are irrelevant to the computational view. It doesn't matter what a system is made of. As long as we have a physical system that goes through the right complex computational information processing steps, we have a mind. Yes this is hard to imagine. So is a lot else that science has revealed.
3. Penrose's Objection

Penrose uses Godel's Theorem to argue that human's are creative in a way that no computer can possibly match.
Godel's Theorem says that for any finite and consistent system of rules you define there will always be mathematical truths which it can never be proven in that system. Since a computer program is always a finite (and hopefully consistent) set of rules, there will always be truths of mathematic which humans know to be true, which a give computer cannot generate.
Pinker responds that no proof has been given that humans can exceed what computers can know. Penrose has assumed that humans can know that their thoughts are consistent, and this is more than we can know for sure.