The Psychology of Communication



The main lesson of thirty-five years of AI research is that the hard problems are easy and the easy problems are hard. The mental abilities of a four-year-old that we take for granted recognizing a face, lifting a pencil, walking across a room, answering a question in fact solve some of the hardest engineering problems ever conceived.... As the new generation of intelligent devices appears, it will be the stock analysts and petrochemical engineers and parole board members who are in danger of being replaced by machines. The gardeners, receptionists, and cooks are secure in their jobs for decades to come.

Steven Pinker

9.1 Logic Of Approach

In Chapter 4, Miller, Galanter and Pribram developed their TOTE unit by using the analogy of the nervous system and the computer. Since then, we have considered ontogenetic development - from child to adult (Chapter 5) and phylogenetic development - from animal to human (Chapter 6) to place the study of the nervous system firmly within biology where it belongs. We are organisms not mechanisms. Biology is not rocket science. Rocket science is easy. Biology is the study of complex systems whereas physics is the study of simple systems.

Two complexities of the nervous system have been considered. In Chapter 7, we explored the implications of the fact that the nervous system can be viewed from the inside (experience) as well as from the outside (behavior). In Chapter 8, we explored the implications of the fact that the nervous system has functional as well as structural disorders. That is, since the nervous system "knows" the environment, there can be disorders of the person-environment system as well as disorders of the person.

Because of such complexities unique to the nervous system, it is tempting to return to the strategy in Chapter 4. That is, to approach the nervous system indirectly by looking at simpler systems and try to gain some insight into it by analogy. One of the reasons that mechanisms are simpler than organisms is that we have built them ourselves, whereas we have been created by nature over hundreds of thousands of years. We could perhaps gain some insight into ourselves by building mechanisms which simulate our functions. This simulation approach - the domain of artificial intelligence (AI) - is the focus of this chapter.

My first introduction to artificial intelligence was a talk by Frank Rosenblatt at Cornell University while I was a graduate student. He had built a Perceptron to simulate the human visual system [ROSENBLATT]. It consisted of a board (representing the retina of the eye) wired up to another board (representing the reception area for vision in the occipital lobe of the brain). The system was wired up randomly (which apparently upset his students in electronic engineering who had been taught to wire systems up very carefully). The system was then "taught" the difference between the letters E and F. It was "rewarded" for a correct response by reducing the resistance in the wires which fired; it was "punished" for a wrong response by increasing the resistance in the wires that fired. Thus the probability of making the correct response was increased and of making the wrong response decreased, just as the probability of responses of Thorndike's cats were increased and decreased. Eventually, the system could distinguish between the letters E and F regardless of their size, orientation and position on the screen. The learning curve of this mechanism was similar to that of an organism and supported the argument that the visual system is initially wired up randomly and hard-wired by learning.

I remember leaving that lecture wondering why he had devoted a decade of study and thousands of dollars to build a machine which could distinguish E and F. I had been created by two people who never went to High School in one delightful moment and I could do much more than distinguish between E and F. Artificial intelligence, like military intelligence and smart bomb, sounded to me like an oxymoron. However, I subsequently learned to understand the logic of this simulation approach and to share the dream of the artificial intelligentsia that we could eventually simulate ourselves. Indeed, I once built a graduate course entitled Artificial Intelligence and Natural Communication. The best test of understanding of a system is that you can build that system. Build it and you will understand.