3.1 Intelligence and knowledge
Most computer systems are written in response to some kind of requirement, for instance the need to process text efficiently or to move robots around, or as a means to answer complex questions. Just to take a few examples at random, the following questions are all ones which we might design computer systems to answer:
- What are the names of all the employees who are left-handed and speak Mandarin Chinese?
- What would be the effect on profits of increasing production by 5%?
- What will the weather be like tomorrow?
Computers solve any problem by stepping through a suitable algorithm. For many requirements, such as sorting, searching or mathematical operations, there are well-known algorithms that work efficiently in real time. Some problems succumb to ‘brute force’ algorithms; however, many serious problems seem to require processes into which some measure of ‘intelligence’ has been blended, if they are to be solved within realistic timescales – or at all.
Give two specific examples of problems that conventional artificial intelligence programs have been written to solve.
All sorts of answers are possible. Computer translation, expert diagnosis, planning, robotic movement, etc. all come to mind here.
I’ve argued that the Al approach to automating processes requiring intelligence originates from an anthropocentric picture of intelligence, one based on knowledge and reasoning.
The Al story is that we are able to understand the workings of our cognition by looking into our own minds, representing what we find there in some formal way, and then translating this representation into a symbolic form that can be handled by a computer. However, a moment’s introspection should convince us that there are limits to introspection itself.
Consider this case. One technique still widely used in Al is to represent human knowledge as a series of production rules of the form:
- IF ...
- THEN . . .
Human experts – doctors, chemists, geologists – are quizzed in lengthy sessions of knowledge elicitation to draw out their knowledge and represent it in the rule-like formalism shown above. This can then be transferred to a computer where it can be processed by algorithms.
However, now consider another human activity, one in which most of us are ‘expert’: riding a bicycle. What are the rules for doing that? Clearly, we could sit together all day and you wouldn’t be able to tell me. It just doesn’t seem to be the sort of thing that can be expressed as rules – or expressed at all, really. We just do it. It doesn’t appear to involve any explicit thought at all.
Try to think of two similar activities you carry out without apparent thought.
Examples are: walking, catching a ball, swimming, etc.
In fact, everyday human living comprises countless activities that we seem to carry out without any reasoning at all. Of course, we had to learn to do them at some time. But this learning process didn’t really resemble formal teaching, of the kind you’re undergoing now. I certainly didn’t learn to walk by being taught the rules of ambulation.
Al stresses the importance of knowledge, but knowledge is not necessarily a single kind of thing. Philosophers have long distinguished between two kinds of knowledge: prepositional and non-propositional. These two can be understood as follows:
- Propositional knowledge is knowledge that can be expressed explicitly in the form of propositions, such as ‘The Battle of Hastings was in 1066’, or ‘The light from distant galaxies shows a red shift.’ It is sometimes called ‘knowing that’, since you can precede each proposition with the phrase ‘I know that … (the sun rises in the east)’.
- Non-propositional knowledge is knowledge that can’t be expressed in this way, usually because it is manifested in some form of ability or skill. Examples include ‘I know how to swim’, ‘I know how to speak French’, etc. So non-propositional knowledge is often called ‘knowing how’.
Non-propositional knowledge is also sometimes called tacit or implicit. Other classifications have been attempted, but I won’t consider them here.
Give two examples of prepositional knowledge you possess, and two of non-propositional.
I know that Java is an object-oriented programming language and that Shakespeare wrote King Lear. I know how to swim and how to speak English.
The last example in the SAQ’s answer is worth a moment’s discussion, since language is so close to the heart of our intelligence. More or less every normally reared human acquires language. Human languages are fiendishly complicated, rule-governed structures – without rules of some kind, a language could hardly be a language. However, if I were to ask an average speaker to state the rules of their language, it’s very unlikely they would be able to do so. Only professional grammarians really have any explicit knowledge of linguistic rules, and even then they seldom agree. So speaking a language seems to be a matter of following rules without even knowing (propositionally) what those rules are! It is a classic example of non-propositional knowledge – at the core of our mental life.