Undergraduate AI and its Non-imperative Prerequisite

AL1--AL3, AI1, AI2, PL11, SE1, SE2, Discrete Mathematics [page 71, Tucker, A., ed. 1991]. The only Knowledge Unit (KU) in the prerequisites which bears on AI is PL11, a Programming Paradigms KU described thus:

• Introduction to alternative programming paradigms (e.g. functional, logic, and object-oriented) and languages (e.g. Lisp, Prolog, and SmallTalk, respectively). Program construction using at least two of these paradigms. Advantages and disadvantages vs. the procedural paradigm [page 63, Tucker, A., ed. 1991].

The Language Barrier

• Some schools ignore the non-imperative languages altogether and usually end up offering an AI course which might best be described as pseudo-AI. Other schools transfer the responsibility of learning Lisp or Prolog to the AI course itself. This confuses AI with the means of implementation. While Lisp and Prolog (and, for that matter, object-oriented programming) have their roots in AI, they are no longer considered ``AI Languages''. Lisp is an example of a functional programming language, at least in its pure form, and Prolog of logic programming.

• The ACM-IEEE guidelines, which recommends ten hours of lectures, do not suffice to provide the needed background in alternative programming paradigms. For example, in a Lisp based AI course which relied on such a background course, the students could be expected to have had a mere four or five hours of prior language instruction. This is entirely inadequate for all but the most trivial of purposes. Moreover, devoting time to functional and logic-based languages in a Programming Language Concepts course takes time away from the important task, which is the focus of such a course, of examining in detail the run-time structure of programming languages. Most of this examination traditionally involves imperative (procedural) languages. Rather than additionally learning the run-time structure of functional, logic, and object-oriented languages, one is forced, if one follows the ACM-IEEE guidelines, to learn ---or merely be exposed to ---just their syntax.

• In most first courses in AI, some students start out with an elementary knowledge of Lisp or Prolog. But many do not; many have never seen a single non-imperative language. If the course is pitched to the former, it tends to exclude the latter, but if pitched to the latter, the former are invariably ``under'' taught. Though such issues arise with all classes, it is especially acute in AI because adequate instruction in non-imperative languages is still not mainstream. In fact, many CS programs graduate students without any knowledge of functional or logic programming paradigms at all.

• When a serious background course more or less devoted to non-imperative languages is offered, it is usually an upper-level elective course. Thus, by the time AI is offered, many students who are otherwise interested in taking the course discover that they are unable to do so because they have not taken the prerequisite. Again, though this issue arises with most courses, it is especially acute in AI. Students, many of them, have a narrow, even naive, view of the discipline. They tend to prefer courses such as imperative programming, networking, and operating systems, rather than those such as theory of computation, non-imperative programming, and AI. Part of the reason, we suspect, is that many students focus too much on the perceived extent to which their courses will afford immediate and clear skills of interest to industry, rather than the longer term benefits of a thorough understanding of the discipline. In any case, students often will not choose as an elective a course in non-imperative programming and, if AI were to require such a course as a prerequisite, would thus find themselves, perhaps inadvertently, excluded from AI.

A Solution

A First Course in AI

The Language Prerequisite

The Core Concepts

The Lab Component

• At least one programming assignment in which students design and implement a non-trivial AI algorithm or program from scratch.
• At least one assignment in which students use a ``real'' AI system.
• At least one assignment in which students use an AI tool to improve or modify an AI system.
• Several demonstrations of existing AI programs or systems.

Ancillary Remarks

AI Background: Because AI involves knowledge of such things as philosophy, cognitive science, linguistics, psychology, logic, etc, undergraduate students rarely have enough of a background to tackle AI problems without at least rehearsing such material. But if one is not careful, one can spend a great deal of time teaching the background knowledge to AI rather than AI itself. But some background is unavoidable. One must balance the additional time the background to a topic takes against the merits of the topic itself.

Philosophical Matters: Is it really intelligence, or just a simulation? Is it still AI if we know how to do it? Is our approach motivated by cognition? We should not be concerned with these matters in introductory AI. Such matters are best dispatched in short order; there is little enough time for the meat of AI. We favour a view of AI which is neutral on such issues.

Formalism: Formalism is an essential aspect of AI, and indeed, of many advanced learning domains. AI students should therefore be exposed to it and have its importance, at the advanced level, impressed upon them. But most undergraduate CS students are not gifted mathematicians nor formalists, and so the use of formalism should be limited.

History: Students should be exposed to a brief history of AI, though this is no doubt best intermixed with the course, rather than treating it as a separate topic. Some coverage of classical AI programs should be included.

References

Tucker, A., ed. 1991: Computing Curricula 1991: Report of the ACM/IEEE-CS Joint Curriculum Task Force. ACM Inc., 1991.

Appendix