Over the years, my view of an artificial intelligence (AI) course has changed significantly. I used to view it as a course that should discuss our efforts to develop an artificial entity that can learn and make decisions in a complex, changing environment, affect that environment, and communicate its knowledge and choices to humans; that is, an entity that can think. I would therefore cover the weak AI methods that failed to scale up. However, as strong methods that solved challenging problems in limited domains became more predominant, my course increasingly concerned these methods. I would cover backward chaining, forward chaining, planning, inference in Bayesian networks, normative decision analysis, evolutionary
computation, decision tree learning, Bayesian network learning, supervised and unsupervised learning, and reinforcement learning. I would show useful applications of these methods.
These techniques have come to be as important to a computer science student’s repertoire as techniques such as divide-and-conquer, greedy methods, branch-and-bound, etc. Yet a student would not see them unless the student took an AI course. So my AI course evolved into a course that undergraduate students would take either concurrently or following an
analysis of algorithms course, and would cover what I viewed as important problem-solving strategies that have emerged from the field of AI. I feel such a course should be a standard component of every computer science curriculum just like data structures and analysis of algorithms.
