Because there is not enough intelligence in the world and humans often ignore relevant consequences of their decisions, AI can provide the means by which decision makers avoid global catastrophes. I believe we can realize this vision by formulating and testing ideas in the context of writing and experimenting with programs.
Every empirical science needs both theoreticians and experimenters. Turing saw that operational tests of behavior would be more informative than arguing in the abstract about the nature of intelligence, which established the experimental nature of AI.
The two major research themes for both theoretical and experimental AI have always been knowledge representation and inference. Clearly an intelligent person or program needs a store of knowledge and needs inferential capabilities to arrive at answers to the problem he/she/it faces in the world. Other big issues, like learning and planning, can be seen as secondary to KR and inference.
Feigenbaum and I were early players in two major controversies: (1) what are the relative contributions of knowledge and inference, and (2) what representation methods are both simple enough to work with and sophisticated enough to capture the kinds of knowledge that experts use? The DENDRAL and MYCIN programs provide experimental evidence on the side of more knowledge, represented simply.
The sense of collegiality in the AI community has always made AI more fun. Most of the time, the statesmanlike conduct of senior people like Al Newell set an example for debate without rancor. The common goal of understanding the nature of intelligence makes everyone’s contribution interesting.