As a young scientist, I found AI’s constant ferment exciting, and I still do. I had previously worked in cybernetics, control theory and pattern recognition, where we modeled intelligence, perception and action as signal processing. However, that view excluded much of what we know intelligence to require, such as symbolic cognition. Modeling cognition as symbolic computation provided a missing link. But we went too far in modeling intelligence as only symbolic. One of our toughest challenges now is to develop architectures that smoothly combine the symbolic and the sub-symbolic. Or, if you like, to synthesize the achievements of logicist AI with those of cybernetics, control theory, neural nets, artificial life and pattern recognition.
Inspired initially by Waltz, Montanari, Huffman, Clowes and Marr, I’ve advocated constraint satisfaction as the unifying model. At both the symbolic and sub-symbolic levels we can specify the internal, external and coupled constraints that agents must satisfy. Those constraints can be static or dynamic. Our development of the robot soccer challenge has forced all of us to develop architectures supporting both proactive and reactive behaviors.
Think of AI itself as an agent. We need a clear understanding of our own goals but we must also be willing to seize opportunistically on new developments in technology and related sciences. This anniversary is a lovely opportunity to take stock, to remind ourselves to state our claims realistically and to consider carefully the consequences of our work. Above all, have fun.