Please note: This seminar will be offered twice on the same date to accommodate schedules: once at 11:00 a.m. in DC 1304, and again at 2:00 p.m. in DC 3317.
Daniel M. Berry, Professor
David R. Cheriton School of Computer Science
This talk concerns building an artificial intelligence (AI) to do a non-algorithmic task that requires real intelligence. The literature and practice of AI development does not clarify what is a requirements specification (RS) of an AI that allows determining whether an implementation of the AI is correct.
The talk begins by listing David Parnas’s concerns about how, in contrast to developers of a traditional computer-based system (CBS), developers of an AI are not able to describe precisely what their AI does and seem to be proud when their AI behaves in unexpected ways.
Arguing from the idea that an AI must mimic human beings, the talk shows, with two related AIs for stop sign recognition, how traditional measures to evaluate AIs can serve as a major part of an RS for an AI. The talk suggests that (1) these evaluation measures, (2) criteria for acceptable values of these measures, and (3) information about an AI’s context informing tradeoffs in these measures constitute an RS of the AI. The talk concludes with some open questions that will be the subject of future work.