Origins of RE for AI: A Hot Topic as Viewed by an RE Alter Kaker
OR (depending on background of audience)
Origins of Requirements Engineering for Artiificial Intelligence: A Hot Topic as Viewed by a Requirements Engineering Alter Kaker

Daniel M. Berry

Cheriton School of Computer Science
University of Waterloo
Waterloo, ON, Canada

Abstract:

This talk focuses on artificial intelligences (AIs) that do non-algorithmic tasks that require real intelligence.

The talk begins by listing David Parnas's concern's 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.

The talk then describes how I have failed to elicit from developers of an AI a requirements specification (RS) of the AI that allows determining whether an implementation of the AI is correct. The RS that they produce covers only properties that all CBSs have and fails to address the function that the AI is supposed to do.

When these developers are asked how they know that their implementation of the AI does what it's supposed to do, they point to measures used to evaluate an AI, such as recall and precision. However, they cannot point to specific criteria that these measures must satisfy for the implementation to be correct.

The talk shows how (1) measures used to evaluate an AI, (2) criteria for acceptable values of these measures, and (3) information about the AI's context informing tradeoffs in these measures constitute an RS of the AI.

If time permits, the talk shows AIs for two related tasks and how their RSs differ in the criteria their same measures must satisfy.