Quality Assurance for AI: An Inevitable Tradeoff

How do you ensure #quality of a #service which uses #AI for #personalization? #QA in that case is all about risk management. pic.twitter.com/IdKcY5noBL
— ivanjureta (@ivanjureta) January 31, 2018
As currently drafted (2024), the Algorithmic Accountability Act does not require the algorithms and training data used in an AI System to be available for audit. (See my notes on the Act, starting with the one here.) The way that an auditor learns about the AI System is from documented impact assessments, which involve descriptions…
“Objective”, as in, for example, “what I’m saying is objective”, or “that statement is objective”, or “we need objective criteria when making these decisions”, is a complicated term. It takes a lot of effort to make sure it is understood as intended (or closely enough). It is therefore a costly word to use. Why is…
In the creator economy, the creative individual sells content. The more attention the content captures, the more valuable it is. The incentive for the creator is status and payment for consumption of their content. Distribution channels are Internet platforms, where content is delivered as intended by the author, the platform does not transform it (other…
A pluripotent information system is an open and distributed information system that (i) automatically adapts at runtime to changing operating conditions, and (ii) satisfies both the requirements anticipated at development time, and those unanticipated before but relevant at runtime. Engineering pluripotency into an information system therefore responds to two recurring critical issues: (i) the need…
If an AI is not predictable by design, then the purpose of governing it is to ensure that it gives the right answers (actions) most of the time, and that when it fails, the consequences are negligible, or that it can only fail on inconsequential questions, goals, or tasks.
Opacity, complexity, bias, and unpredictability are key negative nonfunctional requirements to address when designing AI systems. Negative means that if you have a design that reduces opacity, for example, relative to another design, the former is preferred, all else being equal. The first thing is to understand what each term refers to in general, that…