[marginalium]
How to judge AI performance.
27 Dec 2024
How to judge AI performance. It’s notable that our method of telling which AI model is better than others is to test it on human assessments. But AI’s aren’t human, and concentrating on how human-like they are seems like a good way to miss whatever problems they actually will have. Anyway, this paper reckons that it also makes us think AIs are less useful than they are:
We study how humans form expectations about the performance of artificial intelligence (AI) and consequences for AI adoption. Our main hypothesis is that people project human-relevant task features onto AI. People then over-infer from AI failures on human-easy tasks, and from AI successes on human-difficult tasks. Lab experiments provide strong evidence for projection of human difficulty onto AI, predictably distorting subjects’ expectations. Resulting adoption can be sub-optimal, as failing human-easy tasks need not imply poor overall performance in the case of AI. A field experiment with an AI giving parenting advice shows evidence for projection of human textual similarity. Users strongly infer from answers that are equally uninformative but less humanly-similar to expected answers, significantly reducing trust and engagement. Results suggest AI “anthropomorphism” can backfire by increasing projection and de-aligning human expectations and AI performance.
A very complicated way of pointing out that you won’t think AI is useful unless you figure out where it actually is useful, rather than trying to use it as a drop-in replacement for yourself. At least in the short term, anyway.
Anthologies: Gratification, Wealth Architecture, Digital Architecture, On Being Fruitful, On Thinking and Reasoning, Noetik