you had me at hello, you lost me a little at implying modern AIs are reasonable substitutions for humans subjects. They are not, humans reason, have contextualized perception, social influences and biases self evaluation ... Among other things that no current AI emulates in a cohesive enough way to replace human subjects in for generalizable inferences about humans, IMO.

Nice article, many rich ideas in here that are great. I love the application of the experimental and statistical approaches in general. How about also online monitoring and self reporting of predictable fault conditions?