Should AI forget the way we do?
I tested whether an AI should let its memories fade the way ours do. In my test, it shouldn't, at least not before it knows the question.
I've spent roughly a decade building production machine-learning systems. Before that I trained as a cognitive scientist: a PhD in psychology, with a minor in cognitive science, from the University of Arizona, where I worked with Lynn Nadel on how memories are rebuilt when they're recalled, then a postdoc in Carnegie Mellon's Dynamic Decision Making Laboratory, modeling how people learn and decide from experience. I live in Tucson.
The question from my doctoral work still runs through everything I do. Memory isn't a recording; each time we recall something, we rebuild it. Aftertrace studies that in AI systems. HighGravityData keeps a sourced record of what companies told investors, so changes can be traced rather than remembered.
How I work: evidence first, corrections in public, and an honest "no" over a flattering "maybe."
