—Kosinski’s team used “sanity checks” to analyze how well GPT networks understood the scenario and the human’s predicted response.
Does a typical chess program have a “theory of mind”? From the outside looking in, it looks like it’s trying to figure out what you’re thinking, to predict your response. Or maybe it’s not doing anything like that.
“Does a typical chess program have a “theory of mind”?”
Nah, chess bots are a different animal from these chatbots. The chess bots probably don’t even have any concept of what a human or a mind is. They are only focused on finding the optimum strategies to beat one single problem.
They are basically solving that problem the same way a human chess player is trying to solve it, but the bots have the ability to do the necessary calculations so much faster that while a human grandmaster might be able to think 10 moves ahead, the chess bot can think twenty or twenty five moves ahead in the same alloted time, and therefore the chess bot will find more innovative strategies that get it closer to the optimal outcome.