A Turing test of whether AI chatbots are behaviorally similar to humans
Assessing AI by observing its behaviors.
Will AIs become a major threat to our society? How can we effectively handle the risks associated with AI? These questions are becoming increasingly urgent as AIs gain more power and are applied in various real-world situations. It's no longer just a topic confined to movies or fiction; superintelligence could soon become a real thing. Even those who previously dismissed these concerns now seem to be reaching a point where they must seriously contemplate the possibility of AI threats. Understanding present-day AIs seems crucial in addressing these significant questions. But how can we truly understand AIs?
The paper below
https://www.pnas.org/doi/full/10.1073/pnas.2313925121
follows the long tradition of observing behaviors to understand AI, i.e., to infer properties of AIs from how they work. This kind of behavior approach is not popular nowadays in philosophy. But what else can we do?