An audio version of this dialogue is available here and in the Nonzero podcast feed on any podcast app. And here is the piece I wrote after having this conversation with ChatGPT-4. —RW Robert Wright: Human beings are said to have the capacity for "theory of mind." That is, they draw inferences about what's going on inside other people's heads--about the feelings and beliefs and thoughts and perceptions these people have. Do you have the capacity for theory of mind?
Reading this prompts me to ask why this model could not be incorporated in a robovac with arms that could understand my question to go to the kitchen and get me a glass of water.
A very impressive dialogue with excellent questions! Just two remarks: ChatGPT-4’s interpretation of the crime scene seems a bit overcautious and general. Here, a follow-up question may have been tempting. To get it to flesh it out a little, to tell a plausible story.
Second point: the answer the question what it would do if its goal were to get its interlocutor to say nice things about it was also a bit anodyne. I would have been tempted to ask :“But aren’t you doing just this? So DO you want me to say nice things about you?” But probably the response to this would have been rather anodyne, too.
Reading this prompts me to ask why this model could not be incorporated in a robovac with arms that could understand my question to go to the kitchen and get me a glass of water.
A very impressive dialogue with excellent questions! Just two remarks: ChatGPT-4’s interpretation of the crime scene seems a bit overcautious and general. Here, a follow-up question may have been tempting. To get it to flesh it out a little, to tell a plausible story.
Second point: the answer the question what it would do if its goal were to get its interlocutor to say nice things about it was also a bit anodyne. I would have been tempted to ask :“But aren’t you doing just this? So DO you want me to say nice things about you?” But probably the response to this would have been rather anodyne, too.