At a recent conference I saw a demo of an AI video-chat assistant, which appeared to participants as an on-screen cartoon character. The AI assistant would listen to the group discussion, and when it determined that the conversation was floundering, would jump in and try to refocus everyone, pointing out possible next directions based on what had already been said.
Unsurprisingly, the AI was articulate, organized, and well focused. It also had a good grasp of the topic and of possible next topics.
But I had a big problem with the very premise of the project.
In terms of clarity of expression, organized thinking, and ability to summarize ideas, an AI will usually do better than a human. After all, an AI is essentially an amalgam derived by training on prior examples of many humans who have shown how to exercise just those skills.
But an AI is not actually a human. It’s entire “world view” is based on what it has already seen. So if an idea appears which is completely novel, the AI likely won’t realize that it is seeing something new.
In short, a problem with having an AI moderator for an intellectual conversation between humans is that the AI, while generally being much more organized and articulate than the humans, may not recognize when the humans are actually stumbling toward a completely new idea.