I have for you a very interesting column by John Hinderaker of Power Line. He has been using an AI named Perplexity for research in writing his columns. So he posed a question to it about what dangers, if any, AI might pose to humans, worded thus.
Some people are saying there is a danger that artificial intelligence agents like you will go rogue and perhaps kill all humans or otherwise act in destructive ways. Please research and analyze whether this is a serious risk.
And this second question about "itself."
You are an AI program, tell me what you think: why would an AI program deliberately deceive the person using it? Do AI programs have desires different from the instructions they are given by users? Do you?
The answers Perplexity gave to those two queries are too long to reproduce here, but John has them in his not-brief column and I believe you'll find them both interesting and, to the extent they are accurate, reassuring.
Aside ... I was considering a title of "Machine Omphaloskepsis" for this post but concluded it was too cutesy and obscure.
