I've just started reading a book by Stuart Russell - seems he's one of the main guys in AI. The book is Human Compatible - AI and the problem of control.
Human Compatible: AI and the Problem of Control : Russell, Stuart: Amazon.com.au: Books
From the reviews it seems that he is concerned but there are ways we can develop this without going down some Skynet disaster for humanity. I'm only a chapter or two in, so I don't yet have a grasp on what he suggests is the right route. But he does believe that there are dangers.
I must admit I sat up and took an interest a few months back when I tried ChatGPT. I'd read that it could write computer code which I used to do for draughting software when I was working. So I gave it a simple problem to solve that might have taken me a couple of hours to write and code and check. Three hours tops. I thought it might have given me some basic ideas on how to do it, recommend some sites from which I could learn how to do it, give a rough outline maybe.
It gave me the complete code immediately. And I mean immediately after I hit enter. It took me half an hour to check it. It was better than I could have written. It was...unerving. It had taken me a long time to learn this rather obscure programming language, I didn't know anyone who was competent in it in my line of work (it was why I was in demand at the time) but here was a free facility online that would churn out whatever you wanted whenever you wanted it. Perfectly.
It was a wake up call for me as to where we are going.
