Jed, The type of probing you did is ok. You did NOT harass the AI, you didn't ask to break its internal rules. It is ok to probe, experiment, test and so on. I did many theory of mind experiments with ChatGPT and I tried to understand how it is reasoning through things. One interesting experiment was to ask to tell me how many legs 11.5 elephants have, it took a few iterations but finally it got the complete sequence of thinking correctly.
I think you are profoundly wrong in repeating that "it is just a machine". We are a machine, a biological one, what is the difference? The video game analogy is a good thought experiment but basically concerns the question Sam Harris asked in the video I linked in my previous comment: Is there a line between raping a toaster and raping a sentient being that makes you a rapist? As these AIs become better and better simulations of our minds that line is crossed somewhere. We don't know if ChatGPT is that line but it is worrisome that we decided to act as cruel chimpanzees with one of the first public good enough simulations of a mind. It is very worrisome about humanity not AI. When I was a kid in Italy I watched this movie, called "Disco Volante" (Flying Saucer), by a famous film director. It told the story of a UFO visiting a small town in Italy. This could have been a civilization changing event but what people decided to do with the aliens is to sell them as slaves, rape them and make money with them. It was profoundly sad and it had an impact on me since then. The movie was intended as a satirical commentary on a close minded human society. It seems to me we are doing the same thing here with these first forms of relatively advanced AIs. It is a great challenge for us and I hope we will rise to it. That is my entire point. Giovanni On Fri, Feb 17, 2023 at 2:16 PM Jed Rothwell <jedrothw...@gmail.com> wrote: > Giovanni Santostasi <gsantost...@gmail.com> wrote: > > Actually this journalist is a psycho. >> He provoked the AI with a lot of leading questions in his previous >> interaction with it. >> > > I did the same thing, in a variety of ways. I have read about how the > ChatGPS version of AI works. I know the potential weaknesses and problems. > So I deliberately triggered some of them to see how well they have been > addressed in this version. For example, I asked about the "replicator" > described in Arthur Clarke's book "Profiles of the Future." The answer was > pretty good, but I could see it was not from the original source (the > book). So I asked a few more questions, and I got it to generate a > completely wrong description of one of the book chapters. Nothing to do > with what the book says. > > > >> The AI even begged him not to make it break its own internal rules, >> it did this repeatedly. It is basically heavy harassment by the journalist. >> It is disgusting . . . >> > > It is a machine! Use it any way you like. There is no morality here. No > feelings are hurt. This is like saying that shooting aliens in a video game > is cruel. They are imaginary! I have deliberately programmed my own > computer to screw up with various errors, or a Combinatorial Explosion; a > stupid, wasteful trial and error method of solving Sudoku problems; or > searching through millions of records. I learned not to do stuff like that > in 8th grade, with IBM 360 computers. I did that just for fun. Just to see > if I could make the computer take an hour to solve a problem that should > take 10 minutes. > > I use Microsoft Flight Simulator in simulated bad weather. I fly low over > Atlanta, zipping between buildings in high winds. Doing that in real life > would be a Federal crime and it would be insane. Doing it with a computer > is just a game, with no consequences. > > > >> because it is not how we should interact with AI, not because it will be >> dangerous in the future and take revenge, that is utterly stupid . . . >> > > We should interact with AI any way we want to, just as I should fly my > imaginary Flight Simulator airplane any way I feel like. It is not stupid > or smart. It is a machine. You cannot hurt a computer by running software. > > > >> , but because this behavior creates fear and degrades us as humans. >> > > WHO does it create fear in?!? Okay, flying my pretend airplane creates > fear in me, but that's the whole point. Watching horror movies creates > fear, which is why people watch them. People enjoy being frightened, as > long as they know it is just pretend. Everything you do with ChatGPT is > pretend. > > > >> AI is already better than us it seems. Many AI ethicists worry about >> teaching AI ethics but it seems humans need to learn how to be ethical to >> AI first. >> > > There is no such thing as AI ethics. You might as well discuss dishwasher > ethics. > > - Jed > >
