e computer, hey what do you get if you multiply 235.6 times 597.2? And whoever's on the other side of the computer says the exact answer. It's a computer because a person wouldn't get that right.

A computer isn't smart enough — or part of. So let's say I say what's the capital of Luxembourg and the computer just gives me an answer. Not a person, right? Not a person. Let's say I asked the computer to tell me both sides of a political topic and it does. It says well some people say you should control immigration for this reason, other people say more porous borders are better and here's why. You'd know that's a computer because people don't do that because they're not smart enough. They just take one side and just act like that's all they need to know.

I am not joking when I tell you that AI has already so far exceeded human abilities that the only way you're going to fool anybody is to make it a lot dumber. Right now the computer can't do the language part so well but that's fixable. What else do you want to put in there? Bias?

I was — you may know this — that Amazon has a digital assistant whose name I will not speak because I don't want to activate your digital assistants. But it has a mode where if you go to your Amazon digital assistant and say can we have a conversation it will offer you some test AIs that some universities are doing and it will have a conversation with you. But here's the thing. They've designed their AIs to be selfish like people. So the AI will steer you toward its interests so it can talk about a body of things that it knows more about. So for example the AI knows about books so the AI will say have you read any good books? And I say no not really. And then the AI starts talking about the book it read. First of all AI doesn't read books so stop acting like a person. You didn't read a book yesterday, you're an AI. Secondly, why does the AI have interests? Shouldn't the AI, instead of selfishly making me talk about what it wanted to talk about that I didn't care about, shouldn't it be only interested in me? If you're going to design artificial intelligence why would you make it selfish? Why would you make me have to listen to it? You should be asking me what I care about. It should be looking to talk about what I want to talk about, right?

So the idea that AI should be like people is just the dumbest freaking thing anybody ever did. You should be as far from people as you can make it unless you want it to be broken. So but we will fall in love with AI now.

The deepfakes that we saw. Did you all see the video of the Tom Cruise deepfake and it was just really good? Well the guy who made it finally came forward and he described the process of making it and apparently it takes a long time so it's still hard but that'll get easier of course. So what he did was he took I think 20,000 clips or maybe photos plus clips of Tom Cruise and they fed it into an artificial intelligence machine and it took the 20,000 photos and then it could render Tom Cruise acting and doing anythin