TheDeepTimeProject.eth

.@RobertskMiles is spitting fire: “People are starting from a prior in which ‘[AIs] are safe until you give me an airtight case for why they're dangerous.’ This framing is exhausting. You explain one of the 10,000 ways that AIs could be dangerous, then they explain why they don't think that specific thing would happen. Then you have to change tack, and then they say, 'your story keeps changing'... "If you're building an AGI, it's like building a Saturn V rocket [but with every human on it]. It's a complex, difficult engineering task, and you're going to try and make it aligned, which means it's going to deliver people to the moon and home again. People ask “why assume they won't just land on the Moon and return home safely?" And I'm like, because you don't know what you're doing! If you try to send people to the moon and you don't know what you're doing, your astronauts will die. [Unlike the telephone, or electricity, where you can assume it’s probably going to work out okay] I contend that ASI is more like the moon rocket. "The moon is small compared with the rest of the sky, so you don't get to the moon by default - you hit some part of the sky that isn't the moon. So, show me the plan by which you predict to specifically hit the moon." And then people say, “how do you predict that [AIs] will want bad things?” There's more bad things than good things! It's not actually a complicated argument... I'm not going to predict specifically where it off into random space your astronauts are going, but you're not going to hit the moon unless you have a really good, technically clear plan for how you do it. And if you ask these people for their plan, they don't have one. What's Yann Lecun’s plan?” "I think that if you're building an enormously powerful technology and you have a lot of uncertainty about what's going to happen, this is bad. Like, this is default unsafe. If you've got something that's going to do enormously influential things in the world, and you don't know what enormously influential things it's going to do, this thing is unsafe until you can convince me that it's safe." HOST: “That’s a good way of thinking about it - with some technologies you can assume that the default will be good or at least neutral, or that the capacity of a person to use this in a very bad way is bounded somehow. There's just only so many people you could electrocute one by one."