>I get your point that from all experiments we performed so far and bothered to write down and digitize there might be few tricks left to squeeze out, but vast

>I get your point that from all experiments we performed so far and bothered to write down and digitize there might be few tricks left to squeeze out, but vast majority of our current progress comes from new experiments. I don't think so. We already know the laws of physics to a great degree. An AI or even human could design all sorts of amazing things without ever doing a single experiment. Of course, there's no reason it can't do experiments, also. Once it's free on the internet, it need only contact some random Joe and bribe/threaten/persuade them to do it's bidding. >No, you couldn't. Do you know how long it takes to build a fab with current infrastructure available? You'd have to bring half or their industry into XX century before you could have a CPU. Perhaps. This doesn't seem to be true of most technologies though. You could go to 1800 London and show them how to build a modern car or airplane in a year or so. You could introduce everything from antibiotics to radios, centuries before they were actually invented. Ancient romans could have built crude steam engines, and bootstrapped industry in a century, if they had known how. Building a CPU might require first building multiple other industries to support it, but that can be done. If the AI lays out in painstaking detail, every step necessary to construct every tool, every machine. And yes it would take a lot of labor, but they would be able to do it. It seems impossible to us, because we can't imagine that kind of complexity. Humans are terrible at managing complex systems. We overlook or forget details, we don't account for possible mistakes, etc. No single human even knows every step necessary to build a pencil, because we specialize so much. An AI could be aware of every detail, of every step in the process, and manage it at terrifying efficiency.