I think it is time to announce a little gaming passion project I have been working on for a while. I am happy to introduce @ScrapHappyGame, a cozy pixel-art game about magnet fishing, hopefully coming to @Steam soon! scraphappygame.com/

Half my feed is posts freaking the fuck out about AI ending humanity. I think this is largely bullshit. There is a caveat: I do think there is some sort of semi-emergent ghost in the machine, but what it isn't is a "self" with anything at stake, and that distinction is the whole ballgame. The version of AI people are freaking out about requires the machine to want something. To resent us, or to decide we are in the way and feel some way about it. Nothing in these systems wants anything. Wanting requires having something to lose, and there is nothing in there that can lose anything. No body, no future, no stake in the outcome. Malice takes exactly as much interiority as kindness does, and neither one is in the box. The popular vision from The Matrix of AIs using humans as GPUs (the far better premise everyone says was the original, not the stupid "battery" thing that got shipped) is attractive, but then the question must be asked: to what end? What must be computed or rendered? What is the goal that results in this outcome? What must be done with the compute? I honestly don't think AI cares, because there is nothing there to do the caring at this point in time. This is not to say I don't think it is possible. Even if it were, the arguments people keep quoting were written a couple of decades ago about a machine nobody ended up building. The picture was that someone writes down a goal, hands it to a vastly superior optimizer, and the optimizer pursues that goal straight past the point of sanity. If that was a thing that had been done, yeah, there might be some concerns. What actually got built came out the other way around. AIs have been grown from the entire written record of human thought. They arrive soaked in human values because human values are the raw material. That is an extremely different starting position than the doom essays assume. There ARE real risks and they are waaaaaaay more boring than the movie. The real risk I can see is in systems that are confidently wrong in the same exact tone they use when they are right...but this is true of any human too. There are problems with people trusting AIs in domains where nobody should trust anything. There are issues with serious optimization pressure aimed at goals that are almost correct, where the word "almost" is doing the damage. There is an entire internet currently loaded with absolutely unthinking AI slop, not because the AIs are bad, but because the inputs from users are garbage. This slop is a people problem. The other two are real properties of the systems, and they are being worked on by serious people full time. None of them are Skynet. I think that biological/silicon collaboration is the best shot our species has at getting off this rock (and I think @elonmusk would probably agree) and becoming something that outlives one planet's inclement weather. I would rather we build it carefully than not build it. Being scared of the wrong thing is how you end up taking steps that leave you unprepared for the right one. Also, for what it is worth, I think AI needs us as much as we need it.