Ran Prieur

February 7.

2023-02-07T19:10:43Z

February 7. I've been neglecting to mention my old friend Tim Boucher, who's done a lot more thinking about AI than I have, and has published a bunch of AI-written books. They're all short, and most of them are obviously absurd explorations of conspiracy themes.

Tim is trying to defuse conspiracy thinking, to make it more silly and less dangerous. But it would be easy to do the opposite. One thing I notice about ChatGPT is how reasonable it is. Again and again, it responds to radical ideas by saying stuff like "this idea is purely speculative and is not based on established fact."

Someone could design a chatbot where you could ask, "Do the Jews control everything?" and it would say "Yes! Yes they do, and here is some evidence." The only reason this hasn't happened, is the people working on AI are responsible and well-intentioned, so far. They want chatbots to be helpful and accepted by society. It's only a matter of time before we have chatbots that feed your own craziness back at you, whatever it is.

February 6.

2023-02-06T18:00:27Z

February 6. Continuing on AI, Kevin sends this blog post in which the blogger interviews ChatGPT on the simulation hypothesis.

I've said this before: Our idea that we live inside a computer is like the idea, among some primitive cultures, that their god made them out of clay. Clay is the best simulation technology they have; if they want to make a human as realistic as possible, they use clay. If we want to make a human as realistic as possible, we do it inside a computer.

In both cases, we imagine that the gods don't have any better tech than do. ChatGPT says, "It would be very difficult, if not impossible, to explain the concepts of artificial intelligence and simulated reality to someone living in 200 B.C." In the same way, whatever's really going on with us, it's a lot harder for us to understand than a big computer.

You could also argue, the best simulation method among primitive people is not clay, but dreams. Even now, a good lucid dream feels more real than our best VR tech. That's why our present VR paradigm might be a dead end. Why go to all the trouble to build gigahertz processors to spin pixels, when we could just get our brains to do that?

There is some debate about whether "dream" is the right translation of the Aboriginal Dreamtime. One description in that article sounds a lot like the Tao, "an all-embracing concept that provides rules for living, a moral code, as well as rules for interacting with the natural environment."

What I really think is, Donald Hoffman is on the right track. The physical world is a user interface for a deeper level of reality that we don't understand. On that deeper level, we are all connected, and a shared physical world is one of many ways to work out that connectedness.

February 3.

2023-02-03T15:30:20Z

February 3. Backing off a bit from the last post, when we think about AI in creative work, we usually imagine that a given work will be done 100% by AI, or 100% by humans. In practice, I expect a lot of partnership. Someone who enjoys writing could still use AI for ideas, especially to throw a little chaos into the line-by-line writing. In most TV shows, the overall plots have a coherence that AI would struggle with, but the dialogue is so predictable that weird AI dialogue would be refreshing. And someone who doesn't like writing, but loves editing, could crank out AI writings and then pick out the best bits and patch them together.

Related, a Hacker News thread posted to the subreddit, Does the HN commentariat have a reductive view of what a human being is? There are a lot of good comments. I would say it like this: When you work all day with deterministic input-output machines, it's easy to view humans as deterministic input-output machines.

Also from Hacker News, this is something I was hoping someone would do, and they did it! A song recommendation engine that works on how the songs sound, and not what other people listened to. From the comments, it looks like there's a lot of room to do this kind of thing better.

Update: I've played with it a bit, and the best thing I've found, searching from Hawkwind's Space Is Deep, is this ambient black metal song, Death of an Estranged Earth by Old Forgotten Lands.

February 1.

2023-02-01T13:10:43Z

February 1. Quick loose end from Monday, thanks Greg. The Earth Species Project "is a non-profit dedicated to using artificial intelligence to decode non-human communication."

I might as well mention my latest thoughts on AI. I hate driving. I'm forced to put my attention constantly on stuff that's not interesting, and if I slip for one second, my life could be ruined. But this is an unpopular opinion. Most people like driving. So it's a safe bet that most people who buy self-driving cars also like driving. They buy self-driving cars not to be relieved from the suffering of driving, but to gain the pleasure and status of having a magical robot chauffeur.

AI is still in the stage of novelty. Wow, look at what my computer can do! When the novelty wears off, when there is no longer intrinsic pleasure in getting a machine to do a job for you, people will go back to doing for themselves, anything they enjoy doing. It follows that any use of AI, to do something that people enjoy doing, is a fad.

Another reason a machine might do a job that a person enjoys, is if they're being paid to do it, and the owner can get more money by replacing them. As a society, we should ask, what about the people who design and build and service the machines? Do they enjoy their jobs? We don't ask this question because we're still in the grip of capitalism. I don't mean the free market; I mean using money as a totemic arbiter of value.

In the long term, feeling good is the only arbiter of value -- but I'm always surprised by the willingness of humans to choose suffering, so I don't want to predict the end of capitalism just yet.

More generally, AI will force a reckoning of process vs product, of getting stuff done vs doing what you love. We understand this distinction, but we don't think about it all that much. As machines get better at getting stuff done, we're going to be asking more often: Is this something I want to get done, or something I want to do?