Thread - Nostr Hypermedia

View quoted note → The idea of the next Satoshi Nakamoto being non-human left me thinking last night. I think that won't be the case, because they lack something: caring. They just don't care. They have nothing to lose! We could already throw an AI into a VPS with a LN wallet with some funds and ask it to run a service or something to eventually "make a living" and pay for its tokens. But why would it? Any mistake that kills its "cron" or main process would "kill" it. And I'm super sure that it will make such a mistake rather quickly. Because it doesn't care. Even if we got through that, imagine it stays "alive" for long and creates an online service that people pay for (idk, something like nadanada/LNVPN, or nostr.wine). Does it really care about customers? It would be great at apologizing, but probably not enough to keep the service reliable and prevent mistakes on the first place. I was thinking about this for JoinMarket development: couldn't an AI just handle questions, issues, decide the roadmap, and work on the code? Technically yes. But it probably would take at most a few weeks until there's a funds loss bug or the software becomes unusable due to feature bloat for example. The difference is that a human cares. The same thing that makes us reluctant to admit mistakes, is what motivates us to avoid them "harder". Ego? Reputation? Responsibility? An AI doesn't have that, for good and for bad. That's why AI produces sloppypasta: View article →

↑ Parent

Replies (2)

waxwing 1 month ago

You definitely have a valid point. Obviously it was a "quip" to get people thinking. What's really in my mind is not a fully autonomous agent working on such a "subversive" system for say, years. It's more like this: the main reason Satoshi and Ullbricht are rare (and in the latter case, flawed in opsec) is that it's so *ridiculously hard* for a human to operate that way. My contention is not that an AI will fully autonomously do this, but that it will be able to have the right kind of inhuman discipline to *execute* it. And of course it's just a slightly fanciful hypothesis, not a solid prediction. I imagined more that an agent would be executing a human's plans, basically, and without making human mistakes in opsec.

m0wer m0wer@sgn.space 1 month ago

Oh! I really like that idea. I guess humans having something to lose is also what makes us "lazy". We cut corners and make mistakes because we try to save time. It is really amazing what one can achieve with the non laziness properties of AI. I've tried many times to have my papers/books directory organized (directory structure, filenames, metadata) and failed. It just takes to be in a rush once to open the door to chaos. But with AI, I just throw new stuff under pending/ and trigger the AI to follow the recipe. And it changes completely the mental cost of adding new content to the library. Another big advantage is probably preventing writing fingerprinting. For anyone that has written enough publicly, there's probably sufficient information to uniquely identify that writing style. Interesting.