The A.I. Dilemma - March 9, 2023

Post Reply New Topic

A hour long presentation by Tristan Harris and Aza Raskin discussing the information threats from AI.

They're both talking about a double-exponential curve for LLM progress. The LLM's are also teaching themselves new capabilities that weren't put in so even putting safety checks on various topics is tricky as it's happening without the researchers knowledge. The way they describe this is that everything - symbols, video, audio, pretty much anything that gets passed to it gets translated into language. One of the scarier examples of this is that ChatGPT emerged research chemistry skills and that was open to the public without filtering out whether it's nerve agents and things of the like. They also mentioned, like AlphaGo where an AI was able train against itself at playing Go, you'd have skills like AlphaPersuade, AlphaFlirt, etc.. Add that to deep fakes, being able to take a three second clip of someone speaking and turn it into an entire speech in their vocal style. Also - automated exploitation of weaknesses in code. Automated individual research and blackmail.

It kind of sounds like Batman Begins with the bad acid in the water, just that in this case it'll be on our devices. I'm already thinking I may want to get a much stronger password for my router and see if I can force long timeouts for anything over five failed guesses.

Someone might figure out a way to align this stuff relatively soon but - my honest opinion - I think I'm going to be a lot more careful what kind of information I put out on the web and in particular I do worry if we're going to have some month in the summer or fall where everyone's credit cards get hacked just because a given country figured out a good way to do cyber warfare against their economic and military competitors.

I think the biggest headf--- is how it is you even plan for the future in this environment, or where can you put what money you do have without it getting hacked by AI (is online banking for example sufficiently set up to deal with these threats?). I guess ostrich head-burying has some uses like 'keep doing the same thing until a new convention emerges', and if things really go off the rails don't worry - civilization might be gone so we won't be here to need to worry about it anyway.