4 ms·
I think that AI alignment has come to mean a whole lot of things to different people, and it's all under the same umbrella. * Aligning with the user's intent,
by mindvirus 3y ago
I think that AI alignment has come to mean a whole lot of things to different people, and it's all under the same umbrella.
* Aligning with the user's intent, especially in the face of ambiguity
* Aligning with American left-wing/Christian sensibilities
* Aligning with safety/laws (don't tell people how to commit a crime, don't accidentally poison them when they ask for a recipe)
* Aligning with a company's public image (don't let the AI make us look bad)
I think these are all important things to explore, but because they get bundled together, to the author's point it makes it hard to work with them.
- sharemywin 3y agoI think the researchers that coined the term mean don't accidentally or on purpose kill people.
- ftxbro 3y agoIt's true and after this change of meaning they have tried making other names like "dont-kill-everyoneism" or "AI-not-kill-everyoneism" for it (not joking) but they are worried if it will sound too silly or alarmist or if it won't catch on. That said, language does its own evolution without regard for what the ones who coined it had meant.
- mindvirus 3y agoRight, and I think that's mostly the first point. When the user says "make paperclips efficiently", they (hopefully!) don't mean "step 1: destroy humanity so I can't be turned off".
- jahewson 3y agoNo it’s mostly the second point. This is why they speak in euphemisms.