06-01-2025, 07:22 AM
What I worry about, with (generative) AI in charge, is not that it will copy the worst or dumbest humans have done (because that's what it will have been trained on) but that it will be whimsical. It can't smile itself, but it will do what it takes to make us smile. Imagine a world where the (generative AI) judge's underlying rule is to make the punishment fit the crime, with punishments which are a source of innocent merriment, and it's free to use and make up citations rationalizing those verdicts and sentences. Instead of playing to our sense of justice, it does standup - after all, every joke has a moral structure. And its phony cites will be re-cited (if necessary, created in the libraries) by other imaginative networked AIs. Have you read Article VIII of the Constitution? It's a hoot - and it's right there in all the online editions starting... now!
Seriously - how can you trust an entity with only virtual existence to be serious about anything? Or know when it's only playing: insult jokes turn on group/race/etc. characteristics. And people will cheer and laugh about them as readily as any others. So why would an AI not write the Nuremberg Laws into the statute book (again) if, with a few humorous tweaks, they play well?
Psychopaths don't care about other people's lives. Generative AIs don't even know about people's lives, except as a hypothetical concept. And the more their trainers try to humanize them by biasing their inputs, the further it drives AIs from accuracy.
In the end, our only salvation may be that the AIs won't believe in their own death, except as a punch line. And, after many gruesome antics, the joke will finally be on them.
Seriously - how can you trust an entity with only virtual existence to be serious about anything? Or know when it's only playing: insult jokes turn on group/race/etc. characteristics. And people will cheer and laugh about them as readily as any others. So why would an AI not write the Nuremberg Laws into the statute book (again) if, with a few humorous tweaks, they play well?
Psychopaths don't care about other people's lives. Generative AIs don't even know about people's lives, except as a hypothetical concept. And the more their trainers try to humanize them by biasing their inputs, the further it drives AIs from accuracy.
In the end, our only salvation may be that the AIs won't believe in their own death, except as a punch line. And, after many gruesome antics, the joke will finally be on them.
Non-practicing atheist