AI judges make a lot of sense, that way everyone is treated equally, because eveey judge thinks literally the same way. No corrupt judges, no change in political bias between judges, no more lenient or strict judges that arbitrary decide your fate.
How you decide what AI model is your judge is a whole new can of worms, but it definitely has lots of upsides.
I mean this with the biggest offence possible: AI judges make no sense, atleast with the current way of doing AI (LLMs). It’s been known for years that they amplify any bias in their training data. You are black? Higher chance of going to prison and longer serving time. Getting divorced and are male? Your ass is NOT getting custody. Hell, even without that the LLM might just hallucinate some crime not in the data for a case and give you a life time prison sentence. And if you somehow manage to avoid all that, what’s stopping somebody from just shadowprompting it and getting the judgement they want? It would also be an easy target for corruption, the goverment wants their poltical rivals gone? Tweak the model so it’s just that bit harsher, or just a bit more alligned with some other interpretatnion of the law.
Who would even choose the training data? The judges? Why would they, it would be better for them to sabotage and keep their jobs. Some goverment agency then? Don’t want to do that, or you’re gonna find out separation of power has a reason.
You have been found guilty of jaywalking. I hearby sentence you to 90 days of community service as unicorn titty sprinkles from Valhalla. May Chester have mercy on your handkerchief.
And how will this be done? A proper legal system needs impartiality, which an AI still varies as much or more than a human judge. Not to mention, the way it’s trained, the training data itself, if there are updates to it or not, how much it thinks, how it orders juries and parties, etc.
If, in theory, we have a perfect AI judge model, how should it be hosted? Self host it? Would be pretty expensive if it needs to be able to keep up. It would have to be re-trained to recognise new legislation or understand removals or amendments of laws. The security of it? If it needs to be swapped out often, it would need internet access to update itself, but that produces risk for cyber attacks, so maybe done through an intranet instead?
This requires a lot of funding, infrastructural changes and tons of maintenance in the best case scenario where the model is perfect and already developed. There would be millions, or ideally, billions in funding to produce anything remotely of quality.
AI judges make a lot of sense, that way everyone is treated equally, because eveey judge thinks literally the same way. No corrupt judges, no change in political bias between judges, no more lenient or strict judges that arbitrary decide your fate. How you decide what AI model is your judge is a whole new can of worms, but it definitely has lots of upsides.
I mean this with the biggest offence possible: AI judges make no sense, atleast with the current way of doing AI (LLMs). It’s been known for years that they amplify any bias in their training data. You are black? Higher chance of going to prison and longer serving time. Getting divorced and are male? Your ass is NOT getting custody. Hell, even without that the LLM might just hallucinate some crime not in the data for a case and give you a life time prison sentence. And if you somehow manage to avoid all that, what’s stopping somebody from just shadowprompting it and getting the judgement they want? It would also be an easy target for corruption, the goverment wants their poltical rivals gone? Tweak the model so it’s just that bit harsher, or just a bit more alligned with some other interpretatnion of the law.
Who would even choose the training data? The judges? Why would they, it would be better for them to sabotage and keep their jobs. Some goverment agency then? Don’t want to do that, or you’re gonna find out separation of power has a reason.
Bad idea.
Perhaps when we have real AGI, but I wouldn’t want an LLM to decide someone’s fate.
You could get away with murder if your lawyer talked the charges out of it’s context token limit.
And how will this be done? A proper legal system needs impartiality, which an AI still varies as much or more than a human judge. Not to mention, the way it’s trained, the training data itself, if there are updates to it or not, how much it thinks, how it orders juries and parties, etc.
If, in theory, we have a perfect AI judge model, how should it be hosted? Self host it? Would be pretty expensive if it needs to be able to keep up. It would have to be re-trained to recognise new legislation or understand removals or amendments of laws. The security of it? If it needs to be swapped out often, it would need internet access to update itself, but that produces risk for cyber attacks, so maybe done through an intranet instead?
This requires a lot of funding, infrastructural changes and tons of maintenance in the best case scenario where the model is perfect and already developed. There would be millions, or ideally, billions in funding to produce anything remotely of quality.
All I see are downsides.
if it’s designed right, it would be great. otherwise it would suck