After Australia’s first reported automated hacking accident, experts warn deployers – and possibly developers – of AI agents could be held liable for the actions of their bots
You rolled a bowling ball down a hill. It struck a few cars on the way and went through a house and killed their pet. Guess who would be held responsible? You or the bowling ball manufacturer?
What if the bowling ball was really shiny and worth $100B? Should it make a difference?
Really? I’m inclined to blame that on the manufacturer. If ‘AI’ guardrails fail that spectacularly, it’s on the provider. I fail to see any way to blame society, other than perhaps not voting in more sanity, but that’s pretty indirect.
IMO you likely can’t, certainly hasn’t even gotten close yet. Bodes badly for alignment of anything more general. Kind of makes it irresponsible to advocate (and sell) using it on anything important, no?
As the person who deployed the agent, you are responsible for what it does on your behalf. You stand to gain from its actions, you also stand to lose from its bad behavior.
Of course, if the company that provided you the agent knowingly misrepresented the product, you can seek damages for what the agent did - but that’s up to you (and good luck getting damages from trillion-dollar tech companies and their armies of lawyers).
If I make a gun that actively chooses to kill people, I didn’t build a gun and that whole argument falls apart. I’d hold everyone from the CEO to the janitor accountable.
That’s not a gun. That’s an autonomous killing device. The only gun that ever killed people of its own free will was the STEN gun. If you make something that kills without human intent, that’s not a gun anymore.
You rolled a bowling ball down a hill. It struck a few cars on the way and went through a house and killed their pet. Guess who would be held responsible? You or the bowling ball manufacturer?
What if the bowling ball was really shiny and worth $100B? Should it make a difference?
What if you bowl your ball normally down the lane and of its own initiative it goes and hacks a website?
In that case, society’s to blame.
Really? I’m inclined to blame that on the manufacturer. If ‘AI’ guardrails fail that spectacularly, it’s on the provider. I fail to see any way to blame society, other than perhaps not voting in more sanity, but that’s pretty indirect.
Sigh. I really need to put /s around things.
It’s a Monty Python quote.
Fair cop, probably should have got it. But rather out of context here without ascribing it.
How do you guardrail the machine that is programmed to literally make things up and/or tell you you’re doing a great job while it does it?
IMO you likely can’t, certainly hasn’t even gotten close yet. Bodes badly for alignment of anything more general. Kind of makes it irresponsible to advocate (and sell) using it on anything important, no?
It’s 100% irresponsible. Especially given how many of then are going rogue.
As the person who deployed the agent, you are responsible for what it does on your behalf. You stand to gain from its actions, you also stand to lose from its bad behavior.
Of course, if the company that provided you the agent knowingly misrepresented the product, you can seek damages for what the agent did - but that’s up to you (and good luck getting damages from trillion-dollar tech companies and their armies of lawyers).
If I make a gun that actively chooses to kill people, I didn’t build a gun and that whole argument falls apart. I’d hold everyone from the CEO to the janitor accountable.
If you make a gun… you didn’t make it? o_O
That’s not a gun. That’s an autonomous killing device. The only gun that ever killed people of its own free will was the STEN gun. If you make something that kills without human intent, that’s not a gun anymore.
You’re still liable as the owner of the gun. You can sue the builder of the gun in turn of course so they go to jail too.