The ‘machine defendant’: Who’s to blame when AI makes mistakes?

This has already been tried in at least one court.

There was that story a while back about the guy who was told by an airline's AI help-desk bot that he would get a ticket refund if turned out he was unable to fly, only for the airline to say they had no such policy when he came to claim.

He had screenshots and said he wouldn't have bought the tickets in the first place if he had been told the correct policy. The AI basically hallucinated a policy, and the airline was ultimately found liable. Guy got his refund.

And the airline took down the bot.

The corporation running the AI and the corporation using the AI. They should both pay the same fines. To be clear, two fines of the same size, not a single fine that's split.

The company. This would curb some waste energy.

Given the high costs of AI, isn’t it reasonable to assume that whomever stands to make a profit is equally liable for it’s outcomes?

When it comes to deadly "mistakes" in a military context there should be strong laws preventing "appeal to AI fuckery", so that militaries don't get comfortable making such "mistakes."

I suspect that it's going to go the same route as the 'acting on behalf of a company' bit.

If I call Walmart, and the guy on the phone tells me that to deal with my COVID infection I want to drink half a gallon of bleach, and I then drink half a gallon of bleach, they're going to absolutely be found liable.

If I chat with a bot on Walmart, and it tells me the same thing, I'd find it shockingly hard to believe that the decisions from a jury would in any way be different.

It's probably even more complicated in that while a human has free will (such as it is), the bot is only going craft it's response from the data it's trained on, so if it goes off the rails and starts spouting dangerous nonsense, it's probably an even EASIER case, because that means someone trained the bot that drinking bleach is a cure for COVID.

I'm pretty sure our legal frameworks will survive stupid AI, because it's already designed to deal with stupid humans.

I don't see any legal issue here.
When a person or a company publishes software that causes harm or damages, that person or company is fully liable and legally responsible.

Whether they themselves understand what the software does is completely irrelevant. If they don't have control over its output, they shouldn't have published it.

I feel like self driving cars are going to end up being the vanguard of deciding this, and I basically see it as mirroring human liability with a high standard where gross negligence becomes criminal. If a self driving car can be proven to be safer than a sober human, it will serve the public interest to allow them to operate.

Everyone involved

The morons who decided to use AI in places it shouldn't have been used in the first place.

I imagine there will be limits set, through precedent.

For example, if a customer is chatting with an AI bot regarding a refund for a pair of $89 sneakers, and the bot tells the customer to report to the nearest office to collect 1 million dollars, I can see the courts ruling the plaintiff is not owed $1 million dollars.

Although, if the plaintiff ended up flying a few States over to try and collect, maybe travel costs and lost wages? Who knows.

If a company marketing fee for service legal advice, their might be a higher standard. Say a client was given objectively bad legal advice, the kind that attorneys get sanctioned or reprimanded for, and subsequently acts upon that advice. I think it's likely the courts would take a different approach and determine the company has a good bit of liability for damages.

Those are both just hypothetical generic companies and scenarios I made up to highlight how I can see the question of liability being determined by the courts. Unless some superceding laws and regulations enacted.

Or fuck it, maybe all AI companies have to do is put an arbitration clause in their T&C's, and then contract out to an AI arbitration firm. And wouldn't you know it, the arbitration AI model was only trained on cases hand picked by Federalist Society interns.