Yes, but my point is that the only reason to grant legal personality to a "thing" is if you want to shield its owner from liability. If you're still holding the manufacturer liable for infractions committed by the algorithm, you don't need it.

Ah, but aren't companies themselves entities such that their owners and creators are liable about their actions in certain cases, but not in others?

Not a lawyer and not a founder, but as I understand it, it goes something like:

- If I own a company and it is insolvent or incapable of fulfilling some contract, I am probably not in trouble, the company is. It might still depend on the specifics, though.

- If I own a company and it does something illegal without my knowledge or intent, such as say produce unsafe products, I might be in trouble, or I might not be, depending on how unsafe, why and whether I can be found have been negligent in allowing it. The company itself definitely is in trouble, though, since it has clearly violated the law.

- If I own a company and I willingly task it, or even allow it, to do something illegal, with full knowledge of the illegal actions in question, then I am probably in trouble too.

Operating through a company doesn't make me immune from the law, and neither should working through an algorithm. On the other hand, full personal liability for everything my company/algorithm does, when I am not in the loop at the point the decision happens, seems like the wrong standard as well.

So basically the algorithm can be found guilty of a crime. The company that created it can then be found guilty of selling a defective product (depending on whether there was any reasonable way of catching the defect). The owners can be then charged with negligence or something to that effect if the defect is something that standard industry practice or compliance with regulations should have caught, or if it was introduced deliberately.

If someone has a law background here, feel free to let me know if I am misunderstanding anything...

The thing is: what's the purpose of finding "the algorithm" guilty of a crime? A company owns assets that can be used to pay fines and damages, and as a semi-autonomous entity (in the sense that "it" still exists even if all of its owners and workers are replaced) it can be prohibited or forced to do something. An algorithm, on the other hand, owns nothing and its actions are wholly dependent on the people that program and configure it.

If the purpose is to account for freak events where the algo misbehaves for reasons that couldn't be foreseen, is that any different than a (non-autonomous) vehicle part failing despite being well-designed and maintained? This is not a new situation - an algorithm is just another car part. You don't need to give legal personality to the fuel injection pumps or the steering column in case it fails unexpectedly.

It's entertaining that apparently one of our signifiers of independent personhood is "Can drive a car."

This issue is only confusing to the extent this isn't considered.

>an algorithm is just another car part

Exactly. And the designer is on the hook for negligence in the usual way - but not for malicious, tragic, and/or unforseeable circumstances that may lead to death or injury.

Finding the boundary between those will keep a lot of lawyers very busy. But I don't think we need to start assigning personhood and agency to a steering system just yet.

With the comment that sparked off this thread, I wasn't implying the car had any sense of agency. Rather, that legal personhood might be a good way to sort out the laws.

A corporation is a legal person, which makes it a lot easier to enter a contract with one, but that doesn't imply it has rights. You can own a corporation without it being slavery.

The purpose I had in mind when I first mentioned the idea was just to provide a convenient way to reason about the law. It lets you re-use laws written with human drivers in mind, instead of having to individually recontextualize each as a type of mechanical failure. So you can charge the car with reckless driving, instead of having to show that an algorithmic failure lead to a behavior consistent with reckless driving.

It's still an abstraction over something akin to mechanical failure, like how corporate personhood is an abstraction over a group of people. It doesn't mean the algorithm has rights, any more than a fuel injection pump does. You also can't murder a corporation, even though it's a legal person.

"Operating through a company doesn't make me immune from the law, and neither should working through an algorithm."

That might be technically true but in reality it quite often does. See nearly all bankers from 2K8 melt down for good examples of how money and power buy immunity in what is really an economically based caste system.

Do you really think if this was some garage start-up and not Google that they would have been granted anything but a swift kick in the ass out the office door of whatever bureaucrat handled this?

Could it not also be used to constrain liability? If "Joe" has its license suspended, the manufacturer can slightly alter the algorithm and name it "Sue" and continue on.

I'm sure the car manufacturer would love that, but why would the legislative body - and society in general - create such a system?

Sounds reasonable to me.