Live data from Hacker News

We must not grant AI agents legal personhood

ft.com

81–90 of 122 posts

Re: We must not grant AI agents legal personhood

#81

Earlier quoted context omitted.

Not for the most important things, like murdering people, fraud, etc. Executives are prosecuted for these kinds of things and the mere fact that it can happen shapes their behavior. An AI cannot be punished at all. So they will not behave with that in mind.

I am sick of this 'just so' logic. I could equally argue that AIs are not driven by greed, sexual appetites, revenge, etc., making them considerably more trustworthy than wealthy sociopaths. It's already noteworthy when a very wealthy or politically powerful person is abruptly disempowered, even more so when they end up losing their personal liberty or life. I put it to you that's it's noteworthy because it is such a…

> I could equally argue that AIs are not driven by greed, sexual appetites, revenge, etc., making them considerably more trustworthy than wealthy sociopaths.

Imagine the consequences of trusting thousands of entities that each far exceed human intelligence, have agency, can cooperate, but have no feelings.

Re: We must not grant AI agents legal personhood

#82
I agree with the article's conclusion (though I don't think much of the reasoning that got it there). But I do think we should consider granting additional privileges to real humans as they interact with LLMs.

Suppose I have a local LLM running on my PC. I have close and personal conversations with it. It keeps a chat history. I'm later accused, falsely,* of a crime. During the process of defending myself, should the prosecutor be entitled to a copy of my LLM's conversation history that was seized pursuant to a warrant? Or allowed to interrogate my LLM, again seized while executing a search warrant?

Under current US law, the answers are probably yes. Just as an accused's diary or notebook can be seized and introduced into evidence, an LLM's internal state seems like a paper/effect/document that would be subject to search and seizure. But a conversation with a local LLM also seems more like a conversation with a spouse, therapist, or attorney. There are good public-policy reasons to treat those conversations as privileged; we want people to be able to get good advice without fear of disclosure. Shouldn't those reasons apply to LLM conversations as well?

I'm not saying LLMs should have personal rights. But I can see a case for privileges for real humans arising from conversations between real humans and LLMs. LLMs might not be human, but humans do interact with them as if they were.

*It doesn't matter, of course, whether the accusation is false. Due process means people have certain rights regardless of culpability. But I don't want readers to fixate on culpability, so in this scenario the accusation is unfounded -- let's say someone did break into a house and steal the family jewels, but a security camera identified the wrong person.

Re: We must not grant AI agents legal personhood

#83
post #78

Earlier quoted context omitted.

But you could still prompt it to act like any person, including fictional and deceased ones, which proves it has no real concept of self. It's just a language model for humans to generate whatever text meets the task. You could have it run for a year as different perspectives in GRR Martin's ASOIAF to finish his books for him. And then what, award it with Jon Snow's personhood? An AI agent would need some kind of sta…

Actually I see little problem with first part. If you initially prompt it to have certain personality based on some other personality even fictional one that is fine. I agree on the stability though. My point is that for personhood to start to make any sense it needs to operate that personhood continuously as well ending it would be ending a personhood. Which could be a murder...

So let's say the agent was tasked with being a deceased loved one continuously for a year. Do they then have the rights of the deceased person? What if the surviving loved one(s) then wanted the agent terminated because it was means to help with the grief and were now ready to move on?

There's a ton of scenarios where you can setup an AI to take on the personality of whoever, maybe other living people like someone famous. Granting the legal rights of personhood is going to be extremely problematic.

Is AI Jon Snow legally allowed to exist since George RR Martin has not given permission for anyone else to finish his books?

Will we have to abort our long-running agent persons before they reach the legal period for personhood?

Re: We must not grant AI agents legal personhood

#85
post #78

Earlier quoted context omitted.

Actually I see little problem with first part. If you initially prompt it to have certain personality based on some other personality even fictional one that is fine. I agree on the stability though. My point is that for personhood to start to make any sense it needs to operate that personhood continuously as well ending it would be ending a personhood. Which could be a murder...

So let's say the agent was tasked with being a deceased loved one continuously for a year. Do they then have the rights of the deceased person? What if the surviving loved one(s) then wanted the agent terminated because it was means to help with the grief and were now ready to move on? There's a ton of scenarios where you can setup an AI to take on the personality of whoever, maybe other living people like someone fa…

Ofc, they wouldn't be that person but a legally distinct person. And just because they start as something doesn't meant they are something. Most likely due to varying input they would drift from starting point and become something unique.

And going back to Voyager Doctor. That AI is a clone. Are all those clones same person or different person. Is only one run allowed? Or can you have multiple?

I am not saying that any AI running for a year should be given personhood. Just that such run is bare minimum point for even entertaining the idea at all. As if AI can't keep running at least that long they certainly couldn't upheld any type of obligations that could come up with it. Say just paying for the computing capacity they need or say taxes...

Re: We must not grant AI agents legal personhood

#86
post #60

Earlier quoted context omitted.

No one understands these things, which is why considering machine consciousness as a possibility is entirely reasonable. People who have a shallow model of how they work ("it's just linear algebra!") betray a lack of understanding.

I see no reason to seriously consider that current AI is in any way conscious. It's fun as thought experiment people like to play with. The technology is not some black box, it was designed by humans who understand it well enough to keep building and improving it.

My argument isn't that current AI is definitively conscious or that it will be in the short term. It's that nobody has a cohesive theory of consciousness, and that we should therefore approach questions of what's conscious and what's not with humility.

For LLMs in particular, it's notable that they both 1) originate from a lineage that attempted to model biological brains mathematically and 2) are the only other known entities in the universe able to reproduce many types of human intellectual labor. That is surprising and striking. Does that mean that they definitively share other traits with humans like consciousness? No, but it does make them one of the most interesting candidates, particularly when we have a very limited understanding of how humans themselves are conscious.

Re: We must not grant AI agents legal personhood

#87
post #60

Earlier quoted context omitted.

No one understands these things, which is why considering machine consciousness as a possibility is entirely reasonable. People who have a shallow model of how they work ("it's just linear algebra!") betray a lack of understanding.

I see no reason to seriously consider that current AI is in any way conscious. It's fun as thought experiment people like to play with. The technology is not some black box, it was designed by humans who understand it well enough to keep building and improving it.

We do not actually know what consciousness is, but we know we have it, in spite of our brains also just being a big blob of simple biological weighted functions that react to the hallucination our brains generate from external sensors we evolved. There is no guarantee other humans even perceive the same simulation we do!

The reality is we have no way to measure consciousness or how it emerges, or how to tell if other humans actually experience it the same way we do, so confident statements about consciousness being unable to emerge from simple weighted functions is baseless projection, wishing ourselves more special than an emergent property of the bag of weighted functions in our heads.

Re: We must not grant AI agents legal personhood

#88
post #11

we already have a mechanism to establish artificial persons, it is called incorporation.

IMO it was a mistake and everybody should be a sole proprietor.

This way legal consequences always have personal consequences and people should act accordingly

Re: We must not grant AI agents legal personhood

#89
post #23

Earlier quoted context omitted.

That's the key issue here. AI forces a re-examination of corporate personhood and corporate responsibility. Much of conservative US business theory is based on Milton Friedman's famous op-ed in the New York Times in 1970, "The Social Responsibility of Business Is to Increase Its Profits". [1] Go read the original, please. It's short. Friedman ends with "There is one and only one social responsibility of business — to…

>Anyone arguing against legal personhood for AIs needs to justify holding AIs to a higher standard than that. Or, conversely, allowing corporations to hold to such a low standard. No they don't. If a CEO orders an employee to murder somebody then the employee and CEO will both go to jail; there's always a human in the loop to be punished it the corporation does something blatantly illegal, so there's incentive to beh…

> If a CEO orders an employee to murder somebody then the employee and CEO will both go to jail.

In the US, death through willful corporate negligence very rarely results in criminal punishment. There are few cases.[1][2] The longest sentence was two years, and it wasn't a CEO.

CEOs do serve jail time for fraud now and then. But criminal negligence, no.

[1] https://blog.ehssoftware.io/safetyinsiderblog/summary-of-osh...

[2] https://www.fisherphillips.com/en/insights/insights/federal-...

Re: We must not grant AI agents legal personhood

#90
post #87

Earlier quoted context omitted.

I see no reason to seriously consider that current AI is in any way conscious. It's fun as thought experiment people like to play with. The technology is not some black box, it was designed by humans who understand it well enough to keep building and improving it.

We do not actually know what consciousness is, but we know we have it, in spite of our brains also just being a big blob of simple biological weighted functions that react to the hallucination our brains generate from external sensors we evolved. There is no guarantee other humans even perceive the same simulation we do! The reality is we have no way to measure consciousness or how it emerges, or how to tell if other…

> but we know we have it

Do we? We only know that at most one thinking person believes they have it, and everyone else claims they have it.

Post reply on HN