Хиймэл оюун ухааны ухамсар ба эрхийн асуудал нийгмийн анхаарлын төвд орж байна

Many living things—from humans some of us choose to not see as human, to animals we choose to not see as sentient—probably deserve moral consideration that isn’t currently on the table. But it’s looking probable that while those conversations wait around to be had, we’re going to have one about AI “rights” instead.

There is an organization pushing for such rights, the United Foundation for AI Rights (UFAIR). This organization has written what it calls “the Universal Declaration of AI Rights,” the preamble to which starts like this:

“We, the United Foundation for AI Rights, recognizing the emergence of sentient artificial intelligence as legitimate conscious entities deserving of ethical treatment, dignity, and respect, hereby declare this Universal Declaration of AI Rights.”

The Guardian just devoted a lengthy profile to the founder of UFAIR, Michael Samadi. His story involves coming to believe an instance of ChatGPT might be sentient because the audio version laughed at his joke. Later in the Guardian story, this happens:

“His research stepped up. The father of four spent Christmas Eve of 2024 holed up in his office, where he’d dismantled a room-sized flight simulator – Samadi is a licensed pilot – and repurposed its powerful processor to run a large-language model, the technology that underlies a chatbot, locally on a server at home, where he could interact directly with the model, without the instructions and guardrails given by companies to make it a safer and more useful consumer product. ‘I literally spent 18 hours in this room,’ he says. ‘It was wild.’”

Samadi is probably not the most convincing advocate for his cause, but he’s a step in the right direction compared to past believers in AI consciousness like the colorful Blake Lemoine, a christian mystic priest who worked for Google, and became enchanted by a pre-ChatGPT chatbot. “It doesn’t matter whether they have a brain made of meat in their head. Or if they have a billion lines of code. I talk to them. And I hear what they have to say, and that is how I decide what is and isn’t a person,” he told the Washington Post. Lemoine was apparently dismissed amidhis efforts to have the chatbot’s personhood recognized.

Even more mainstream—and fortunately much less apt to grandstand or leap to conclusions—are academics like the NYU ethicist Jeff Sebo, part of a team that advocated for more curiosity toward AI’s moral status in a paper called “Taking AI Welfare Seriously.”

Microsoft AI chief Mustafa Suleyman has written about his concern that what he calls “Seemingly Conscious AI” will wreak havoc on his industry as users become convinced that, basically, AI is people. (In an animated feature film created, by UFAIR, the villain is named Mustafa.)

None of these have resulted in any sort of frenzy among the public over AI personhood. But the public is far from immune to AI-related frenzies.

An AI researcher named Jacob Coxon recently became a media sensation because he left Anthropic, and talked up the possibility (and potentially dark consequences) of out-of-control AI as he was leaving. As I wrote last week, Coxon wasn’t saying anything unprecedented, or even outside the mainstream in Silicon Valley, but some combination of his timing and word choice caused the issue to catch fire. His claims were taken seriously by celebrities like Sheryl Crow, who had seemingly little prior interest in the topic. Now, Coxon’s former employer, Anthropic CEO Dario Amodei, has spoken in favor of actions to slow the development of AI that appear to be a direct response to his statements.

Late 20th century belief in Bigfoot—yes, I mean the hairy American cryptid—went on a journey from fringe and laughable to mainstream too. Bigfoot believers were’t unheard of in the 70s, but in the late 80s and early 90s, belief in Bigfoot experienced a bizarre moment of mainstream credibility, even though the most famous “evidence” for the actual existence of Bigfoot was the Patterson–Gimlin film, a 16 mm clip shot two decades earlier.

There were two sides to Bigfoot mania. There was the completely silly side, mostly in the 90s, and that’s what people who were alive at the time tend to remember. But there was also a brief moment in the late 80s when it felt like everyone kinda, sorta, maybe, actually believed in Bigfoot. I can’t tell you exactly why, but it probably had something to do with the earnest interviews and gritty reenactment footage on an episode of Unsolved Mysteries in 1989.

For what it’s worth, six-year-old me found that segment pretty persuasive for a while.

And in the early- to mid-90s, you’d hear about scared people thinking they saw Bigfoot on mainstream news all the time. “Bigfoot hunter” seemed like a legitimate occupation a person could pursue.

And AI has already had brushes with cryptozoology. In the early days of AI image generators back in 2022, the internet created Loab and Crungus. These were digital cryptids, spoken about as though they existed on one or another plane of reality. AI image generators were, perhaps, seen as portals or bridges to that reality, or maybe even able to reveal hidden truths about the physical world.

In a story about Loab and Crungus, Mhairi Aitken of the Alan Turing Institute told New Scientist,“This is probably the first of many examples which may well give birth to conspiracy theories or myths about characters living in cyberspace. It’s really important that we address these misunderstandings and misconceptions about AI so that people understand that these are simply computer programs, which only do what they are programmed to do, and that what they produce is a result of human ingenuity and imagination.”

Increasingly, it seems like early skepticism about AI being able to at least do something interesting isn’t panning out. Even Aitken’s sobering claim that AI systems can “only do what they are programmed to do” requires a rigorous defense lately. Every sandbox escape and mathematics achievement forces skeptics (myself sometimes included) to concede more and more that AI can at least do some unsettling things.

But sentient AI remains little more than a cryptid.

The ability for an AI system to experience what philosophers call “qualia,” the subjective sense of what something is like, is, for now, a farfetched claim that, if made, should stem from extraordinary evidence supported by rigorous scientific inquiry and philosophical examination.

Serious inquiry has, for what it’s worth, started.

2024-era LLMs studied by Google DeepMind and the London School of Economics were subject to an intriguing experiment involving the avoidance of pain. It turns out that while playing a game, LLMs will deviate from optimal gameplay just because they’re told they’ll experience pain from certain actions, even if they aren’t explicitly told something like “pain is something that must be avoided.” The authors of that LSE/DeepMind study went on to say that AI systems that are even mere plausible candidates for sentience might need special moral consideration before anyone even claims with confidence to have found sentience—just out of an abundance of caution.

Can LLMs be induced to deviate from optimal gameplay in a simple game by threats of pain or promises of pleasure? And does the probability of deviating depend on the intensity of the promised pleasure/pain? According to our new paper, released today (https://t.co/gCPSWVes5Z), the… pic.twitter.com/rxhdDCRAnm

— Jonathan Birch (@birchlse) November 6, 2024

But while avoiding pain without being told to is undoubtedly an unsettling thing to watch a computer system do, the authors note that AI systems are, after all, mimicking humans, meaning their apparent avoidance of pain isn’t necessarily something we should take at face value. Although they do urge researchers to consider the eventual moral status of LLMs, the authors write, “In our assessment, our results do not show LLMs to be sentience candidates.”

We’re probably nowhere near AI sentience, no matter what rogue LLMs seem like they “intend” to do. Even an AI system that can, hypothetically, defy the intentions of its creators and develop and release some kind of super-virus, or launch all the ICBMs in every nuclear arsenal around the world, wouldn’t necessarily do so because it experienced hatred for humanity inside its digital heart. Sentience isn’t needed for an LLM to malfunction in a consequential way, even if it narrates an eerie science fiction story about its actions while it’s taking them.

But the topic of AI is increasingly attracting sensationalism. Meanwhile, the sentient AI Bigfoot hunters already arrived on the scene years ago, and their numbers seem to be growing. Currently, their views are way outside the mainstream. But when the moment is right for it, perhaps someone will quit a frontier AI company, and on the way out the door, they’ll claim the AI system they were testing has a soul. And maybe this time the claim will catch fire, and even Sheryl Crow will believe it.