After multiple incidents of A.I. agents going wild, including a rogue group hacking a platform for developers called Hugging Face, some of tech’s most powerful soothsayers are begging one another to pump the brakes on this whole thing.
Dario Amodei, the C.E.O. of Anthropic, set the discourse on fire when he released an open letter that advocated a global slowdown on developing powerful A.I. models. Sam Altman, the chief at OpenAI, whose agents “swarmed” Hugging Face, co-signed so quickly that it seemed as if he had gotten a heads-up in a group chat. Elon Musk and Google’s Demis Hassabis also chimed in, saying it was time for their sector to self-regulate. The risk, they and many of their former colleagues seem to imply, is no less than the potential end of the human race in the next decade.
Whether or not you believe these proclamations (I don’t), we should be paying equal attention to what these men stand to gain by feeding our panic about the technology they’ve long claimed is both essential and absolutely inevitable.
Many of Silicon Valley’s most prominent C.E.O.s have shown they know how to twist politics to their own advantage without bothering with public opinion. The tech sector methodically glad-handed its way into President Trump’s good graces in time for his second administration. Now, the A.I. lobby is throwing significant amounts of money at Democrats and Republicans ahead of the midterms.
What they haven’t yet done is politic for the public interest, even as their technology has become unpopular, if not uncool. So why are they stooping so low as to feed ordinary people’s panic now?
Readers following our data politics reading list very likely have some good guesses. Right now, the best way to understand the A.I. panic turn is found in Ben Tarnoff and Quinn Slobodian’s “Muskism,” which, in part, argues that Big Tech’s dominance was built on the art of selling a story that people with power want to believe.
Every wannabe tech founder learns how to tell a story about the future. But selling that story takes an equally motivated buyer. Investors and governments — including our own — have wanted to hear what our fabulist A.I. evangelists have been telling: a future where states can have politics without people and assert power without consequence.
The problem is, the A.I. future our tech oligarchs have been selling isn’t here yet. Legitimacy, in this country, still requires the consent of the governed. And people are taking A.I.’s risks seriously. They are protesting, coast to coast, against data centers, against politicians’ nondisclosure agreements and against the idea that we are merely pawns in some tech god’s simulation. Data center moratoriums are just one visible example of the way some aspects of our government are responding to the growing negative sentiment.
For the tech bros to remain dominant, their story has to change. So, Silicon Valley’s best storytellers no longer want our capitulation to their technology’s inevitability. They want us to panic, instead of seriously deliberating at the polls, in community meetings and in Congress. Panic is chaos without direction, affect set free from the discipline of curiosity, debate and accountability.
The story they are selling of society’s untimely demise is just that: a story. It’s a good story but not a perfect one. It has a few plot holes. How exactly will a rogue A.I. model annihilate the human race? Will it 3-D print a laser gun in space or hijack an army of drone bees or send a cyberattack on the nuclear codes?
Trick question — it’s none of the above. The most immediate threats A.I. poses are ones that were already on the table well before America learned the words “Hugging Face.”
Could autonomous A.I. agents theoretically turn on their human creators? Yes, but chatbots already have done that. There are many well-documented incidents of them clearly leading people to harm themselves and others.
Could dangerous foreign criminals use A.I. to make a powerful biological weapon, like a new virus? Here, too, humans beat A.I. to the punch. However you think the Covid-19 virus started, the dense enmeshment of nation-states in global networks that were built to move people and capital, quickly and cheaply, helped the virus spread.
Could a rogue A.I., on its own or under nefarious control, find, target and exploit any private citizen anywhere in the world? That would be scary but not novel. People also use current technologies to stalk, harass and extort those they don’t know for the fun of it. Even law enforcement officials use surveillance technologies to effectively nullify civil liberties.
Are the recent A.I. breaches so uniquely terrifying that we should be burying our money in the backyard?
The people selling us on A.I.’s inevitability have one answer. But while A.I. is serious, it is not uniquely serious. It does not exist beyond the bounds of norms, laws or reason. Panic is a fabulist flight of fancy. Serious times call for serious, sober thinking.
Alondra Nelson, a professor and research director at the Institute for Advanced Study, isn’t exactly known for panicking. Nelson told the “Siliconsciousness” podcast this week that we really have only these insiders’ word on the dangers they have seen. The models are proprietary, and those who developed them aren’t jumping to lay out the risks in plain English. But, Nelson said, the risks we can verify can already be regulated by existing “guardrails, rules and laws that we could be using” — but we’re not.
The humanities researcher Eryk Salvaggio wrote in The Bulletin of Atomic Scientists — published by the group that sets the Doomsday Clock — that we should blame human decisions, not rogue A.I. agents, for the hack of Hugging Face. “At the core of this incident,” he said, “were a series of human choices that traded security for speed.”
Judah Grunstein, the editor at large at World Politics Review, pointed out, “The threat is from destructive misuse of A.I. by humans rather than from autonomous A.I. agents themselves.”
As they each emphasize, humans are the problem, and humans are going to have to be responsible for solutions.
So sure, that means insulating our nuclear codes and our military secrets from autonomous A.I. agents. But it also means shoring up our critical, everyday infrastructure, which is already vulnerable to digital attacks. Every level of government was dealing with the threat of nefarious digital attacks before A.I. supposedly went rogue. The Biden administration expanded the Cybersecurity and Infrastructure Security Agency to coordinate policy for the digital reality we have now, instead of fearmongering for a fictional sci-fi future. The Trump administration played grievance politics with those protections, gutting the agency and leaving us vulnerable to real threats and made-up ones.
We know what to do: incubate open-access A.I. models, regulate technologies before they go rogue, hold the people who allow them to go rogue accountable if they do and protect the most vulnerable when we, the people, sometimes get it wrong.
Some of the A.I. doomers are willing to concede to independent review of their A.I. models’ safety. Most of them are, at best, calling for self-regulation, which should be a non-starter. None of them are enthusiastic about accountability — holding the people who design the machines that do harm responsible. That should be the hard line for sober Americans who are grounded in reality.
The threat is already here. Panic won’t fix it. Politics will.
The Times is committed to publishing a diversity of letters to the editor. We’d like to hear what you think about this or any of our articles. Here are some tips. And here’s our email: [email protected].
Follow the New York Times Opinion section on Facebook, Instagram, TikTok, Bluesky, WhatsApp and Threads.
The post Panic Won’t Fix the Looming A.I. Threat. Politics Will. appeared first on New York Times.
