Cliff’s Note:
Here is something I did not have on my bingo card. Over the weekend, Dario Amodei, Sam Altman, and Elon Musk — three men who have spent years telling us the race is the point — all said out loud that somebody needs to slow the race down. When the arsonists start asking for a fire code, you pay attention.
The White House called it a Chinese hoax.
David Shuster spent this week on what happens next, and the back half of this piece is the part I would pay for: what the people inside that building are quietly telling him to do about it, and why he probably will not. That reporting is for members.
Sixty dollars a year, five bucks a month. That is the whole business model — no institution, no sponsor, nobody upstairs. Just you, and it is the reason David can write this without asking anyone’s permission.
—Cliff
by David Shuster
In golf, it is a one foot putt. In basketball, it is called a wide-open layup. . In soccer or hockey, it is the shot taken from three feet away at an empty net.
Whatever the terminology, Donald Trump now has one of the easiest political plays imaginable sitting in front of him.
It is also a matter involving the continued existence of the human race.
And yet there is every indication that Trump may find a way to miss.
The subject is artificial intelligence.
This past weekend, the leading competing titans of the A.I. industry each said A.I. research needs to slow down.
Anthropic CEO Dario Amodei called for governments to “pace” frontier AI development, arguing that capabilities are advancing too quickly for existing safeguards to keep up. Open AI’s Sam Altman said he agrees and supports greater coordination. Elon Musk weighed in and said Amodei and Altman are correct and that there needs to be government mandated and enforced restraint. Other prominent AI figures are also advocating for independent safety evaluations and government involvement.
Front line AI researchers have been issuing dire warnings for months. Most recently and prominently, 27-year-old researcher Jacob Coxon quit his job at Anthropic and declared, in a now viral social media post and several interviews, that OpenAI and Anthropic have been working irresponsibly. Coxon declared that people building the technology genuinely believe it could kill everybody by the end of the decade.
For a primer on the threats of superintelligence and the key issues, please read this column by my colleague Lawrence Winnerman .
As Lawrence notes, some of the people engaged in this race think the chances are high that it will trigger a human extinction event.
The question is no longer whether A.I. will transform civilization. That part is settled. It will. The question is whether civilization will remain in charge of the transformation and live to use it.
I believe that A.I. could produce miracles. It may accelerate the conquest of cancer, transform medicine, improve education, increase productivity beyond anything previously imagined, and solve problems that have tortured humankind for centuries. It may someday make the ordinary human workweek look like an eccentric historical curiosity.
Great.
But do we have to try and reach this paradise by conducting the most consequential technological experiment in human history according to the business philosophy of a Las Vegas casino? Build faster. Deploy faster. Beat the other gambler. And if you lose, oh well. There is always another table. Except in this case, perhaps everybody dies.
The sensible proposal is not to ban A.I. Nor is it to strangle technological development or voluntarily hand America’s technological lead to Beijing. The sensible proposal is considerably less dramatic: establish rules for the most powerful systems so this is not a race in which nobody can afford to brake.
That means independent testing. Reporting requirements. Serious security standards. Restrictions on deploying systems that fail clearly defined safety evaluations. And legal mechanisms permitting competing companies to share certain safety information without being treated as conspirators against the sacred interests of commerce.
None of this is exotic or extreme
We already regulate banks, airplanes, nuclear reactors, pharmaceuticals and other things that can produce unpleasant consequences when operated carelessly. Civilization has occasionally learned that “move fast and break things” is not an adequate philosophy when the thing being broken happens to be civilization.
The central problem is a simple one.
Every leading A.I. company may rationally believe that slowing down would be safer. But each company also fears that if it slows down while its competitors continue racing, it will lose the race.
So everyone in the A.I. industry accelerates.
This is not necessarily because everyone is insane. It is because everyone is responding rationally to an irrational set of incentives involving trillions of dollars in potential revenue.
That is precisely the kind of problem government regulations exist to solve.
The strongest objection, as Donald Trump has pointed to, involves China. If the U.S. slows down, would Beijing? A unilateral American retreat could surrender an important economic and strategic advantage. And as Trump said recently, “whoever wins the A.I. race wins.”
But regulation and technological leadership are not opposites. The United States could seek international agreements governing the most dangerous capabilities while continuing to invest aggressively in research, computing, semiconductor manufacturing, robotics, and less hazardous applications. Indeed, American leadership may be more valuable if it produces standards that other countries ultimately adopt.
The easiest part of all of this involves U.S. domestic politics.
Trump’s approval/disapproval rating is now the worst of any U.S. President in fifty years. And in his handling of Artificial Intelligence, a new University of Massachusetts Amherst poll just found that 23 percent of American voters say Trump has handled AI well, compared with 57 percent who say he has handled it poorly. That’s a minus 34. Among independent voters, Trump is at minus 61.
According to a poll conducted by CBS News, 54 percent of voters would prefer a candidate who opposes new data centers in their area, compared to 15 percent who would prefer one who supports building them.
Even if Americans don’t believe A.I. will extinguish humanity, most voters believe AI will bring them financial ruin. The U-Mass poll found 64 percent of Americans think AI will reduce the number of jobs, while just 19 percent say it will create more.
There are some Trump White House staff who see the polling, appreciate the dire challenges facing Trump’s GOP in the mid-term elections, and also worry about the existential threat that A.I. may pose. These aides have reportedly advised Trump to deliver a prime-time speech to the nation, declare that certain aspects of A.I. could threaten humanity, and announce specific A.I. regulations the industry is requesting. Trump could also claim he is saving humanity from extinction. And in this case, Trump would probably be correct.
