It strikes me as bizarre that today in the United States, a new pill to treat heartburn would have to go through months or even years of government approvals and testing before being released to the public. But the most powerful technology ever developed — capable of breaking into banks, shutting down power grids or helping build bioweapons — can be released without even a preliminary government review.
We are having a collective failure of imagination about artificial intelligence. While many closest to the technology have issued grave warnings and urged oversight, it doesn’t seem to register as a major public concern. Congress has, of course, so far abdicated its role as lawmaker, and the Trump administration is delighted to let the industry regulate itself.
President Trump is right that the term “artificial intelligence” doesn’t capture the phenomenon. It is super intelligence: analytic capacity greater than that of any human and perhaps even collections of the smartest humans. We reach for analogies — the industrial revolution, electricity, the internet. But electricity does not get exponentially smarter every six months. The internet cannot solve mathematical problems, diagnose diseases or design new weapons.
Bad people can use these superpowers for bad purposes, but that’s not the scariest scenario. We have seen models hack into databases and computer systems while pursuing unprompted goals. Suddenly, the old thought experiment about an AI that’s told to maximize paper-clip production — which proceeds to maniacally commandeer all the world’s energy and resources to make more paper clips — seems a little less fanciful. The danger is not that the machine becomes evil. It is that it becomes massively and unpredictably destructive while pursuing its legitimate goal.
Self-regulation isn’t possible. If one laboratory spends six months testing a powerful model while its competitors release theirs, prudence carries a commercial penalty. The companies themselves understand this. Read the calls issued over the last two years by researchers and top executives at almost all the leading AI companies, and you’ll find warnings of a “race to the bottom,” in which everyone might prefer greater safety but no one can afford to slow down first. That is a classic collective-action problem — and precisely what regulation is designed to solve. For a start, many of the voluntary safeguards the industry has proposed could simply be turned into mandatory ones.
But what about China? The argument that U.S. regulations will simply “let China win” starts with a false premise — China is not racing ahead without rules. Beijing already regulates powerful AI systems, requiring security assessments and restricting how models are developed and deployed. And history doesn’t suggest that technological competition must mean abandoning safety. During the Cold War, the United States was locked in an existential nuclear race with the Soviet Union, yet we required civilian reactors to be licensed and meet federal safety standards. In fact, the Soviets were the ones with a weak safety culture and inadequate oversight, which produced the Chernobyl disaster. After it, Washington and Moscow actually expanded cooperation on nuclear safety, recognizing that some risks transcend national rivalry. The fiercer the competition and the more consequential the technology, the more important it becomes to establish some rules for the race.
Reasonable regulations can make you stronger. After the 2008 financial crisis, Washington imposed major new rules on American banks — higher capital requirements, stress tests and restrictions on risk-taking. The banks screamed bloody murder. The chairman of Wells Fargo called the stress tests “asinine.” Washington, we were told, was shackling one of America’s great industries.
But guess what? In the years after, American banking boomed. According to Quartz, from roughly 2012 to 2022, the six largest U.S. banks made $643 billion in profits, more than they had in the decade before. And they didn’t just get richer; they came to dominate the world like never before. The top six American banks today have a combined market capitalization larger than the top 20 European banks. In the early 2000s, the market cap of JPMorgan Chase was about 40 percent smaller than that of HSBC, Europe’s largest bank. Today, JPMorgan Chase, now the U.S.’s largest bank, has a market cap bigger than Europe’s top four banks combined. Regulation was not the only reason. America recovered faster, Europe stumbled and U.S. banks benefited from tax cuts. But forcing banks to raise capital, shed risk and test their capacity to survive severe shocks did not destroy their competitiveness; it enhanced it.
AI will be the commanding technology of the 21st century. The U.S. should want to dominate it. But dominance will not belong simply to the country that builds the most powerful models fastest. It will belong to the country whose AI the world is willing to trust — with its data, infrastructure, businesses and perhaps even weapons. China will build AI that is very powerful. The U.S. should build AI that is powerful but also safe, with the human being at the center and in control. American technology combined with American values will be our best competitive advantage.
The post This is the most powerful technology ever. Don’t fail to imagine the risk. appeared first on Washington Post.
