The Other AI Alignment Problem

Artificial intelligence promises extraordinary gains. It can raise productivity, transform medicine, reorganize supply chains, improve manufacturing and create products and services we cannot yet imagine. I am an optimist about those possibilities.

But optimism is not the same as complacency.

The question that interests me is not simply how much wealth AI will create, but who will prosper from it—and who will gain power as a result.

AI is a scalable technology. Building the most powerful systems requires vast amounts of computing power, data, talent and capital. Deploying them rewards firms with enormous customer networks and distribution platforms. These forces can make successful companies extraordinarily productive. They can also make them extraordinarily large. The Magnificent Seven accounted for roughly 8 percent of the S&P 500 in 2012 and about one-third by 2025.

But economic concentration is only the beginning.

My work focuses on a broader cascade. Economic power can become political power. Political power can influence the rules governing markets. And both can reshape the beliefs and social norms through which societies understand inequality, opportunity and legitimate authority. My AI presentations therefore treat economic concentration, political influence and social stratification not as separate problems, but as connected ones.

This leads to what I call the other AI alignment problem.

Most discussions of AI alignment ask how we can ensure that increasingly capable machines behave in ways consistent with human values. That is an important question. But there is an earlier one that we must tackle first, before it is too late.

How do we align the companies building those machines with the interests of the societies they increasingly shape?

That question is partly about regulation. But it is also about ethos: the beliefs that determine which policies seem reasonable, which risks seem acceptable and whose interests are understood as serving the public good.

Today, the Silicon Valley Hacker Ethos is dominant. It says that free speech trumps competing rights; that regulation kills innovation; that better AI requires Big Tech to have broad access to data; and, increasingly, that powerful American AI companies are needed to prevail in the strategic competition with China.

The US-China competition makes this especially potent. The argument is straightforward: America must win the AI race; regulation could slow American innovation; therefore constraints on the largest American AI companies risk strengthening China.

There is real geopolitical competition behind this argument. But it also creates a convenient alignment between the commercial interests of the largest technology companies and the language of national security. The danger is that “America must win” quietly becomes “let the largest American technology companies decide how AI should be governed.”

That is not technological necessity. It is a fabricated ethos.

And once an ethos takes hold, it can shape which regulatory choices societies even regard as possible. That is why the alignment problem cannot begin and end with engineers designing guardrails for machines. We also need institutions capable of aligning the immense private power created by AI with broader social objectives.

The objective is not to slow AI until its risks disappear. Nor is the choice between technological leadership and responsible governance. The harder task is to capture the enormous benefits of scale without allowing scale to determine who writes the rules.

This is the argument I have been developing in talks around the world, often in China. It grows out of a question that has run through much of my research on trade, technology and institutions: When does economic change generate broad-based prosperity, and when does it instead concentrate economic, political and social power?

AI makes that question more urgent. The other alignment problem is whether we can harness its extraordinary capabilities without allowing those who control the technology to define the purposes society asks it to serve.