In Defense of “Crackpot Schemes” for AI Governance

AI is unpopular. And nearly a billion people use ChatGPT.

AI is destroying jobs. And fields predicted to have been eliminated by AI, like radiology, continue to grow and leverage the technology to improve their work.

AI is wrecking the environment. And state officials are learning from hyperscalers how to run the grid more efficiently.

A black or white conception of AI promises to either deny its benefits or exacerbate its shortcomings. Yet, there’s no easy way to ensure a proper balancing of AI’s promises and perils. While every major AI CEO has pledged to "maximize the benefits and mitigate the risks,” they have left out the details of how to do so.

It’s unlikely they or anyone will ever have such a definite plan. But, if we want to avoid repeating what we did with nuclear power — locking a technology away and letting its best uses go unrealized — we need bold ideas from outside the rooms where AI policy currently gets written. We need “crackpot schemes.”

Consider Social Security. Most Americans, asked who we should thank for it, would name FDR. The better answer is a doctor in Long Beach. In 1933, he wrote a letter to the editor proposing a $200 monthly pension for every citizen over 60, funded by a sales tax, with one catch: recipients had to spend the money within thirty days. Two goals, one scheme. Support the elderly. Sustain a Depression-era economy.

The idea caught fire. Clubs formed across the country, 2.2 million members strong. Ten million Americans signed petitions. By 1935, a majority of the country backed it. The plan itself never became law. Critics called it a “crackpot scheme.” But the Social Security system we have today is unimaginable without that op-ed from a doctor with minimal policy chops or political clout but plenty of experience seeing people struggle to achieve financial security in their golden years.

The best policies often come from unexpected places.

Now consider the policy challenges posed by AI. The Bureau of Labor Statistics projects 3.1 percent job growth by 2035. Forecasters at Metaculus project a 1.4 percent loss. Expert forecasters also expect unemployment for recent college graduates to hit 13.5 percent in the near future, with underemployment climbing to nearly 60 percent. Nobody knows for certain how this economic transition will play out. We are entering a volatile decade, and the people most exposed to it are the people with the least say in how it gets governed.

AI must be developed for us and by us if it is going to earn our trust. People support what they help build. As of now, there are few clear mechanisms for people to share their perspective on what they need to feel like AI is a net positive for their wellbeing and the wellbeing of their community. That needs to change.

Here are a few of the “crackpot schemes” I’ve heard about when talking to everyday Americans eager to know that AI progress does not have to come at the expense of Average Joes and Janes.

First, a mandatory national service program with civil and military tracks for Americans between 17 and 19. It would signal that the country values young people's judgment at exactly the moment AI is making judgment more valuable than ever. It would build the interpersonal experience no model can replicate. It would help revive the civic glue that holds this country together.

Second, recurring citizen assemblies at the local, state, and federal levels that provide a representative set of Americans with the chance to share their viewpoints in a calm, deliberative fashion. Polls focused on likely voters may hold sway with political campaigns but do little to help guide actual policymaking. Gathering people in a room to discuss tangible, specific policies and related trade-offs may bolster the ability of politicians to plan for and respond to many different possible futures.

You may regard these as "crackpot schemes" with no odds of passing. Good. The doctor's pension plan was wrong on the numbers, wrong on the mechanism, and wrong on the funding. It was also the most important policy proposal of the decade, because it forced a country that had decided old age was a private problem to admit it was a public one. Your AI proposal does not have to be right. It has to be loud enough to make the wrong answer indefensible. Now share yours.

Kevin Frazier is the Director of the AI Innovation and Law Program at the University of Texas School of Law.