Opinion | Can A.I. and Democracy Fix Each Other?

Democracy isn’t working very well these days, and artificial intelligence is scaring the daylights out of people. Some creative people are looking at those two problems and envisioning a solution: Democracy fixes A.I., and A.I. fixes democracy.

This week I spoke with several people who are hard at work on various aspects of this endeavor. Among them: Audrey Tang, the minister of digital affairs for Taiwan; Divya Siddarth, a founder, with Saffron Huang, of the Collective Intelligence Project; Colin Megill, president of the board of the Computational Democracy Project; and Beth Simone Noveck, who directs the Governance Lab at Northeastern University.

They’re exemplars of a larger movement that has its heart and head in the right place. Attitudes about A.I. are polarized, with some focusing on its promise to amplify human potential and others dwelling on what could go wrong (and what has already gone wrong). We need to find a way out of the impasse, and leaving it to the tech bros isn’t the answer. Democracy — giving everyone a voice on policy — is clearly the way to go.

Democracy can be taken hostage by partisans, though. That’s where artificial intelligence has a role to play. It can make democracy work better by surfacing ideas from everyone, not just the loudest. It can find surprising points of agreement among seeming antagonists and summarize and digest public opinion in a way that’s useful to government officials. Assisting democracy is a more socially valuable function for large language models than, say, writing commercials for Spam in iambic pentameter.

The goal, according to the people I spoke to, is to make A.I. part of the solution, not just part of the problem. Noveck, whose lab uses information technology to assist democracy, told me she opposes Elon Musk’s and other tech leaders’ proposed six-month moratorium on the development of A.I. systems that are more powerful than the GPT-4 chatbot. “We should be doubling down on efforts to use A.I. for social good and strengthening democracy,” she said.

Noveck’s lab has used artificial intelligence tools — including one from Iceland called Your Priorities — to help the people of Oakland, Calif., come to agreement on building tiny houses for the homeless, among other projects. She wrote by email, “Ultimately, we can use A.I. to regulate, make laws, govern and to develop institutions that are flexible enough to address the challenges of A.I.”

Likewise, the Collective Intelligence Project argues that “collective flourishing” requires progress, safety and participation of the public in artificial intelligence.

Siddarth, who worked for Microsoft’s chief technology officer until January, told me that while elites’ views on A.I. are highly polarized — from “let ’er rip” to “bomb rogue data centers” — the general public is still uncertain. Since most people aren’t dug in, there’s still an opportunity for consensus building, she said.

I told Siddarth that her project reminded me of the efforts of Daniel Yankelovich, a pollster who died in 2017. In 2010 he and his co-editor, Will Friedman, published “Toward Wiser Public Judgment,” which called for building consensus on key public issues, starting with focus groups at the grass roots.

One difference, Siddarth told me, is that her group wants to move much faster. “We want to go to the global community and start big,” she said. “I think we can be ambitious. We have to take that plunge at some point.”

At President Biden’s Summit for Democracy last week, Tang announced plans to begin a series of citizen assemblies that would work to make sure that A.I. is aligned with human needs. One danger of A.I. is that it gives you exactly what you ask for, which may not be what you really want. We risk becoming like King Midas, who got his wish that everything he touched would turn into gold, then turned his own daughter into gold when he embraced her.

Tang, a self-described hacker who addressed the summit by video (her remarks start at 3:56:00), is worth listening to on A.I. and democracy because Taiwan is a world leader in incorporating public opinion into decision making through a consultative process called vTaiwan. One of its signal successes was forming a consensus around how UberX, Uber’s standard private ride service, was allowed in. (UberX, recall, uses a form of artificial intelligence, albeit not a general purpose one.)

Taiwan is a young democracy. Its first direct election of a president was in 1996. Still, Tang told me that Taiwan has several institutions that are widely regarded as trustworthy and impartial, including National Taiwan University; the national academy, called Academia Sinica; and PTT, a Reddit-like forum operated by students of National Taiwan University. Their participation makes consensus building around complex issues easier than in the United States, she said.

“I’m very interested in depolarization of these conversations” around A.I., Tang told me. She said that “it’s very difficult to hit the rewind button” on A.I., so the focus should be on mitigating the risks of A.I. systems that have already been released.

Tang and the Collective Intelligence Project use software called Polis to solicit and digest public opinion. The widely used software, which is in the public domain, was built by the Computational Democracy Project, a nonprofit based in Seattle. “We are providing a map of the opinion space,” Megill told me. “You can comment, and you comment on other people’s comments. That’s the matrix.” He said it’s somewhere between Twitter, which is a completely unstructured way of gathering opinion, and a conventional survey. An early generation of artificial intelligence is used to digest inputs, including by discovering clusters of people who are more or less like-minded.

Polis has been researching how to take advantage of large language models since 2020, Megill said. Until now, there were concerns that the models weren’t powerful enough, but the new ones definitely clear that hurdle, he said. The other concern is “to make sure we understand how it will behave in delicate use cases,” Megill said.

“We’re in zoo construction, but we haven’t ordered the lions,” he said.

That’s a useful way of thinking about A.I. and democracy. One way to avoid getting eaten by lions is to have no lions. A much better way is to have lions but make sure they can’t get out of their cages. I’m not sure if a moratorium on development is a good idea or not. What I am sure of is that answering the question is a job for everyone, and A.I. itself can help us sort out a solution.

Elsewhere: Old and Can’t Get a Mortgage

The likelihood of being rejected for a mortgage loan rises with age, but that’s not proof of age discrimination, according to a study last year by Natee Amornsiripanitch, a senior financial economist at the Federal Reserve Bank of Philadelphia. Regulations allow lenders to consider a borrower’s age under some circumstances, according to a brief about his work for the Center for Retirement Research at Boston College. The brief, which was written by center staff members and approved by Amornsiripanitch, also says that “mortality risk has real economic implications for lenders for which they might require additional collateral.” A second brief from the center looked at whether older people pay more for mortgages. It found that they did, but only up to 0.08 percentage point on a typical purchase loan of about 4 percent. The gap might be caused by older applicants doing less shopping around, it said.

Quote of the Day

I’ the commonwealth I would by contraries
Execute all things; for no kind of traffic
Would I admit; no name of magistrate;
Letters should not be known; riches, poverty,
And use of service, none; contract, succession,
Bourn, bound of land, tilth, vineyard, none;
No use of metal, corn, or wine, or oil;
No occupation; all men idle, all;
And women too, but innocent and pure;
No sovereignty; —

— William Shakespeare, “The Tempest” (1610)