If AI is allowed to drive discourse, can democracy survive?

Say political engagement could be increased in a certain society by the introduction of effective artificial intelligence tools — but those living there would have fewer in-person interactions. Should such tools be adopted?

A panel discussion featuring leading voices in computer science, tech, economics, law, and philosophy offered a range of viewpoints, some of them contrasting, on how such an AI would affect public discourse. The September 24 conversation was hosted by the Edmond & Lily Safra Center for Ethics as part of its “Ethics In Real Life” series.

Introducing the event was Eric Beerbohm, Alfred and Rebecca Lin Professor of Government and Ethics Center director. “AI doesn’t just give us new things to disagree about — although it certainly does that,” he began. “It also is changing how we encounter disagreement, whom we listen to, and how we make up our minds together.”

Building on this framing, moderator Lawrence Lessig, Roy L. Furman Professor of Law and Leadership at Harvard Law School, invited panelists to consider the pros and cons of making AI chatbots available to voters in a hypothetical city ahead of a mayoral election. The bots, designed with different personalities representing different types of constituents, would allow citizens to explore candidates’ political programs and potential impacts in an accessible and digestible way.

Dubious about technology’s ability to solve complex social problems, Shannon Vallor, Baillie Gifford Chair in the Ethics of Data and Artificial Intelligence at the Edinburgh Futures Institute, drew a parallel to social media. “Twenty years ago, we were all told that social media tools were going to be the magic democracy sauce that was going to be the cure for democracy’s ills,” she answered. “How well did that work out?”

James Mickens, Gordon McKay Professor of Computer Science at the Harvard John A. Paulson School of Engineering and Applied Sciences, brought up issues of access and representation. “In theory, these AI-based tools can be these great democratizers for information,” he said. “But there are all these practical implementation problems that end up being the real determinant of whether they’re helpful.”

David Deming, Danoff Dean of Harvard College and William Henry Bloomberg Professor of Economics, shared a more optimistic perspective on the chatbot idea. “People’s appetite for understanding the politics of their local area is pretty low,” said Deming, who studies the intersections of AI with higher education and work. “If you could use a technology to increase people’s engagement in local politics … my view is we should be open to ways to increase engagement in our local communities.”

After noting that she was not speaking on behalf of her employer, Manon Revel, senior research scientist at Google DeepMind, pushed back on Deming’s position. “I’m not sure I want to be deprived of the joy of going to my neighbor and going through the very messy and complicated process of engaging and getting their thoughts and maybe disagreeing,” she said.

Also on the panel was Brian Jeon, a senior at Harvard College who has served as a course instructor for the past two years. The economics concentrator reported seeing fewer and fewer students attending his office hours. “I did feel very sad that I wasn’t able to have that human-to-human connection,” Jeon said.

Even more concerning to Jeon was the possibility of undermining human interdependence, which is, in his estimation, the core of democratic society. Jeon worried that “the understanding of that dependence might get corroded by this tool because we suddenly feel that ‘I alone can understand the world via this chatbot.’”

“This technology is pushing us further down a road we’ve already gone too far down,” Vallor added, explaining the concept of “moral deskilling”: the idea that, over time, the less we force ourselves to do morally difficult things, like talk to a neighbor, the harder it becomes to do them and to tolerate any subsequent friction.

Revel shared Vallor’s concerns about deskilling. “A little bit of this technology is not bad in most cases,” Revel offered. “But the frictionlessness of the technology makes it really hard to have a little bit of it and not too much of it. How do we make effort fun?”

In a second scenario, Lessig asked panelists to consider the adoption of AI chatbots as stand-ins, not just supplements, to human interactions. These so-called “digital delegates” would act as representations of residents based on interviews and behavioral data. Digital delegates would specifically debate thousands of possible municipal housing plans and predict which one residents would be most likely to support.

“When I confront someone and realize how my political beliefs are affecting them … there is no guarantee that I will listen, there is no guarantee that I will understand,” Vallor answered. “But we cannot foreclose that possibility, because that is what democracy is for.”

Rather than view this technology as displacing human connection, Deming offered the opposite scenario. “As somebody who likes arguing with people and likes politics and likes socializing, maybe I’ll get to do more of it because I’ll have the AI organize my files … and a robot that mows my lawn and I can spend more time talking to people,” he said.

“We always talk about how AI is going to be the thing that is going to free us up finally and give us leisure and give us time with our families and let us do things for pleasure,” Vallor countered. “Rarely do we say, ‘What AI is going to be great for is fighting corruption among the most powerful and wealthy.’ Why are we not asking how AI can actually help hold the representatives themselves to account?”

Mickens closed out the event with an appeal to all in attendance. “I encourage people to be curious, to ask questions, and to think about what are the long-term, downstream effects of allowing any technology, but AI in particular, to infiltrate various parts of our lives.”