Can AI Systems Be Designed to Support Human Autonomy and Societal Well-Being?

In 2021, an internal whistleblower alleged that Meta knew its Facebook algorithms were amplifying harmful content, including the promotion of posts that worsened body image, anxiety and exposure to self-harm material among teens. But those same systems also drove engagement. Meta struggled to act.

The allegations triggered a wave of media coverage and prompted regulatory and legal scrutiny that continues to this day.

Today, AI far more advanced than simple algorithmic content curation is expanding human potential at a breathtaking pace. But without sufficient guardrails in place for safely experiencing it, the technology could have negative impacts on society at a similar or greater scale than those often attributed to social media algorithms.

Darden experts are exploring the balancing act between AI’s promise and perils through research on how AI systems promote human flourishing — and squash it. “These are groundbreaking changes in our society. It is critical not just to be at the technological edge, but to be at that human-AI intersection edge as well,” says Professor Roshni Raveendhran. She studies how novel technologies affect flourishing in individuals and workplaces and teaches the Darden elective “Minds and Machines: Flourishing in the Age of AI,” which helps students explore what it means to harness the power of AI for the well-being of themselves, the workplace and society.

Human Psychology Defines AI Opportunities

Professor Roshni Raveendhran She studies how novel technologies affect flourishing in individuals and workplaces.

AI users flourish when the technology is integrated thoughtfully, Raveendhran says. They have an enhanced sense of individual autonomy, feel they are perceived as competent by others and improve their social relations — all basic psychological needs that humans share.

A consistent finding in her research is that people believe engaging with AI is less socially risky than interacting with humans. Because AI is a machine, the thinking goes, it doesn’t carry the judgments or other social motivations that people have.

In one study, people who talked to AI about minor problems — a boss was rude, a driver cut them off — reported feeling less stress and anxiety afterward because they believed AI was less likely to form negative judgments of them than a human would.

Another study looked at the effect of technologies that track worker behavior. A laptop software might provide feedback on time spent on a task or the number of breaks taken in a day. Users of the technology reported feeling much as they would using a smart device that tracks fitness goals.

“With the technology, we experience that as informational and not as evaluative. We feel more autonomy. It feels like information you would like to know about your behaviors,” Raveendhran says. “But when we hear the exact same things from humans, even if they’re not objectively judging us or causing harm in any way, we experience that context as evaluative — just because humans have the capacity to form judgments.”

The AI Perception Factor

Darden professor Luca Cian is an expert on consumer psychology.

A key driver of AI’s role in supporting human flourishing is people’s beliefs about the technology. While AI is perceived as a supportive tool for many things, it also can be eyed with suspicion when it steps in uninvited. For example, a news story written by AI or a chatbot that insists on handling your question when you’d rather speak with a human agent.

“AI doesn’t fail because the code is wrong, though that is sometimes part of it. It fails because it isn’t designed for how the human mind actually receives it,” says Professor Luca Cian, who studies AI strategy and psychology.

“In companies and government, very few people understand the psychological aspects of AI. You can write the most beautiful and perfect AI recommender or agent, but if people don’t trust it and don’t use it, then it’s useless.”

People are more likely to trust AI that is developed or adopted in ways that support human thriving at every step, Raveendhran says. In an ideal scenario, stakeholder input is reflected at all levels — from designers, technologists organizations to policymakers, employees and other end users. Psychological impacts are noted with intent to ensure the rollout is driven by purpose.

“All stakeholders have to figure out the bounds of these technologies,” Raveendhran says. “How can we embed and integrate these technologies in a way where people feel trusted and valued?”

In a study that measured news readers’ trust of AI-written content, Cian and his co-authors found that people rated the same news headlines as less accurate when labeled as AI-written than when labeled as human-written. The effect held regardless of the headline’s actual veracity. AI authorship dragged down perceived accuracy even for objectively true news.

In another study, Cian and his co-authors found that consumers prefer AI-based recommendations when they’re choosing for utilitarian reasons (e.g., a vacuum cleaner) but prefer human recommendations for hedonic choices (e.g., a fragrance) because they assume AI is good with facts but bad at experience and emotion. Companies like Netflix could respond to the finding effectively, Cian suggests, by emphasizing AI picks for informational content like documentaries and picks from “similar users” for emotional content like horror movies.

“The single most reliable trust restoration move I found in my research is that it’s very important to keep a human in the loop most of the time,” Cian says. “People think a human would compensate where the AI is less competent, and the other way around.”

System Support

AI is disrupting entrenched systems, but it also brings enormous potential to fuel flourishing systems, Cian says. The opportunity calls for leaders with a moral compass.

In one study, participants were treated unfairly by either a human or AI in a money-allocation game. Participants then had a chance to call out a different, unrelated human who had also behaved unfairly. People who had been wronged by the AI were significantly less likely to push back on the second person — what Cian and his co-authors call “AI-induced indifference.” Because people blame AI less for unfairness, exposure to unfair AI seems to dull our reaction to unfairness in general.

“It’s fundamental for us to study the moral changes AI is bringing with it. If people are less sensitive to AI — less likely to blame it for unfair actions — we risk a whole class of people using AI to do unfair things at arm’s length,” Cian says.

At the same time, the use of AI can also enhance how others perceive us — again, because it’s thought to have lower social motivations. In a study, Raveendhran looked at perceptions of women working in male-dominated industries. When the screening process involved AI, people evaluated women employees more favorably than if AI were not used, believing the technology is less likely to discriminate.

“The important thing to understand here is that these are people’s perceptions. These are not objective realities,” she notes. “People believe that AI, compared to a human, is less likely to make judgments based on bias and discrimination.”

Society’s understanding and expectations of AI will keep evolving, Raveendhran adds. Lessons learned from previous groundbreaking tech advances — the internet, social media — will help center the human experience to some extent. But AI’s unprecedented reach into every facet of daily life calls for stronger guardrails and new approaches, such as greater collaboration between academia and practitioners.

“Humans are the fundamental piece of this puzzle,” Raveendhran says. “It’s a human question, and that’s what we need to be thinking about.”