Who wants to be a millionaire? Probably the better question is, who doesn’t? Unfortunately, people often suffer from financial misbeliefs that could end up hurting their wealth accumulation and retirement goals, according to MIT professor and financial economist Eric So.
As people lean on AI tools to help answer their financial questions—from investment choices to retirement planning—So’s latest research explores whether using AI as a financial advisor can help mitigate some of their deeply held but potentially flawed financial thinking.
Testing AI Financial Advisors
Certainly, people seem more willing than previously to give the technology a chance. Artificial Intelligence at the Consumer Inflection Point, a 2026 TD AI Insights report, found that more than half of survey respondents said they use AI to help manage their finances, compared to only 10% last year.
The good news is that AI does indeed hold promise for influencing people’s financial perceptions in useful ways, So explained to attendees at the 2026 MIT IDE Annual Conference. He presented research, which he has undertaken with MIT colleagues Andrew Lo and Jillian Ross, that involved creating an AI financial advisor chatbot designed to intervene to correct users’ mistaken thinking about financial matters. In the paper, Breaking Bad Financial Habits: How LLM Conversations Correct Financial Misconceptions, the researchers observed a significant and lasting shift away from mistaken beliefs among users who leveraged the tool.
But So, who leads an IDE research group about how AI and its pervasive use influence our decision making, cautioned that this effect depends on the type of chatbot you’re using. The LLM that was specifically designed to engage people with their misconceptions showed a greater shift, vs. an off-the-shelf LLM, which has much less impact.
“What our research shows is that by far and away, the most effective intervention is an LLM specifically designed and primed to push against financial misbeliefs,” he said. An off-the-shelf LLM has essentially only the same effect as having a text box appear for people to reflect on why they believe their misconception or simply asking them to pause and reevaluate.
One reason for off-the-shelf LLMs’ weaker influence in changing financial misconceptions is that they often have a sycophantic nature, he noted, which reinforces user beliefs rather than challenges them. On the other hand, purposefully designed AI systems can durably correct financial misconceptions, suggesting they could become scalable tools for financial education.
Human vs. AI Financial Advisors
So recently also conducted research with Abigail Sussman and Fiona Yang of the Booth School of Business at the University of Chicago that explored the trajectory of AI adoption for professional services where people often turn to human advisors–including the financial domain. That paper, AI Advisors and the Competence-Judgment Tradeoff in Information Disclosure, revealed that participants consistently rated human advisors as more competent than AI advisors when evaluating who could best solve their problem.
While participants perceive humans as having a competence advantage relative to AI, social embarrassment can complicate things.
“Revealing an embarrassing mistake to a knowledgeable person means risking their judgment of your character, and that risk can suppress the very disclosure [of information] that would make their advice worthwhile,” the paper notes.
Participants are willing to share information with a financial advisor when dealing with a medical emergency, for instance, but that willingness decreases if they are reaching out for advice because of frivolous spending.
“AI adoption in professional services,” the paper concludes, “will depend not only on closing the capability gap but on how domains differ in the social costs of disclosure.”
Putting Trust in AI
What these two papers tell us is surprisingly human. People are willing to trust AI when it shows up as an agnostic expert, without judgment or an agenda to please. When an AI advisor is deliberately designed to challenge faulty thinking, people let it change their minds in ways a human advisor often can’t. And when a financial situation carries social embarrassment, people are more willing to open up to AI precisely because it won’t judge them for it. In both cases, the trust isn’t built on capability alone. It’s built on the absence of the social friction that makes human advisors complicated.
What AI Is Doing to Us
So’s work exploring AI and financial decision-making is part of a larger body of work exploring the risk AI presents to human skills, decision making and building leadership. While AI offers valuable capabilities, the more we use it, the greater our risk of reducing our own thinking, practicing, and refinement of our own skills, which can lead us to lose them.
So will address what he has been studying about the way AI is changing how people think and make decisions in his upcoming book, The Collision: What AI Does to Us, due out in October.