More than 59 million people in the U.S. suffer from mental illness, or approximately one in every five adults. Despite this staggering figure, barely half receive treatment, often due to a lack of access to care.
According to Columbia Business School professor Boaz Abramson, this fact quietly imposes costs on the U.S. economy comparable to a recession each year. His research alongside the University of Wisconsin’s Professor Job Boerma and Yale University’s Professor Aleh Tsyvinski estimates that mental illness reduces aggregate consumption by roughly 1.2 to 1.3 percent annually, translating into $260 billion in lost economic activity each year.
In this way, mental health is not just a medical or social issue, but an economic one. Mental illness shapes how much people work, the kinds of jobs they take, how much risk they are willing to bear, and how much wealth they accumulate over time. Plus, unlike a typical downturn, its effects do not naturally reverse.
Now, as AI is rapidly introduced across healthcare, can the technology address one of the economy’s most persistent challenges by expanding access to mental healthcare and improving outcomes? Or will it simply shift costs and risks in ways that make the problem worse?
Why Mental Health is a Persistent Economic Challenge
Mental illness is far more prevalent and economically consequential than many traditional economic models assume. Included in the 20 percent of the U.S. population that experiences some form of mental illness, is about 5 percent who live with severe mental illness that substantially interferes with daily functioning.
These conditions influence economic behavior in systematic ways. People suffering from mental illness tend to work fewer hours, sort into less demanding roles, and avoid risk. They invest less in risky assets, accumulate less wealth over time, and consume less as a result. What begins as an individual health challenge becomes a macroeconomic one.
Abramson’s research shows why the costs are so persistent. Mental illness often goes untreated, not because care is ineffective, but because access remains limited. As a result, its economic effects compound rather than dissipate, creating a steady drag on growth rather than a temporary shock.
“Around 35% of the US population lives in areas without any mental health providers and as a result doesn't have access to in-person mental health services,” Abramson said.
Abramson believes the strongest economic case for AI lies not in replacing clinicians, but in helping them serve more patients. Used as a complement, AI can streamline administrative work such as billing and record summarization, freeing clinicians to spend more time with patients. It can also support treatment through digital therapeutics that reinforce cognitive behavioral therapy between visits, or through monitoring tools that give clinicians better insight into patients’ conditions outside the clinic.
"There's consensus among psychologists and psychiatrists that tools that work in cooperation with the clinician are well suited to improve the availability of mental health services,” Abramson said.
From an economic standpoint, this distinction matters. The binding constraint in mental healthcare is not demand, but supply. Tools that make clinicians more productive effectively expand access without sacrificing quality. Over time, improved access can translate into better labor outcomes, higher productivity, and a reduction in the long-run economic costs of mental illness.
There is broad consensus among clinicians, Abramson notes, that this kind of AI-enabled augmentation holds real promise.
When AI Substitutes for Care, the Risks Grow
The picture changes when AI is positioned as a substitute rather than a complement. Virtual mental health tools, including AI chatbots, are often justified as better than nothing in underserved areas. But Abramson urges caution.
He points to recent research showing that AI systems can respond inappropriately to signs of suicidal ideation, offering factual information where clinical intervention is needed. Bias and stigma embedded in training data can further compromise care.
"There are a lot of dangers associated with AI chatbots providing mental healthcare in place of clinicians, such as failing to address suicidal ideation and exhibiting biases and stigma,” Abramson said.
The economic risks are significant. AI-only solutions may lower short-term costs, but poor outcomes can generate higher downstream expenses through disability, unemployment, and crisis-driven healthcare use. What appears efficient in the moment may ultimately prove costly. The challenge, Abramson emphasizes, is that AI capabilities are advancing faster than the safeguards needed to deploy them responsibly at scale.
Who Saves and Who Pays?
Powerful incentives are pushing organizations toward AI adoption. Employers and insurers are under pressure to control healthcare spending, while health systems are trying to stretch limited clinical capacity. AI offers an appealing way to do both.
But who captures the savings when AI lowers costs, and who bears the risk if outcomes worsen?
In many cases, organizations benefit financially, while patients and families absorb the consequences of inadequate care. At the macro level, the broader economy pays as well, through reduced productivity and higher public costs.
AI, in this sense, is neither inherently a solution nor a threat. Its economic impact depends on whether it improves effective care or merely shifts responsibility. Used to complement clinicians and expand access, AI could help address one of the economy’s most persistent drags. Used as a low-cost substitute without sufficient safeguards, it risks deepening the problem it aims to solve.
