Alex Imas, an economist at the University of Chicago and author of “The Power Ghost,” has developed an optimistic view of artificial intelligence (AI). He occupies a unique position in academia as both a leading scholar studying AI’s impact on the labor market and an active practitioner of the technology.

Unlike many of his peers, Imas takes doomsday predictions seriously, particularly the discussions around “ghost GDP” and spiral deflation raised by the independent research organization Citrini Research. This theory posits that if automation replaces most jobs and labor’s share of income declines significantly, wealthy capital holders will reach a saturation point in consumption, while unemployed workers will lack the means to consume.

Under this hypothesis, demand would collapse, leading the economy into a recession. Although Imas has written that the likelihood of negative economic growth is low, he emphasizes the need to take high unemployment rates and their potential to drag down the economy seriously.

“My first reaction was one of fear,” Imas told Fortune. “I needed to overcome that fear through rigorous reasoning, which is not self-comforting but rather a conclusion that integrates various factors based on historical patterns and human preferences.”

Wall Street also values Imas’s warnings. Morgan Stanley included him in a recent research report as a primary resource for investors to understand AI’s impact on employment, calling him a valuable third-party expert in the field.

Imas is not just a theorist. His research has been published in the American Economic Review, the Quarterly Journal of Economics, and the Proceedings of the National Academy of Sciences, and he collaborated with Nobel laureate Richard Thaler to update the classic work on behavioral economics, “The Winner’s Curse.” Perhaps most notably, he runs a widely read Substack column titled “The Power Ghost.” Upon learning he was featured in a Wall Street report, he joked, “That’s interesting… I hadn’t even noticed.”

The influence of “The Power Ghost” has exceeded his expectations. When starting the column, Imas aimed to write with the rigor of academic papers for a broader audience than journal editors, reaching economists, AI researchers, tech experts, and policymakers. He noted that the response has far surpassed his expectations, with even his mother-in-law’s friends providing feedback. Recently, he helped a neighbor install the AI agent Claude on her computer and witnessed her develop an application from scratch in just one afternoon. “These ideas need to be widely disseminated and reach more people,” he said.

The Signals from Starbucks

After months of writing and revising, Imas offers a thought-provoking perspective for those who adhere to doomsday theories, suggesting that an AI-driven economy may not necessarily lead to a grim outcome. He uses Starbucks as an example.

With a market capitalization of $112 billion and a focus on highly standardized products, Starbucks has long had plans to reduce labor through technology. However, despite years of layoffs and automation, the company has only managed to maintain thin profits. Recently, CEO Brian Niccol has completely changed strategy, re-emphasizing handwritten notes on cups, ceramic mugs, and comfortable seating—details that hold more value for customers than efficiency. Starbucks is now hiring more baristas and slowing down its automation efforts.

In Imas’s view, Starbucks’s transformation is highly instructive. He pointed out in a recent Substack article that as AI makes the production of goods cheaper and more abundant, “what becomes truly scarce?” In the age of AI, some things are destined to remain non-commodifiable. Niccol seems to understand this; the human presence, social connections, and the provenance of products will become increasingly scarce and thus more valuable.

While Starbucks is testing ChatGPT for drink recommendations, this is entirely different from its operational strategy. Starbucks stated to Fortune that its approach to AI is “pragmatic and grounded,” referencing previous public information. The company added, “If AI can help employees showcase their skills, deepen connections with customers, and optimize café operations effectively, it will be widely adopted; if not, it will be abandoned.”

From Farms to Relationship Industries

The theoretical support for this viewpoint is structural change theory, which examines how the economy evolves when technology significantly increases productivity in a sector. A classic case, praised by Fundstrat analyst Tom Lee, is that around 1900, 40% of the U.S. labor force was engaged in agriculture, a figure that has now dropped below 2%.

People did not stop eating; rather, as food became commoditized and inexpensive, they no longer spent most of their time on food production. The economy did not collapse but transformed, with labor gradually shifting from agriculture to manufacturing and then to services as workers’ incomes rose. Imas believes AI will bring about a similar dynamic: “The economic logic of scarcity will not disappear; it will merely shift.”

He cited a significant paper published in 2021 by Diego Comin, Daniel Lashkari, and Martti Mestieri in Econometrica, which pointed out that historically, over 75% of labor reallocation across industries has been driven by the “income effect” rather than the “price effect.” In other words, as people become wealthier, they do not simply buy more of the same inexpensive goods but seek goods and services with higher “income elasticity,” meaning demand for these goods and services grows faster than income.

Imas’s behavioral economics perspective is rooted in the “mimetic desire” theory proposed by French philosopher René Girard, which suggests that sometimes people desire something not merely for its practical value but because others want it and cannot easily have it. Research has shown that when subjects learn that a random subset of people is restricted from purchasing a certain item, the price they are willing to pay nearly doubles. When AI is involved in producing goods, the premium on those goods diminishes significantly, as people perceive AI-produced items as essentially infinitely replicable, weakening the scarcity that drives desire.

This means that as AI drives more areas of the economy toward commodification, consumption and employment will gradually shift toward what Imas terms “relationship industries.” This circles back to his analogy with Starbucks: people are willing to pay for things that have a strong human touch. The future middle class’s consumption patterns will resemble those of today’s affluent classes. Many financially free billionaires now spend considerable time on podcasts, live performances, and social platforms, enjoying consumption and creating interpersonal interactions.

“(The billionaires) could easily stay on an island, enjoying all the movies, playing all the games, and using all the tech products,” Imas said. “But most of the time, these billionaires are recording podcasts, interacting with others on X, and attending performances, essentially consuming relationship goods or striving to provide them, such as socializing and being among people.”

He believes that the demand for interpersonal connections has no natural limit, as it is fundamentally a comparative need that can never be fully satisfied.

Not Artists, but Nurses, Teachers, and Baristas

Imas emphasizes that he is not envisioning a romantic world filled with artists and performers. He stated, “Starbucks employees are just ordinary people; they are not performers or artists. People value their interactions with them, not from a highbrow, artistic, or entertainment perspective, but simply from a basic social need.”

In his theoretical framework, relationship industries include nurses, doctors, teachers, therapists, childcare workers, private chefs, and service personnel, among others. These sectors employ nearly 50 million people in the U.S. Many existing jobs will not disappear entirely but will transform. As AI takes over the repetitive tasks of teachers or doctors, the core of the job will shift to emotional support, care, and interpersonal connections, becoming the true source of economic value. Imas points out that one overlooked aspect is how these roles will evolve toward being more relationship-oriented as AI develops.

“Manufacturing workers and truck drivers may disappear because those jobs do not involve relational interaction,” Imas explained. “But many existing jobs with relational components will evolve into relationship-oriented roles.”

The Useless Sports Car

Imas’s theory has been tested in a large nonprofit healthcare organization. A senior data scientist, who wished to remain anonymous, told Fortune that despite management’s strong promotion of the enterprise version of ChatGPT over the past six months, employees found almost no other applications beyond writing emails and summarizing reports.

This data scientist stated that their actual work involved statistical analysis of cancer patient data for one of the largest medical databases in the U.S., but due to strict legal protections of privacy information, current AI tools have no access.

He agrees with Fortune’s analogy of AI as a “sports car,” but for most jobs, the reality resembles the traffic congestion of Manhattan. Years ago, shortly after ChatGPT was launched, he developed a cancer survival risk calculator using the tool in less than a month. However, due to the interpersonal nature of the tool, it has been stuck in legal review and paperwork processes, unable to be deployed.

He is not a Luddite. He acknowledges that AI can help him convert statistical code between different programming languages faster than building prototypes alone. However, he believes that compared to regression analysis, his most irreplaceable value lies in handling interpersonal collaboration, including communicating with international surgical oncologists from Yale, MD Anderson Cancer Center, and the University of Toronto, covering various cancers from thymic carcinoma to orbital sarcoma, bridging the gap between clinical intuition and strict statistical requirements.

“Experts have packed schedules, and being able to communicate with them for 15 minutes in a day is considered very lucky. Therefore, I must make everything precise and concise,” he added. Currently, no AI can replicate the communicative context required for such relationships. This complex, irreducible human judgment is key to maintaining the operation of complex institutions.

The Speed Issue

Imas has not completely dispelled his concerns; the optimistic scenario he describes depends on the speed of transformation. If the shift from a commodity economy to a relationship economy occurs gradually, historical experience suggests that the labor market can absorb and adapt. However, if the pace of AI automation far exceeds the speed at which workers and institutions can retrain and reallocate, the demand collapse he warned about may still occur.

“The speed of change is crucial,” he said. “It determines whether humanity ultimately moves toward a hopeful or more worrying future.”

Imas warns that those who still treat AI as an overhyped phenomenon are deceiving themselves, possibly because they are still using chatbots from years ago rather than cutting-edge models. He pointed out that current AI remains “uneven.” While this term is widely used to describe the probabilistic nature of AI and its tendency to hallucinate, Imas emphasizes that “one day, even the lowest points will be at an extremely high level… even the worst performance will be quite impressive.”

Morgan Stanley warned in its March report that as the capabilities of large language models improve faster than expected, the impacts of AI could become “increasingly severe,” with large-scale layoffs possible across industries. On one side are serious predictions, while on the other, a cancer statistician quietly waits for corporate enthusiasm for ChatGPT to wane. This stark contrast reflects the uncertainty that Imas, after much struggle, still cannot completely resolve despite his optimistic judgment.

Imas remains concerned for those who adopt an ostrich mentality toward AI, stating that the current priority is to provide one-on-one coaching to help people master cutting-edge technologies. He believes that his theory of relationship industries is both reasonable and positive, but he also admits, “It took me a long time to come to this realization.”