What this covers

Eric Topol, a cardiologist and author, joins Russ Roberts on EconTalk to examine how artificial intelligence might reshape medicine's human core. Topol argues that contemporary medicine suffers from insufficient time with patients—appointments averaging 7–12 minutes with physicians functioning as data-entry clerks—which produces over 12 million serious diagnostic errors annually and fuels doctor burnout. His central claim is that AI and deep learning, by automating routine pattern recognition and clerical tasks, can recover time for the activities machines cannot replicate: listening, empathy, and touch. Yet this recovery is not automatic. The Gift of Time must be deliberately redirected toward patient connection, or healthcare systems will simply accelerate throughput. Topol frames this as the essential move from Shallow Medicine toward what he calls "deep medicine"—care centered on human interaction as itself therapeutically vital.

The conversation ranges across AI's actual performance and its hype. Topol documents where technology has been oversold—IBM's Watson oncology project stalled because electronic health record data is "farcical" and error-laden—and where it shows genuine promise, such as deep learning reducing radiologist false-negatives from over 30% to single digits, or passive metrics detecting depression through voice and keystroke patterns. He argues that Retrospective vs Prospective Evidence matters enormously: most AI validation rests on historical datasets, not real-world clinical trials. The discussion also surfaces structural barriers: erroneous data perpetuated through records, unnecessary testing spawning costly incidentaloma findings, and patients' lack of Data Ownership. Topol contends that medical schools should select for Empathy Selection over Cognitive Selection, and that the placebo effect—substantially driven by doctor presence and touch—reveals why human connection cannot be digitized. Roberts raises the Third-Party Payment Problem, noting that absent Skin in the Game, no mechanism compels movement toward a more humane model.

Sharpest takeaway

Topol argues that current 'shallow medicine' fails patients through rushed, error-ridden, impersonal care, and that AI/machine learning—by automating diagnostic and clerical work—can liberate doctors' time to restore the human empathy and connection that is itself therapeutically essential.

  • AI excels at narrow pattern recognition (scans, depression detection) but is unproven prospectively and has been over-hyped
  • The gift of time freed by AI must be deliberately redirected to patient relationships, or it will be used to squeeze more throughput
  • Human connection and touch drive healing (placebo effect, trust), and cannot be digitized

The argument · threads4 threads · 34 claims
0.85

Third-party payment severs the cost signal between decision-makers and outcomes, preventing movement toward better models.

2 pointscentrality 5/5
  • Unnecessary tests have hidden costs—biopsies, unnecessary surgery, delayed radiation harm, and lost peace of mind—that are not free even when out-of-pocket cost is zero, and a core distortion is that the doctor who makes the testing decision bears none of the cost when it proves unnecessary.

    “people who make that decision--the doctor--who don't bear any cost if it turns out to be unnecessary, it still has these other costs which are harder to notice.”

  • The incentives to move toward a more humane medical model don't exist because the system lacks feedback loops and skin in the game—most care is paid for by a third party rather than the patient or provider—making it a strange, convoluted system from which it is hard to get to the better model.

    “the lack of skin in the game for, which should be the ultimate skin in the game, the fact that most money is being paid for by a third party. That just is such a strange, convoluted system. I think it's really hard to get there from here.”

0.56

U.S. medical innovation is real and globally significant, despite being funded through hidden costs borne by others.

1 pointcentrality 4/5
  • The U.S. is extremely innovative in medicine because it pays for these innovations with other people's money, and the rest of the world often enjoys the resulting innovations.

    “We are extremely innovative, because we pay for all these innovations out of, with other people's money--that the rest of the world often enjoys.”

0.43

International health comparisons are misleading because population differences and unique U.S. challenges confound the measures.

1 pointcentrality 3/5
  • International health comparisons are partly misleading because the U.S. population differs from others and faces unique problems—easy opioid access, more cars with cheap gasoline causing crash deaths, and later-age births subsidized by fertility treatments—and for a complicated medical situation one would likely rather be treated in La Jolla than in London's NHS, though this may not hold for a poor person.

    “Some of those outcomes are complicated by the fact that our population is not like their population... We have too easy access to opioids... We have a big country with lots of cars and inexpensive gasoline, so we kill a lot of people in our cars... We have women delivering babies at much later ages because we subsidize fertility treatments.”

0.29

Data property rights reform, not antitrust, addresses the incentive problems posed by large data-collecting firms.

1 pointcentrality 2/5
  • Traditional antitrust solutions are orthogonal to the problems posed by large data-collecting firms; the better approach is to change the legal and regulatory environment for property rights and ownership of personal data, forcing companies to share gains with the people who are not really their customers but their inputs.

    “the traditional antitrust solutions to these problems aren't really, don't seem relevant; they seem literally orthogonal to what the issues are. I think we ought to be looking at changing the regulatory, the legal environment for property rights and ownership of that data.”