Opinion: Will AI soon diagnose politicians’ mental health conditions from afar?

For psychiatrists, Donald Trump’s third run for the presidency means that we’ll once again face the challenge of diagnosis: Should we, or shouldn’t we? For 50 years, the American Psychiatric Association’s code of ethics has forbidden any member from diagnosing public figures unless the doctor has interviewed them and obtained their permission, a constraint known as the Goldwater Rule. During the 2016 presidential race, some psychiatrists skirted the rule and, through indirection, labeled Trump paranoid, sociopathic, or pathologically narcissistic. In justification, they referred to a competing obligation to warn the public about impending disaster.

Ever since, the field has debated the merits of diagnosis from a distance. I won’t rehash the arguments. Instead, I want to make a strange, speculative prediction: that the need for resolution will soon be behind us. We won’t have to rely on doctors to diagnose public figures. Bots will do that job.


One objection to diagnosis based on public behavior is that it lacks the nuance and specificity of face-to-face interviews. But it’s not as if the record contains no relevant information. Calling Trump narcissistic, one psychologist pointed to Trump’s tendency to assert unique superiority — “Only I can fix it” — and to claim expertise in a host of different subject areas. (The constraints on psychologists resemble those for psychiatrists.) The media, this and other experts said, contained ample evidence on which to base a diagnosis.

Reviewing public data and media coverage is something that artificial intelligence does well. In fact, various groups have already claimed to have the ability to use AI to diagnose conditions like depression, bipolar disorder, and dementia by using speech pattern recognition (and sometimes analysis of transcribed text). In time, couldn’t AI bots learn to mimic whatever it is that psychiatrists and psychologists attempt when they comment on candidates’ fitness for office? Given the limits of prediction in psychiatry, it’s not hard to imagine that bots will outperform us.

This thought came to me when I read a news report about a new precaution that venture capitalists make before investing a startup. They hire consultants to assemble a psychological profile of the CEO, based in part on a face-to-face interview. They’re looking for a personality triad, comprising egotism, manipulativeness, and sociopathy, that has lately been called Machiavellian narcissism. The concern is that Machiavellian narcissists may end up committing fraud and, when caught, drive companies into bankruptcy.


The stimulus for this interest is the collapse of companies like Theranos, FTX, Frank, Outcome Health, Ozy Media, Slync, Headspin, and others whose executives have been accused or convicted of falsifying data and misleading investors. For AI, that sort of list would provide an inroad. And when there’s money at stake, research will follow.

I’m imagining a scenario in which programmers identify companies whose worth has plummeted and whose CEOs are thought to have exhibited Machiavellian narcissism. Using that data as a starting point, AI would review speeches, journalistic interviews, video presentations, news reports, and opinion pieces and create algorithms that detect Machiavellian narcissism at a level that, when found in corporate leaders, confers risk. Perhaps investors would subscribe to such a service.

It’s a short leap from finance to politics. Once a relevant algorithm emerges, wouldn’t it be applied to candidates for public office? (And what do we imagine? Would the algorithm pick out Trump?) If investors find red-flag warnings valuable, perhaps certain voters will as well.

AI’s output might not be specific labels like narcissistic personality disorder or sociopathy, but regarding Trump there has never been agreement on one diagnosis. “The Dangerous Case of Donald Trump,” the widely read compilation of essays about his condition edited by Bandy X. Lee, considered a half-dozen forms of pathology.

I’m not saying that diagnosis by AI would be beneficial. It’s easy to foresee disastrous scenarios, starting with character assassination by machine. But I can also imagine trusted, transparent programs developing a good track record and playing a leading part. My point is that in short order, for better and for worse, mental health experts may be consigned to secondary and perhaps more comfortable roles, explaining the significance of diagnostic judgments that they do not make.

I have been thinking about diagnosis for a second reason as well. I spent much of the past five years writing a novel whose hero, a psychiatrist, is dragooned into treating a buffoonish egotist known as the Great Man, a national leader floundering in his second term. In part, the exercise was my private attempt to come to terms with Donald Trump. What did I want to know about him?

The novel that emerged, “Death of the Great Man,” features a fictional psychiatrist named Henry Farber. He specializes in the treatment of paranoid men, but he uses the label loosely. A certain level of grandiosity, sense of injury, and suspiciousness will suffice. When it comes to diagnosis, Farber cares most about conditions like sleep apnea that may respond to specific, technical treatments. For the rest: He’s engaged in psychotherapy, intent on understanding his patient through empathy.

Farber’s curiosity largely reflects my own. I treated psychiatric patients for well over 40 years. I found distinctions between personality disorders — the sorts of conditions that Trump was tagged with in the 2016 campaign — to be of limited value. For one thing, the diagnoses are unstable. Patients called narcissistic on one visit may soon qualify for the label sociopathic. To the therapist, the patient will seem the same person throughout.

Once treatment is underway, the specifics of personality disorder diagnoses often wane in importance. In the novel, the Great Man is contrasted with a patient who suffers from very substantial paranoia but who when put to the test, turns out to be self-sacrificing.

Donald Trump’s nature, who he is in broad terms, is obvious. I knew that, to interest readers, I would have to depict granular traits beyond the reach of diagnosis. I was creating a comic character — the Great Man is his own person — but we would need to know about him what the public cannot know about Trump. Does he lie to his wife? In private, is he amusing or terrifying?

Writing the book made me think that categorization, especially when it comes to personality disorders, is among the least interesting uses of psychiatric expertise. In one way, it is strange for psychiatrists to insist on the importance of diagnosing candidates for office. We want our mentally ill patients to be able to discuss their disorders openly, to exercise their talents fully, and to be judged by their achievements rather than the ailments we treat them for.

I suppose that’s why, reading about Machiavellian narcissism, I was not overly alarmed — not in my narrow, professional role — at the prospect of outsourcing the task of public diagnosis. Psychiatry’s best function is explaining the meaning and limitations of our various labels and the range of expected behaviors that they predict.

I am extremely worried at the prospect of Trump’s return to the Presidency. In my novel, the Great Man’s second term is dystopian. But Trump has a track record. In this go-round, won’t psychiatric labels be superfluous? The best predictor of future behavior is not diagnosis, it’s past behavior. We’ve seen Trump in action. One way or another, isn’t that enough?

Peter D. Kramer is the author of eight books including the international bestseller “Listening to Prozac” and, most recently, the novel “Death of the Great Man.” Dr. Kramer is an emeritus professor of psychiatry at Brown University.

Source: STAT