It’s hard to know what’s real, it turns out.
Sometimes things are obvious. When it’s raining, there’s no doubt about whether it’s raining or not. A flat tire, in all likelihood, really is a flat tire. But much of the world is not directly verifiable in this way. Are there really millions of dead people collecting Social Security? Does China secretly control the Panama Canal? Was that video on TikTok actually Tom Cruise? How do you know for sure?
The rapid ascent of generative AI is only making things worse. Within the past few years, it’s been used to create political deepfakes from Moldova to South Africa to the United States. An AI-generated Imran Khan, Pakistan’s ex-prime minister, gave a speech while the real Khan was in jail. AI-generated evidence is now being introduced in divorce court, and last year AI was used to impersonate a company’s CFO and steal $25 million. According to one survey, deepfake incidents rose in 2024 by 300% in the United States, 1,625% in South Korea, 2,800% in China, and 3,000% in Bulgaria. Even the suggestion of an AI deepfake is now enough to cast doubt on anything, from the size of political rallies to legal culpability in criminal court.
Last year, in an attempt to curtail the spread of misinformation, Meta launched a campaign called “Know What’s Real,” encouraging users to verify things before sharing them. A few weeks later The Atlantic launched a podcast called, similarly, “How to Know What’s Real.” Focusing on technology, the series would examine “how to determine what is authentic and true.” It’s an exceedingly important idea, of course, especially now that conspiracies are mainstream, deepfake content is exploding, and nobody can agree on what to believe. But is it even possible to know anymore?
Before he became prime minister of the U.K., a recording allegedly of British politician Keir Starmer swearing at staff went viral on X. Aides to Starmer suggested the clip was a deepfake, as did several members of the opposition party. But analysis by an expert was inconclusive. “Unfortunately there is no definitive yes/no,” Tony Thompson wrote for the British non-profit Full Fact. “It remains very difficult to confirm deepfakes with total certainty.”
The problem isn’t necessarily that we lack certainty. In some ways, it’s that we have too much of it. Many people are certain that COVID-19 originated in a lab, or not. Or that vaccines are safe and effective, or not. Or that climate change is real, or not. Verification or fact-checking may seem like antidotes to such polarization. But, in deepfakes and everything else, there is often not a definitive yes or no answer. As Oliver Wendell Holmes, Jr. wrote in 1897, “certainty generally is illusion.”
The problem isn’t necessarily that we lack certainty. In some ways, it’s that we have too much of it.
A better approach, instead, is to abandon our attachment to certainty. In its place, we should learn to embrace uncertainty, and think in terms of probability. It isn’t necessarily a new idea. Noted theoretical physicist Richard Feynman, for one, was an apostle for it. “It is absolutely necessary, for progress in science, to have uncertainty as a fundamental part of your inner nature,” he said in 1956. Feynman was echoing Albert Einstein, who said in 1921 that “as far as the propositions of mathematics refer to reality, they are not certain; and as far as they are certain, they do not refer to reality.” And Einstein was echoing French mathematician Blaise Pascal, who wrote in the 17th century that “we sail within a vast sphere, ever drifting in uncertainty.”
As a physician, I’m used to uncertainty. Most medical problems transpire inside the body, hidden from view. I can’t touch the clot in a patient’s coronary, or see their stroke in the same way I can see a flat tire. At work, I can rarely be totally certain about anything. That’s just the nature of medicine. As Canadian physician William Osler said, medicine is “a science of uncertainty, and an art of probability.” But it’s also the nature of the world. I can’t personally verify the weather in Tucson, or general relativity, or the reality of climate change, either.
Instead, I weigh and consider a constellation of indirect information, and make a judgment about what I think is likely true. To do this, I’ve been taught to think in probabilities, as Osler suggested, because probability is a yardstick for uncertainty. Physicians and scientists everywhere are trained to think this way. But this kind of thinking, called Bayesian inference (after the English mathematician and clergyman Thomas Bayes), is something that can benefit everyone. Especially in a time when the basic facts of what’s real, and what’s not, have come into question.
Want more health and science stories in your inbox? Subscribe to Salon’s weekly newsletter Lab Notes.
Here’s an example from my world in the ER. A man in his 60s comes in with crushing chest pain and trouble breathing. I’m worried about a heart attack, so I order an EKG. The EKG is normal. Should I send the patient home? I know that EKGs can be falsely normal — given a hundred heart attacks, they might miss a few dozen. By thinking like a Bayesian, I recognize that a normal EKG lowers the probability of heart attack, but not to zero. It’s more like evolving degrees of maybe, rather than a definitive yes or no. I’m still calling the cardiologist.
Chances are, you already do this sometimes. When you speed past a police car, you probably look for brake lights in your mirror. Why do you do that? Answer: if the officer isn’t preparing to turn around, then the likelihood that they noticed you is less. You feel reassured. New information, new probability. That’s Bayesian inference.
Recently, I was sitting with my grandfather on his porch in Texas. He had just turned 90, and it was hot. The topic of climate change came up. My uncle, sitting nearby, shook his head. He didn’t think God had given man the power to change the world like that. They’re both churchgoing, but my grandfather didn’t seem so sure anymore about the degree of human influence. There’s how many billions of people now? Eight, I said. He nodded. Maybe that is enough to change it, he said. That’s Bayesian inference, too.
Using this kind of thinking more deliberately can help us navigate the world in a more careful, measured way. After the alleged clip of Kier Starmer appeared, fact checkers took a closer look at the X account behind it. It had made other unsubstantiated claims about Starmer, supported by what looked like digitally manipulated images. Whatever our initial suspicions, this information should probably swing us toward skepticism. Rather than reaching for certainty, we can instead smartly manage our uncertainty.
Last year, I saw an oncologist for recurrent fevers. I was concerned about lymphoma, because my father had it. My white blood cell count was low too, which can also be a sign of cancer. With unexplained fevers, neutropenia and a family history, my suspicion that I might have cancer was fairly high. I was worried about the possibility. But then the tests all came back normal, and I was relieved. Not because cancer had been ruled out — it’s possible that the tests might have missed a subtle malignancy, still brewing. My relief came instead from recognition that the probability of cancer was now much lower. Whatever uncertainty remained, I could live with it.
We need your help to stay independent
A central challenge of our time is that many have come to conflate what’s possible with what’s probable. Is it possible that Starmer lost his temper, vaccines are deadly, and climate change is a hoax? Yes, all of these things are possible, insofar as they are conceivable. But that shouldn’t be enough to command belief. How likely is it that the government is suppressing evidence that vaccines kill, or that the entire field of climate science is brainwashed, part of a vast conspiracy, or just plain wrong? What incredible cascade of unlikely things would have to happen for these things to be true?
Instead of reflexively defaulting to one position or another, we can cultivate a sense of likelihood by weighing all available evidence, incorporating new information as it arrives. And information can arrive at any moment, demanding a degree of perpetual open mindedness. “I have approximate answers and possible beliefs in different degrees of certainty about different things,” Feynman said in 1981, “but I’m not absolutely sure of anything.”
It isn’t easy to think this way; simple, emotionally compelling narratives are far more seductive than probabilities. Certainty is way more palatable than doubt. “We burn with desire to find solid ground,” Pascal wrote more than three hundred years ago, “but our whole groundwork cracks, and the earth opens to abysses.” We don’t like it. We want to know for sure. The Internet, hailed as a great democratizer of knowledge, has only made things trickier. By fielding an infinite churn of data, clues can be found to support any notion, no matter how outlandish. AI, with its ability to conjure deepfakes on command, will make things trickier still.
But uncertainty is of the fabric of the world. Quantum mechanics runs on it. Some neuroscientists believe our brains, on a deeply subconscious level, already do too. All the more reason to embrace it — our “true state,” in Pascal’s words — and arm yourselves with probability, as we move deeper into this strange new future.
