New AI systems such as ChatGPT, the overhauled Microsoft Bing search engine, and the reportedly soon-to-arrive GPT-4 have utterly captured the public imagination. ChatGPT is the fastest-growing online application, ever, and it’s no wonder why. Type in some text, and instead of getting back web links, you get well-formed, conversational responses on whatever topic you selected—an undeniably seductive vision.

But the public, and the tech giants, aren’t the only ones who have become enthralled with the Big Data–driven technology known as the large language model. Bad actors have taken note of the technology as well. At the extreme end, there’s Andrew Torba, the CEO of the far-right social network Gab, who said recently that his company is actively developing AI tools to “uphold a Christian worldview” and fight “the censorship tools of the Regime.” But even users who aren’t motivated by ideology will have their impact. Clarkesworld, a publisher of sci-fi short stories, temporarily stopped taking submissions last month, because it was being spammed by AI-generated stories—the result of influencers promoting ways to use the technology to “get rich quick,” the magazine’s editor told The Guardian.  

This is a moment of immense peril: Tech companies are rushing ahead to roll out buzzy new AI products, even after the problems with those products have been well documented for years and years. I am a cognitive scientist focused on applying what I’ve learned about the human mind to the study of artificial intelligence. Way back in 2001, I wrote a book called The Algebraic Mind in which I detailed then how neural networks, a kind of vaguely brainlike technology undergirding some AI products, tended to overgeneralize, applying individual characteristics to larger groups. If I told an AI back then that my aunt Esther had won the lottery, it might have concluded that all aunts, or all Esthers, had also won the lottery.

Technology has advanced quite a bit since then, but the general problem persists. In fact, the mainstreaming of the technology, and the scale of the data it’s drawing on, has made it worse in many ways. Forget Aunt Esther: In November, Galactica, a large language model released by Meta—and quickly pulled offline—reportedly claimed that Elon Musk had died in a Tesla car crash in 2018. Once again, AI appears to have overgeneralized a concept that was true on an individual level (someone died in a Tesla car crash in 2018) and applied it erroneously to another individual who happens to shares some personal attributes, such as gender, state of residence at the time, and a tie to the car manufacturer.

This kind of error, which has come to be known as a “hallucination,” is rampant. Whatever the reason that the AI made this particular error, it’s a clear demonstration of the capacity for these systems to write fluent prose that is clearly at odds with reality. You don’t have to imagine what happens when such flawed and problematic associations are drawn in real-world settings: NYU’s Meredith Broussard and UCLA’s Safiya Noble are among the researchers who have repeatedly shown how different types of AI replicate and reinforce racial biases in a range of real-world situations, including health care. Large language models like ChatGPT have been shown to exhibit similar biases in some cases.

Nevertheless, companies press on to develop and release new AI systems without much transparency, and in many cases without sufficient vetting. Researchers poking around at these newer models have discovered all kinds of disturbing things. Before Galactica was pulled, the journalist Tristan Greene discovered that it could be used to create detailed, scientific-style articles on topics such as the benefits of anti-Semitism and eating crushed glass, complete with references to fabricated studies. Others found that the program generated racist and inaccurate responses. (Yann LeCun, Meta’s chief AI scientist, has argued that Galactica wouldn’t make the online spread of misinformation easier than it already is; a Meta spokesperson told CNET in November, “Galactica is not a source of truth, it is a research experiment using [machine learning] systems to learn and summarize information.”)

More recently, the Wharton professor Ethan Mollick was able to get the new Bing to write five detailed and utterly untrue paragraphs on dinosaurs’ “advanced civilization,” filled with authoritative-sounding morsels including “For example, some researchers have claimed that the pyramids of Egypt, the Nazca lines of Peru, and the Easter Island statues of Chile were actually constructed by dinosaurs, or by their descendents or allies.” Just this weekend, Dileep George, an AI researcher at DeepMind, said he was able to get Bing to create a paragraph of bogus text stating that OpenAI and a nonexistent GPT-5 played a role in the Silicon Valley Bank collapse. Microsoft did not immediately answer questions about these responses when reached for comment; last month, a spokesperson for the company said, “Given this is an early preview, [the new Bing] can sometimes show unexpected or inaccurate answers … we are adjusting its responses to create coherent, relevant and positive answers.”

[Read: Conspiracy theories have a new best friend]

Some observers, like LeCun, say that these isolated examples are neither surprising nor concerning: Give a machine bad input and you will receive bad output. But the Elon Musk car crash example makes clear these systems can create hallucinations that appear nowhere in the training data. Moreover, the potential scale of this problem is cause for worry. We can only begin to imagine what state-sponsored troll farms with large budgets and customized large language models of their own might accomplish. Bad actors could easily use these tools, or tools like them, to generate harmful misinformation, at unprecedented and enormous scale. In 2020, Renée DiResta, the research manager of the Stanford Internet Observatory, warned that the “supply of misinformation will soon be infinite.” That moment has arrived.

Each day is bringing us a little bit closer to a kind of information-sphere disaster, in which bad actors weaponize large language models, distributing their ill-gotten gains through armies of ever more sophisticated bots. GPT-3 produces more plausible outputs than GPT-2, and GPT-4 will be more powerful than GPT-3. And none of the automated systems designed to discriminate human-generated text from machine-generated text has proved particularly effective.

[Read: ChatGPT is about to dump more work on everyone]

We already face a problem with echo chambers that polarize our minds. The mass-scale automated production of misinformation will assist in the weaponization of those echo chambers and likely drive us even further into extremes. The goal of the Russian “Firehose of Falsehood” model is to create an atmosphere of mistrust, allowing authoritarians to step in; it is along these lines that the political strategist Steve Bannon aimed, during the Trump administration, to “flood the zone with shit.” It’s urgent that we figure out how democracy can be preserved in a world in which misinformation can be created so rapidly, and at such scale.  

One suggestion, worth exploring but likely insufficient, is to “watermark” or otherwise track content that is produced by large language models. OpenAI might for example watermark anything generated by GPT-4, the next-generation version of the technology powering ChatGPT; the trouble is that bad actors could simply use alternative large language models to create whatever they want, without watermarks.

A second approach is to penalize misinformation when it is produced at large scale. Currently, most people are free to lie most of the time without consequence, unless they are, for example, speaking under oath. America’s Founders simply didn’t envision a world in which someone could set up a troll farm and put out a billion mistruths in a single day, disseminated with an army of bots, across the internet. We may need new laws to address such scenarios.

A third approach would be to build a new form of AI that can detect misinformation, rather than simply generate it. Large language models are not inherently well suited to this; they lose track of the sources of information that they use, and lack ways of directly validating what they say. Even in a system like Bing’s, where information is sourced from the web, mistruths can emerge once the data are fed through the machine. Validating the output of large language models will require developing new approaches to AI that center reasoning and knowledge, ideas that were once popular but are currently out of fashion.  

It will be an uphill, ongoing move-and-countermove arms race from here; just as spammers change their tactics when anti-spammers change theirs, we can expect a constant battle between bad actors striving to use large language models to produce massive amounts of misinformation and governments and private corporations trying to fight back. If we don’t start fighting now, democracy may well be overwhelmed by misinformation and consequent polarization—and perhaps quite soon. The 2024 elections could be unlike anything we have seen before.

To be a newborn in the year 2023—and, almost certainly, every year that follows—means emerging into a world where the coronavirus is ubiquitous. Babies might not meet the virus in the first week or month of life, but soon enough, SARS-CoV-2 will find them. “For anyone born into this world, it’s not going to take a lot of time for them to become infected,” maybe a year, maybe two, says Katia Koelle, a virologist and infectious-disease modeler at Emory University. Beyond a shadow of a doubt, this virus will be one of the very first serious pathogens that today’s infants—and all future infants—meet.

Three years into the coronavirus pandemic, these babies are on the leading edge of a generational turnover that will define the rest of our relationship with SARS-CoV-2. They and their slightly older peers are slated to be the first humans who may still be alive when COVID-19 truly hits a new turning point: when almost everyone on Earth has acquired a degree of immunity to the virus as a very young child.

[Read: Is COVID a common cold yet?]

That future crossroads might not sound all that different from where the world is currently. With vaccines now common in most countries and the virus so transmissible, a significant majority of people have some degree of immunity. And in recent months, the world has begun to witness the consequences of that shift. The flux of COVID cases and hospitalizations in most countries seems to be stabilizing into a seasonal-ish sine wave; disease has gotten, on average, less severe, and long COVID seems to be somewhat less likely among those who have recently gotten shots. Even the virus’s evolution seems to be plodding, making minor tweaks to its genetic code, rather than major changes that require another Greek-letter name.

But today’s status quo may be more of a layover than a final destination in our journey toward COVID’s final form. Against SARS-CoV-2, most little kids have fared reasonably well. And as more babies have been born into a SARS-CoV-2-ridden world, the average age of first exposure to this coronavirus has been steadily dropping—a trend that could continue to massage COVID-19 into a milder disease. Eventually, the expectation is that the illness will reach a stable nadir, at which point it may truly be “another common cold,” says Rustom Antia, an infectious-disease modeler at Emory.

The full outcome of this living experiment, though, won’t be clear for decades—well after the billions of people who encountered the coronavirus for the first time in adulthood are long gone. The experiences that today’s youngest children have with the virus are only just beginning to shape what it will mean to have COVID throughout a lifetime, when we all coexist with it from birth to death as a matter of course.

At the beginning of SARS-CoV-2’s global tear, the coronavirus was eager to infect all of us, and we had no immunity to rebuff its attempts. But vulnerability wasn’t just about immune defenses: Age, too, has turned out to be key to resilience. Much of the horror of the disease could be traced to having not only a large population that lacked protection against the virus—but a large adult population that lacked protection against the virus. Had the entire world been made up of grade-schoolers when the pandemic arrived, “I don’t think it would have been nearly as severe,” says Juliet Pulliam, an infectious-disease modeler at Stellenbosch University, in South Africa.

Across several viral diseases—polio, chicken pox, mumps, SARS, measles, and more—getting sick as an adult is notably more dangerous than as a kid, a trend that’s typically exacerbated when people don’t have any vaccinations or infections to those pathogens in their rearview. The manageable infections that strike toddlers and grade-schoolers may turn serious when they first manifest at older ages, landing people in the hospital with pneumonia, brain swelling, even blindness, and eventually killing some. When scientists plot mortality data by age, many curves bend into “a pretty striking J shape,” says Dylan Morris, an infectious-disease modeler at UCLA.

The reason for that age differential isn’t always clear. Some of kids’ resilience probably comes from having a young, spry body, far less likely to be burdened with chronic medical conditions that raise severe disease risk. But the quick-wittedness of the young immune system is also likely playing a role. Several studies have found that children are much better at marshaling hordes of interferon—an immune molecule that armors cells against viruses—and may harbor larger, more efficient cavalries of infected-cell-annihilating T cells. That performance peaks sometime around grade school or middle school, says Janet Chou, a pediatrician at Boston Children’s Hospital. After that, our molecular defenses begin a rapid tumble, growing progressively creakier, clumsier, sluggish, and likelier to launch misguided attacks against the tissues that house them. By the time we’re deep into adulthood, our immune systems are no longer sprightly, or terribly well calibrated. When we get sick, our bodies end up rife with inflammation. And our immune cells, weary and depleted, are far less unable to fight off the pathogens they once so easily trounced.

Whatever the explanations, children are far less likely to experience serious symptoms, or to end up in the hospital or the ICU after being infected with SARS-CoV-2. Long COVID, too, seems to be less prevalent in younger cohorts, says Alexandra Yonts, a pediatrician at Children’s National Hospital. And although some children still develop MIS-C, a rare and dangerous inflammatory condition that can appear weeks after they catch the virus, the condition “seems to have dissipated” as the pandemic has worn on, says Betsy Herold, the chief of pediatric infectious disease at the Children’s Hospital at Montefiore, in the Bronx.

Should those patterns hold, and as the age of first exposure continues to fall, COVID is likely to become less intense. The relative mildness of childhood encounters with the virus could mean that almost everyone’s first infection—which tends, on average, to be more severe than the ones that immediately follow—could rank low in intensity, setting a sort of ceiling for subsequent bouts. That might make concentrating first encounters “in the younger age group actually a good thing,” says Ruian Ke, an infectious-disease modeler at Los Alamos National Laboratory.

COVID will likely remain capable of killing, hospitalizing, and chronically debilitating a subset of adults and kids alike. But the hope, experts told me, is that the proportion of individuals who face the worst outcomes will continue to drop. That may be what happened in the aftermath of the 1918 flu pandemic, Antia, of Emory, told me: That strain of the virus stuck around, but never caused the same devastation again. Some researchers suspect that something similar may have even played out with another human coronavirus, OC43: After sparking a devastating pandemic in the 19th century, it’s possible that the virus no longer managed to wreak much more havoc than a common cold in a population that had almost universally encountered it early in life.

Such a fate for COVID, though, isn’t a guarantee. The virus’s propensity to linger in the body’s nooks and crannies, sometimes causing symptoms that last many months or years, could make it an outlier among its coronaviral kin, says Melody Zeng, an immunologist at Cornell University. And even if the disease is likely to get better than what it is now, that is not a very high bar to clear.

Some small subset of the population will always be naive to the virus—and it’s not exactly a comfort that in the future, that cohort will almost exclusively be composed of our kids. Pediatric immune systems are robust, UCLA’s Morris told me. But “robust is not the same as infallible.” Since the start of the pandemic, more than 2,000 Americans under the age of 18 have died from COVID—a small fraction of total deaths, but enough to make the disease a leading cause of death for children in the U.S. MIS-C and long COVID may not be common, but their consequences are no less devastating for the children who experience them. Some risks are especially concentrated among our youngest kids, under the age 5, whose immune defenses are still revving up, making them more vulnerable than their slightly older peers. There’s especially little to safeguard newborns just under six months, who aren’t yet eligible for most vaccines—including COVID shots—and who are rapidly losing the antibody-based protection passed down from their mothers while they were in the womb.

A younger average age of first infection will also probably increase the total number of exposures people have to SARS-CoV-2 in a typical lifetime—each instance carrying some risk of severe or chronic disease. Ke worries the cumulative toll that this repetition could exact: Studies have shown that each subsequent tussle with the virus has the potential to further erode the functioning or structural integrity of organs throughout the body, raising the chances of chronic damage. There’s no telling how many encounters might push an individual past a healthy tipping point.

Racking up exposures also won’t always bode well for the later chapters of these children’s lives. Decades from now, nearly everyone will have banked plenty of encounters with SARS-CoV-2 by the time they reach advanced age, Chou, from Boston Children’s Hospital, told me. But the virus will also continue to change its appearance, and occasionally escape the immunity that some people built up as kids. Even absent those evasions, as their immune systems wither, many older people may not be able to leverage past experiences with the disease to much benefit. The American experience with influenza is telling. Despite a lifetime of infections and available vaccines, tens of thousands of people typically die annually of the disease in the United States alone, says Ofer Levy, the director of the Precision Vaccines Program at Boston Children’s Hospital. So even with the expected COVID softening, “I don’t think we’re going to reach a point where it’s, Oh well, tra-la-la,” Levy told me. And the protection that immunity offers can have caveats: Decades of research with influenza suggest that immune systems can get a bit hung up on the first versions of a virus that they see, biasing them against mounting strong attacks against other strains; SARS-CoV-2 now seems to be following that pattern. Depending on the coronavirus variants that kids encounter first, their responses and vulnerability to future bouts of illness may vary, says Scott Hensley, an immunologist at the University of Pennsylvania.

[Read: Are our immune systems stuck in 2020?]

Early vaccinations—that ideally target multiple versions of SARS-CoV-2—could make a big difference in reducing just about every bad outcome the virus threatens. Severe disease, long COVID, and transmission to other children and vulnerable adults all would likely be “reduced, prevented, and avoided,” Chou told me. But that’s only if very young kids are taking those shots, which, right now, isn’t at all the case. Nor are they necessarily getting protection passed down during gestation or early life from their mothers, because many adults are not up to date on COVID shots.

Some of these issues could, in theory, end up moot. A hundred or so years from now, COVID could simply be another common cold, indistinguishable in practice from any other. But Morris points out that this reality, too, wouldn’t fully spare us. “When we bother to look at the burden of the other human coronaviruses, the ones who have been with us for ages? In the elderly, it’s real,” he told me. One study found that a nursing-home outbreak of OC43—the purported former pandemic coronavirus—carried an 8 percent fatality rate; another, caused by NL63, killed three out of the 20 people who caught it in a long-term-care facility in 2017. These and other “mild” respiratory viruses also continue to pose a threat to people of any age who are immunocompromised.

SARS-CoV-2 doesn’t need to follow in those footsteps. It’s the only human coronavirus against which we have vaccines—which makes the true best-case scenario one in which it ends up even milder than a common cold, because we proactively protect against it. Disease would not need to be as inevitable; the vaccine, rather than the virus, could be the first bit of intel on the disease that kids receive. Tomorrow’s children probably won’t live in a COVID-free world. But they could at least be spared many of the burdens we’re carrying now.