AI has a public relations problem. A technology that could do more to improve our lives than anything invented in decades is being demonized to the public as something that might wipe us out, taking our jobs and even ending humanity.

Researchers recently left leading laboratories with warnings of catastrophe. One Anthropic researcher puts the odds at above 10 percent of AI killing everyone within a decade. Attach a percentage to the apocalypse, and speculation starts to sound like a scientific finding.

Apparently, we’re all doomed. But I don’t buy it. Even if the negative headlines are the ones that get clicks, I believe AI will leave us considerably better off.

We have heard similar predictions before. In 1985, some warned that Excel would eliminate accountants. Since then, we have added about 600,000 new accounting jobs. In 1995, Ethernet pioneer Bob Metcalfe predicted a “catastrophic collapse” of the internet the following year. When it failed to happen, he put a copy of his column in a blender and literally ate his words.

Y2K was supposed to end the world as we knew it, causing failures in banking and power because of a date change. Midnight came and went — the apocalypse did not.

In each case, technologies that were supposed to make human beings useless helped make us richer, healthier and more capable. We didn’t just survive technological progress — we built better lives with it.

Global output per person has risen many times over since industrialization. Electricity, refrigeration, sanitation, and medicine all transformed ordinary life. Global life expectancy rose from roughly 32 years in 1900 to about 73 in 2023. Literacy spread. Poverty fell. In many countries, working hours fell substantially, giving people more leisure time. Progress wasn’t always neat and orderly, but it was real.

Most of us would hate living in the world our great-grandparents knew. We take it for granted every time we turn on a light, open a refrigerator or get an infection treated.

Of course, progress does cause lost livelihoods. But eliminating tasks is not the same as eliminating the need for people. Many jobs were replaced with higher-quality, better-paying jobs. More productive workers can do things that weren’t previously possible or worth trying. We find it much easier to name the jobs a machine might replace than the ones that don’t exist yet.

In fairness, AI’s speed of adoption makes that adjustment harder compared to previous technological revolutions. Workers cannot retrain at the speed of a software-release schedule. That is a serious problem to solve, but not evidence that the whole enterprise ends badly.

This fearmongering has many of the frontier labs calling for increased regulation to mitigate the potential negative effects of AI. However, the cost of complying with additional rules can fall unevenly across an industry. Consider the same compliance bill landing on the desks of a global corporation and a five-person startup. For one, it may be a routine expense. For the other, it may mean not hiring the next engineer. Safety standards can support more trustworthy products; fixed compliance costs can also change who can afford to compete.

There is a name for an industry gaining undue influence over its own oversight: regulatory capture. The history of banking offers examples. For AI, the practical questions are what a rule prevents, what compliance costs and which firms can pay. Even a rule with a genuine safety benefit can leave customers with fewer suppliers, less competition and innovation, and higher prices.

Ultimately, the marketplace will determine which products people find useful enough to keep using. Does the tool work? Is it reliable? Is it worth the money? A customer doesn’t owe a technology company loyalty just because it has a famous founder. Competition gives that customer somewhere else to go and gives a startup a chance to win the business. When only the big boys can afford to play, those choices shrink.

And this race does not stop at America’s borders. The U.S. and China are competing for leadership in AI, with enormous commercial and strategic stakes. A rule adopted in Washington does not automatically bind a developer in Beijing. Who builds the technology will help determine whose products businesses depend on, where investment goes and which countries develop the next generation of companies. America is not the only country deciding what happens next.

Meanwhile, the work that makes me optimistic is already happening. AI is helping scientists predict protein structures and molecular interactions, identify promising antibiotic compounds and test ways to predict which cancer drugs will work for individual patients. These are early advances, but they concern diseases that have defeated us for decades. They deserve more than passing mention between extinction warnings.

At one company, AI-assisted customer-support staff resolved 15 percent more issues per hour on average, with bigger gains among less experienced workers. For all the talk of making people obsolete, people here are getting better at their jobs.

AI is also helping researchers develop better forecasts of floods and wildfire risks. For a family in danger, an earlier warning could mean time to get out. For someone waiting to hear whether a cancer treatment will work, a better way to match drugs to patients could mean another chance. That’s what I have in mind when I talk about AI changing lives: better medicine, less drudgery and opportunities no one can yet imagine.

Yes, these systems need to be tested and monitored. But we have integrated powerful and potentially dangerous technologies into our lives before. Nuclear energy can power a city, even if nuclear weapons can destroy one. The capacity to do harm is real, but it’s not the whole story. That is up to us — the humans in the room.

Dr. Aaron Poynton is executive director of the American Society for AI and chief revenue officer of Applied General Intelligence. His views do not necessarily reflect those of these or other organizations.