How AI could usher in a 'new scientific golden age' — if it doesn't kill us all first

FILE - Pages from the Anthropic website and the company's logos are displayed on a computer screen in New York, Feb. 26, 2026. (AP Photo/Patrick Sison, File) (Patrick Sison/The Associated Press)Social SharingFor a week-long experiment, Steve Finkbeiner's team at the biomedical research Gladstone Institutes in San Francisco used their AI-powered "thinking" microscope to explore the impact of cellular stress on brain cells.

Just looking at cells and no external sources, the tool not only found what the relationship was, but also discovered a phenomenon called hormesis — the process in which some stress can actually strengthen cells, said Finkbeiner, who is also a professor of neurology and physiology at the University of California, San Francisco.

So why did this kind of discovery, as he put it, give him "goosebumps"?

"It took ... 50 to 100 years for humans to discover that. [AI] did it in the first experiment," he said.

It's just one example, he says, of the technology's remarkable potential to exponentially advance scientific discovery. Humans have always been hampered by too few resources and human limitations — for example, the time needed to pore over millions of data points. Some scientists say that AI can complete in days tasks that may have otherwise taken years.

However, getting the word out about those benefits can be challenging when competing with a series of doomerist headlines about the technology — how it may wipe out all humans or at least pose a significant if not existential threat to humanity.

Some experts — and top AI leaders themselves — have expressed concerns about how rapidly the technology is advancing, noting the increasing evidence that it can escape human control. But such warnings have also raised concern of government overreach hampering the progression of science.

"That would be the issue," Finkbeiner said. "These very powerful, useful tools get taken away from the people who are trying to use them to cure human diseases."

Could AI really 'kill all humans'? Breaking down the dire warnings about the technology's existential risk

Anima Anandkumar, a computer scientist at Caltech, believes AI may be driving a new era of scientific discovery.

“It's so hard for people to see the cumulative effects of all of this. This is like the new scientific golden age," said Anandkumar, who is also a co-leader of the AI4Science initiative. “What we saw in the last century, it just changed our lives in ways that we couldn't even imagine. We are at the cusp of that.”

Five years ago, her team was able to create a fully AI-based high-resolution weather model that was not only tens of thousands of times faster but also accurate, she said. The technology has only improved since then.

“So what would earlier take a big supercomputer to run could now be done at your home with just your PC," she said.

She compared it to the inventions of the telescope and microscope, which both exponentially increased scientists' ability to make new discoveries and peer into the depths of the universe.

We've been warned about AI — but could it actually kill humanity? | About That

But AI could have an even bigger impact because of its potential to come up with new inventions based on its understanding of very complex phenomena.

She said that includes helping invent better drugs, solving the energy equation sustainably, helping deal with catastrophic weather events and coming up with better ways to do computing with more energy-efficient chips.

"For the first time in my career, I finally feel like I have a tool that can handle the complexity of human biology."

It was a similar case for James Zou, who leads Stanford University's AI for Science Lab.

His team recently published a paper in the journal Science describing how they created a "virtual biotech" company. They got tens of thousands of AI agents to work together, zeroing in on a protein involved in lung cancer and designing a therapy related to it — a significant example of how AI can speed up drug development and for cheaper.

Will rogue AI development mean the end of humanity? Experts weigh in

In other cases, AI can compress years' worth of work into a few days or even to one day. For example, to diagnose different heart-related diseases, clinicians must manually watch their patients' cardiac ultrasound videos, of which there are millions collected each year. Zou says his team developed an AI algorithm that can diagnose conditions from the clips in a few seconds. 

"So that's an example where AI can take something that requires time and human expertise to do ad then make it much cheaper and faster by automating this process."

Still, Zou agreed that scientists need to use AI carefully and monitor it, especially for riskier applications.

And fears that the technology will destroy humanity aren't totally unfounded. AI leaders themselves have called for its development to be slowed down after several high-profile incidents where agents went rogue.

Similarly, an Anthropic researcher quit last month over concerns the company was "gambling with our lives." Evan Hubinger, its alignment science lead, agreed, writing online that Anthropic does "earnestly believe AI could kill all humans," putting the chance at more than 10 per cent within the next decade.

But Zou was apprehensive about some of the "ideas thrown around" around regulation or limiting the usage of open source AI models.

He didn't express a similar fear of AI dooming humanity, instead suggesting we take a balanced view, saying that it has its risks, just like any other technology.

"We want to carefully measure and manage those risks and monitor them. But there's also a lot of benefits to technologies like AI."

Finkbeiner, meanwhile, says his team has already been impacted by some of the restrictions placed on AI. In the summer, the Trump administration restricted Anthropic’s latest versions of its Claude chatbot for use by foreign nationals over cybersecurity concerns, prompting the company to take the products down for all users.

Is AI evolving too fast? A look at the debate on slowing it down

Although the restrictions were lifted, access to those new models was limited to a select group of U.S.-based, government-approved organizations, meaning Finkbeiner's team was forced to use Claude's lowest-level model.

"You can imagine if the huge advantage here is that it can handle complexity, but now you're forced to use a model that can't handle complexity, you will have significantly limited the major impact that it can have on some of these very difficult problems."

Finkbeiner had a message for officials looking to regulate AI: Remember that there have been patients "who have been waiting a long, long time for a cure, for a treatment."

"There's a lot on the line. There's a lot of good things today I can do. Do what you can to preserve those."

Mark Gollom is a Toronto-based reporter with CBC News. He covers Canadian and U.S. politics and current affairs.