The scientific sprint: How AI is rewriting discovery timelines

A new breed of artificial intelligence threatens to upend centuries of scientific tradition by conducting experiments faster and at scales impossible for humans to match.

While scientists have long relied on intuition, methodical research and painstaking trial and error, companies like Lila Sciences are now building what they call "Science Factories"—autonomous labs where AI systems generate hypotheses, design experiments and analyze results with minimal human intervention. This shift could fundamentally transform how scientific discoveries are made, potentially accelerating breakthroughs in everything from clean energy to medicine.

"We define scientific superintelligence as the ability to conduct the scientific method at a level beyond human intelligence at every step of the process," Geoffrey von Maltzahn, Co-Founder and CEO of Lila Sciences, tells IBM Think in an interview. "The advantage of doing that is self-evident. It could be as consequential as the leap from human science to industrial science."

Recently spun out from Flagship Pioneering with USD 200 million in funding from investors, including General Catalyst, March Capital and the Abu Dhabi Investment Authority, Lila Sciences exemplifies this emerging paradigm. The company's approach integrates generative AI with robotics to conduct thousands of experiments simultaneously—a scale far beyond traditional research methods.

"If you look at how we've approached science for centuries, it's been through human intuition and trial and error," von Maltzahn says. "With AI-driven experimentation, we can scale that process exponentially. Instead of testing one hypothesis at a time, Lila's autonomous labs can generate and execute thousands of experiments simultaneously, optimizing results in real-time."

The results are already proving impressive. In one demonstration, von Maltzahn said, Lila's platform discovered novel, non-platinum-group metal catalysts for green hydrogen production in just four months, with new catalyst compositions emerging roughly every two weeks. Using conventional research methods, experts had estimated this would take a decade.

This vision of AI-accelerated science aligns with what some have called a "compressed 21^st century"—the idea that AI could enable humanity to achieve decades of scientific progress in just a few years. In a recent essay, Anthropic CEO Dario Amodei argued that powerful AI could dramatically compress the timeline of scientific discovery.

"My basic prediction is that AI-enabled biology and medicine will allow us to compress the progress that human biologists would have achieved over the next 50-100 years into 5-10 years," Amodei wrote. "I'll refer to this as the 'compressed 21st century': the idea that after powerful AI is developed, we will in a few years make all the progress in biology and medicine that we would have made in the whole 21^st century."

Lila isn't alone in this pursuit. Tech giants have invested heavily in AI-driven scientific research, leveraging machine learning to predict molecular interactions, generate new materials and streamline drug discovery. However, Lila argues that its competitive edge lies in its ability to scale the experimental process itself.

"Our focus is on scaling not just AI models but the experimental process," von Maltzahn says. "The companies that win in this space won't just be the ones with the biggest AI models—they'll be the ones that can run the most brilliant real-world experiments, at the largest scale and fastest speed."

The rise of scientific superintelligence raises profound questions about the future role of human scientists. Will AI eventually replace them? Or will it serve as a powerful collaborator enhancing human ingenuity?

IBM Principal Research Scientist Payel Das told IBM Think she believes AI will augment rather than replace human creativity. "The human-AI interaction remains central to turbocharging the discovery process," she says. "Human experts define the problem space and provide guidance, while AI agents perform tasks like learning, reasoning and suggesting viable solutions."

Not everyone shares this optimistic vision of AI's transformative potential in scientific discovery. Hugging Face's Co-Founder, Thomas Wolf recently challenged the idea that AI will deliver a compressed 21^st century.

"What we'll actually get, in my opinion, is 'a country of yes-men on servers,'" Wolf wrote in a recent X post. He argues that true scientific breakthroughs come not from answering known questions but from challenging established paradigms. "A real science breakthrough is Copernicus proposing, against all the knowledge of his days—in ML terms we would say 'despite all his training dataset'—that the earth may orbit the sun rather than the other way around."

Wolf's skepticism contrasts with the real-world progress that Das has observed in laboratories worldwide. She points to AI's already demonstrable impact across multiple scientific fields. "AI has contributed to discoveries of broad-spectrum antibiotics and SARS-CoV-2 inhibitors, though these breakthroughs have come from multiple research efforts across the scientific community rather than a single organization," she notes. "These are problems that would have taken far longer using traditional methods."

However, she emphasizes that AI's effectiveness is contingent on proper integration with human expertise. "The ability of human experts to define problem spaces and provide contextual guidance is crucial. AI models can interact with knowledge bases and domain-specific tools, but their effectiveness relies on human oversight to ensure meaningful and responsible scientific discovery."

For all its promise, AI-driven science remains an unproven business model. While Lila has demonstrated early breakthroughs, monetizing those discoveries presents a different challenge.

"Our roadmap is to partner with established companies to bring these discoveries to market," von Maltzahn explains. "We don't plan to manufacture drugs or commercialize clean energy solutions ourselves. Instead, we aim to work with industry leaders who can scale these breakthroughs into viable products."

Researchers and ethicists are raising red flags about potential dangers as companies race to deploy these powerful scientific tools. AI's use in scientific research comes with inherent risks. "On their own, AI models are at risk of coming up with potentially unsafe solutions, such as bioweapons or toxic materials," Das cautions. "Humans with malicious intentions could also steer these models in dangerous directions."

To mitigate these risks, Das advocates for strong governance frameworks. "We need rigorous auditing of model development, deployment and downstream usage," she says. "A full-spectrum governance framework should include user intent detection, model coverage awareness and access constraints on AI-generated discoveries."

Lila insists it takes safety seriously, implementing layers of oversight to prevent unintended consequences. "AI-generated discoveries don't go straight into the world—they go through rigorous validation before they're applied in real-world settings," von Maltzahn says.

Concerns remain about whether these safeguards will be sufficient as AI-driven scientific discovery advances and scales.

Von Maltzahn predicts AI-driven experimentation will soon become the norm. "Our belief is that something like 100% of science is going to change hands in the next decade," he says. "The scientific process, which has been largely unchanged for centuries, will shift from being human-led to AI-assisted at every stage."

Whether Lila's vision of autonomous science factories becomes a dominant paradigm or remains an ambitious experiment will depend on the power of technology and how society chooses to wield it.

"We're standing at the edge of something transformative," Das says. "But the choices we make now—about regulation, access and ethical oversight—will determine whether this transformation benefits everyone or only a select few."