Chemist Lee Cronin Slams the AI Hype: “AI Can’t Do Science — But We Mustn’t Miss Our Chance to Evolve With It”

Friday, 21 August 2026 at 07:00

Chemicus Lee Cronin haalt uit naar AI-hype 'AI kan geen wetenschap bedrijven, maar we moeten de kans niet missen om samen met AI te evolueren.'

According to British chemist Lee Cronin, artificial intelligence can’t do real science, isn’t truly creative, and lacks independent agency. The University of Glasgow professor and Chemify founder pushes back against the notion that ever-more powerful AI systems will one day make scientific discoveries on their own or fully replace human researchers.

Cronin made the striking remarks in a long conversation with astrophysicist David Kipping on the Cool Worlds Podcast. His critique is notable because he himself leans heavily on AI and automation in his lab and company. He sees AI as a potent way to accelerate exploration of what’s already possible—but draws a hard line between searching within existing knowledge and creating fundamentally new ideas.

“AI is going to do science? No. That’s not going to happen. In fact: never.”

That line anchors a broader debate over AI, creativity, and science. Cronin argues we too often confuse today’s AI capabilities with understanding, consciousness, and scientific autonomy.

1. Why Cronin says AI can’t do science

AI can support scientists, but, Cronin argues, it cannot independently take over the scientist’s role. Systems can help design experiments and search existing spaces faster, but that’s fundamentally different from doing science itself.

His sharpest critique targets the hype around “AI for science.” AI now analyzes literature, writes code, explores protein structures, and proposes candidate molecules. Useful, yes—but none of that proves AI understands science, he says.

“Today’s AI systems are what I call shallow explorers. They’re fast, shallow explorers.”

With that label, Cronin separates speed from intellectual depth. An AI can churn through far more options than a human, but it still moves within a possibility space bounded by existing data, representations, and rules.

A human scientist, he argues, can do something else entirely: reframe a problem, invent an unexpected abstraction, or open up a completely new possibility space.

That difference, he says, is exactly what drives real scientific breakthroughs.

2. AI can look smart without actually thinking

Cronin also resists the human traits too readily projected onto AI. A chatbot can persuasively explain how it arrived at an answer, but that doesn’t mean it thinks like a person.

“AI systems are not conscious. They do not think.”

The distinction matters more as generative AI mimics human behavior with increasing fluency. Modern models can converse, code, create images, show reasoning steps, and execute complex instructions.

To users, that output can easily feel like there’s a thinking mind behind it.

Cronin says that leap is premature.

He makes the same objection around “agency,” usually defined as pursuing goals and making decisions independently.

“AI has no independent agency. AI does not make decisions.”

That claim needs nuance. Modern AI agents can string together actions, adapt plans, and use tools inside software environments. Cronin’s point is deeper: these systems still operate within human-set goals, rules, and action spaces.

In his view, functional autonomy is not the same as biological—or human—agency.

3. The real AI problem: our definition of intelligence

For Cronin, AI’s rapid progress mostly exposes how little we understand intelligence, consciousness, and creativity. Computers now display behaviors we’ve long labeled as “thinking” and “creating,” even as the underlying processes remain unclear.

“We’re seeing things we can’t explain because we don’t yet have the right theory of mind, intelligence, and creativity.”

That’s a more important part of his argument than his starkest one-liners might suggest.

He isn’t saying modern AI is useless or simple. The issue is that these systems are now impressive enough to strain our definitions.

An AI model can generate an image that has never existed, write a new program, or propose an unfamiliar molecular structure. To a user, that output may feel undeniably creative.

But that still doesn’t prove, says Cronin, that the underlying process is creative.

He draws a clear line between creative output and a creative process.

4. Is generative AI truly creative, according to Cronin?

Generative AI produces new combinations, but Cronin doubts that anything fundamentally new emerges. He sees today’s models mainly as systems that remix existing elements and move probabilistically through known possibilities.

“AI isn’t creative. It creates nothing genuinely new.”

It’s likely the most contentious claim in the conversation.

After all, “generative AI” refers to systems that can generate novel text, images, audio, video, code, and more. Yet the mere fact that a specific image or sentence has never existed before isn’t enough for Cronin to call it true creativity.

His definition goes deeper.

Cronin looks at the size of the possibility space in which a solution must be found. When that space becomes vast, simple search can’t explain how humans arrive at certain inventions.

He uses a Boeing 747 as a thought experiment. An aircraft comprises an enormous number of parts that must fit together in specific ways. It didn’t emerge because people blindly tried every conceivable configuration until a working plane appeared by chance.

Humans created intermediate concepts, built abstractions, and invented new ways to decompose the problem.

Cronin suspects that’s precisely where something happens that current AI cannot reproduce.

5. Chemistry shows where digital AI hits its limits

The constraint becomes stark in the design of new molecules, says Cronin. Generative AI can propose vast numbers of candidates on a computer, but a digital design only has scientific value once it’s clear the molecule can actually be made and tested.

“Drawing a molecule on a computer doesn’t prove it exists.”

Cronin even argues that part of today’s digital chemistry boils down to generating molecules that look promising in silico, without clarity on whether they can be produced in practice.

His company, Chemify, aims to close that gap between digital prediction and physical reality.

The company is building programmable chemistry, where chemical procedures are described digitally and executed by automated lab systems. Such a system can probe candidate molecules, determine viable routes, make the compounds, and feed experimental results back into the loop.

AI plays a key role in that pipeline, but not as an autonomous scientist.

Instead, the technology is part of a broader infrastructure where software, robots, chemical rules, measurement equipment, error correction, and human expertise come together.

6. Why the physical world matters for AI

Experiments force AI to confront reality. A language model can produce a plausible answer without the claim ever being physically tested. In a lab, that freedom is far more limited.

A chemical reaction either works or it doesn’t. A predicted compound either forms or it doesn’t. Equipment clogs, feedstocks vary, and a theoretically elegant reaction can fail in practice.

That’s why Cronin warns against the notion that scientists can simply wire up AI, robots, and instruments and expect fully autonomous science.

“People think you can just unleash robots, automation, and a closed feedback loop on a problem and it solves it. That’s not true.”

According to Cronin, such a system needs an explicit structure to describe actions, experiments, and errors. His chemical automation work is about building exactly that programmable base layer.

The same distinction matters beyond chemistry.

AI can fire off answers at high speed. Science also demands testing predictions, explaining deviations, and forming new theories when reality breaks the existing model.

7. Could AI ever become conscious or truly creative?

Cronin sees no clear reason to believe that simply scaling today’s AI models will somehow yield consciousness or human creativity.

His argument is tightly linked to his research on life and evolution. Human intelligence, he notes, emerged within living systems shaped by billions of years of natural selection.

Survival sits at the core of that story.

“Evolution demands we do one thing at all costs: survive.”

Cronin suspects traits like intelligence and agency are tightly bound to that evolutionary history. Organisms must constantly solve problems because their survival depends on it.

A language model doesn’t face that biological pressure.

That doesn’t mean creativity is supernatural. Cronin calls himself a materialist. In his view, human intelligence ultimately arises from physical processes.

The question is whether the same qualities can emerge via today’s path of ever-larger, data-trained AI models.

Cronin is extraordinarily skeptical.

In the interview, he even says he’s “99.9999 percent” certain that silicon-based AI, using the current approach, will never be creative “in the true sense of the word.” At the same time, he admits he can’t rule it out with absolute certainty.

That nuance matters: this is Cronin’s prediction, not a scientifically proven limit of artificial intelligence.

8. Critical of AI hype, not of AI itself

Despite his sharp criticism, Cronin sees AI as a technology humans must work with closely. He rejects both the extreme optimism around AI and the opposite scenario in which intelligent machines inevitably replace humans.

“We shouldn’t miss the chance to evolve alongside AI.”

That makes his stance less anti-AI than his soundbites might suggest.

Cronin actually expects technology and people to become increasingly interdependent. AI can help researchers explore larger search spaces, design experiments, and apply existing methods faster. Robots can then automate more and more physical lab work.

But in his view, human researchers won’t vanish from that process.

Their role will shift.

9. What this means for jobs and scientists

AI can automate tasks without eliminating entire professions. In the interview, Cronin explicitly pushes back on the idea that workers will mainly be left to “check” AI’s outputs.

In software development, for instance, AI may generate more code. The human job isn’t just to review lines of code, but to redesign systems, understand problems, and determine the right architecture.

The same goes for scientists.

Chemists may run fewer manual experiments. Researchers can analyze literature faster, and models can test thousands of hypotheses or candidates before a person ever could.

That shifts the researcher’s value to asking better questions, defining problems, making sense of surprises, and formulating new explanations.

Cronin’s own company underscores his stance. While tech firms often pitch AI as a way to cut headcount, Cronin said Chemify has actually hired hundreds in recent years. He says the company now employs about 230 people.

In his view, automation and human labor don’t have to be at odds.

10. The AI debate is ultimately about us

Cronin’s critique points to a deeper issue than the latest chatbot’s performance. The AI industry is building systems that increasingly mimic behaviors we associate with intelligence, even as there’s no broad agreement on what intelligence, consciousness, and creativity actually are.

That gap fuels both hype and fear.

An AI-generated video can be indistinguishable from reality without the model understanding what it depicts. A chatbot can produce a scientific-sounding hypothesis without proving the system is doing science. A model can propose a novel molecule without knowing whether it can ever be made.

That doesn’t mean such systems lack value. Far from it.

Cronin’s point is that capability is not understanding, generation is not creativity, and automation is not scientific autonomy.

Whether that line is permanent remains an open question. His prediction that today’s AI will never be truly creative is very much up for debate, not a settled scientific fact.

But precisely because Cronin uses AI, robotics, and automation to push the boundaries of chemical research, his critique offers a compelling counterpoint in the current AI debate.

He’s not warning that AI can’t do anything.

He’s warning that we’re too quick to draw conclusions about what AI is from what it can do.