From cat pics to quantum breakthroughs – how generative AI is inventing new matter

MIT’s SCIGEN lets generative AI design new materials that could transform technology, energy, and electronics.

AI already has a reputation for churn. Whether it's memes, songs, or videos, the synthetic current has been flowing for a few years now, to the point that it doesn’t really feel like it dazzles anymore.

But what if, as well as creating the artistically sublime, it were able to spin towards producing transformative materials?

When I say transformative, I don’t mean a material transforming into another. I’m talking about a game-changing matter.

"We wanted AI to stop playing it safe and start aiming for materials that could actually change the world," says Ryotaro Okabe, a PhD student at the Massachusetts Institute of Technology (MIT), who worked on the project.

SCIGEN, the MIT system, directs AI models to build structures with exotic quantum properties.

"It’s like giving the AI a rulebook of what’s actually interesting to scientists," says Mingda Li, the lead researcher.

Here, the word “material” needn’t be used in a mere academic context. We’re talking brand new materials – think an AI-designed material that could make your smartphone charge in minutes rather than hours.

The problem with materials discovery

As you might recall from history lessons, the discovery and refinement of a material can literally take an age, like with stone, bronze, and iron.

Even in modern history, a synthetic polymer like nylon took over a decade of experimentation from inception to commercialization, happening between 1927–1938.

AI has no problem churning out possibilities of new compound materials to find synthetic novelties. However, so far, there’s been no vanguard.

"Even with AI, you get millions of candidates, but most are just… safe," says Mouyang Cheng, a co-author of the study.

As AI generates sheer numbers at breakneck speed, scientists in the labs often struggle to keep up.

"We needed a way to tell AI, ‘Don’t just guess. Follow the shapes that matter,’" Li adds.

How SCIGEN changes the game

As superconductivity and quantum computing could be the next disruptive industries on a grand scale, harnessing AI’s potential has become the real play.

Quantum behaviours have strict parameters, known as Kagome or Archimedean lattices (like repeating triangles or polygons).

So instead of letting AI run wild, SCIGEN narrows the design spec so that AI can flourish properly.

"These patterns are rare, but they’re exactly what we need to unlock new physics," commented Cheng.

Initial testing has produced two curious compounds with unfamiliar magnetic behaviour, showing high potential for the team’s efforts.

"It’s exciting to see the AI’s predictions line up with what we can actually make in the lab," Okabe said.

Why it matters for tech

Currently, it remains unclear if and when quantum computing will make its mark. Forget the talk of dystopian progress; the predominant conversation lies in a design framework.

"If we get the right lattice, the whole field could accelerate overnight," says Weiwei Xie of Michigan State University.

AI could also help design materials for better batteries, superconductors, or carbon capture systems.

Dependence on rare-earth elements and preventing pandemic-level bottlenecks in the supply chain could be mitigated with a kind of AI intervention.

A rapid breakthrough could give us something like a new energy grid, and researchers should welcome this prospect.

"We’re not replacing scientists. We’re giving them a supercharged assistant," says Li.

Risks and reality check

Generating thousands of no-go options remains the biggest obstacle, as production and testing on all the ideas is unrealistic.

And the matter of ownership could also come into play, as the patent of a groundbreaking new compound may not be as simple as Wallace H. Carothers and his team at DuPont for nylon in 1937.

Despite this, SCIGEN shows AI moving from a tool to a co-inventor.

"The ratio of stable materials goes down, but the number of exciting possibilities goes way up," Li says.