What’s the Value of Scientific Superintelligence?

Crossing the Atlantic by ship used to take the better part of a week. Propeller planes cut that trip to under 20 hours, then the jet engine cut it to under seven. For both people and goods, an ocean crossing transformed from an event to an errand.

Something similar happened with communications. Before the telegraph, orders for goods were sent as letters back and forth on ships. When a transatlantic cable linked continents in 1866, that same exchange collapsed into minutes: a trader in New York could confirm a shipment with a supplier in Liverpool before either had finished their morning coffee.

Neither breakthrough set out to reshape commerce; each just made something new possible, and the world reorganized around it. That's what scientific advancement does, sometimes loudly, often quietly, again and again. It is the engine behind nearly every leap in how we live.

That engine ran on government funding and direction for decades before broadening to major industrial labs, such as Bell Labs, then to a wide field of corporate R&D. In each place, the scientific method does the slow, expensive work of turning ideas into things that help people. It's a good, reliable engine, like an old V8, but long overdue for an upgrade.

At Lila Sciences, we believe that upgrade is scientific superintelligence. In the same way cloud computing fundamentally reshaped data storage, scientific superintelligence, if achieved, will reshape the scientific method to achieve unprecedented speed, scale, and impact. Imagine a future where any organization can fire up a software platform that processes information above the level of human intelligence at every step of the scientific process: generating hypotheses, designing experiments, running physical labs, and iterating. That is what we are building at Lila Sciences.

But what does scientific superintelligence mean in practical outcomes for humanity? It means a new era of entrepreneurs can, for far lower capital, de-risk and develop new technologies that could have a meaningful impact on health, materials, and energy. It means corporations can dramatically accelerate their output and achieve more immediate outcomes with a higher probability of success, especially when translating discoveries downstream. It means national labs and federal agencies have a reliable partner for ambitious, long-horizon science and pressing national security challenges, such as pandemic response.

At Lila, we’re pouring our efforts into areas that matter. In health, only about one in ten medicines that enter a clinical trial ever reaches a patient, and each new drug costs one to three billion dollars to develop. Most of that risk gets locked in early, in choosing which disease target to chase and which molecule to build, out of a space of chemical possibility no human mind(s) can fully reason through on their own. That's exactly the kind of problem scientific superintelligence is built for: a system that learns from hundreds of thousands of real lab experiments (not just published papers), can search the full chemical space, and raise the odds of identifying a molecule that will be successful. In a recent in vivo CAR-T program, Lila’s approach produced a therapy that lasted longer in non-human primates than existing best-in-class treatments, at an estimated tenth of the usual R&D cost.

The same logic holds in materials and energy. The pandemic, and conflicts like the war in Iran, showed us how fragile global supply chains are. Nearly every one of those sources of fragility comes down to materials: what something is made of and where you can get it. The number of ways to combine elements on the periodic table into materials is nearly infinite, , which is, again, exactly the kind of search a system that gets smarter with every experiment can take on. That's why we're hunting for rare-earth-free magnets and new catalysts for some of today's tightest bottlenecks, such as those behind electric vehicles and satellites. We’re also identifying and testing superconducting materials that could prevent electricity loss to resistance, with ripple effects from clean power to medical imaging.

And not every opportunity starts with discovering something new. Some of the most interesting problems are in formulation, in how to combine materials that already exist on an industrial scale in unique ways. For example, can we formulate better battery electrolytes, lubricants, or controlled-release fertilizers? The chemistry is known, but with scientific superintelligence, we aim to improve on the combinations, ratios, and processes for a given formulation.

When people talk about automated science, they often mention benefits such as increased speed and reduced cost. Those matter, but they're not new—the industry has chased faster, cheaper experiments for years—and they’re not the hard part. The hard part is having a higher probability of success. That's what scientific superintelligence is for, not just making discoveries but optimizing them, learning from every experiment so the next one is smarter than the last.

I am not under the illusion that hard problems in R&D will go away tomorrow, or even in the coming years. A molecule that works beautifully in the lab still has to become a therapy that reaches a patient, and that gap remains real and challenging. But I do believe that by upgrading the scientific method with scientific superintelligence, we will raise the odds of success, one experiment after another.