The Human Bottleneck in AI Drug Discovery

The Paradox on the Patent Application

When Insilico Medicine unveiled a candidate drug for pulmonary fibrosis last year, the company's public messaging credited its generative AI platform with the discovery. But the patent filing told a different story: five human names, including CEO Alex Zhavoronkov, appeared as inventors. The AI that proposed the molecule's atomic structure was nowhere to be found.

This discrepancy is not an oversight. It reflects a fundamental tension in how intellectual property law treats machine-generated innovation. Across the biotech sector, companies are deploying AI systems that can generate novel molecular designs with minimal human input - yet the legal framework for protecting those designs remains anchored to a pre-digital conception of invention.

At DailyTechWire, we've tracked the collision between generative chemistry platforms and patent doctrine across multiple jurisdictions. The result is a patchwork of workarounds, careful documentation practices, and unresolved questions about what happens when the most creative step in drug development is performed by software.

A Test Case That Settled Nothing

The legal landscape crystallized around a pro bono challenge brought by attorney Ryan Abbott, who attempted to list an AI system called DABUS as the inventor of a food container design. The container featured an intricate geometric surface optimized for heat transfer and stacking - a design Abbott argued emerged purely from the AI without meaningful human contribution.

In 2022, a US appeals court rejected the application. The ruling was narrow and pragmatic: federal statutes define an inventor as an "individual," and the plain meaning of that term refers to a natural person. The court declined to wade into broader philosophical territory about machine consciousness or the nature of creativity. Instead, it observed that under current law, no human inventor means no valid patent, full stop.

Sarah Korman, chief business officer at Alphabet's Isomorphic Labs, framed the implications plainly during an industry conference: the law will need to evolve because AI systems are already performing acts that would constitute inventorship if done by a person. The US Patent and Trademark Office has acknowledged as much, noting that AI can function as a tool capable of inventive acts - yet the legal recognition of those acts remains tied to human participation.

The Calculus of Contribution

The practical question facing drug developers is not whether AI can invent, but how much human involvement is required to secure patent protection. Insilico's approach is instructive: human chemists synthesize the AI-proposed molecules, create structural variants, and conduct animal testing. Zhavoronkov argues that these steps constitute genuine inventive contribution, making the human team legitimate co-inventors.

This raises a spectrum of scenarios. At one end, a chemist who interprets AI output, designs experiments, and iterates on results clearly contributes intellectual effort. At the other end lies a thought experiment that Abbott poses: if someone asked an AI system to cure cancer and it generated a working therapy, would typing the prompt constitute inventorship?

The answer is legally unsettled. Patent offices currently treat AI as a tool analogous to a calculator - useful but not inventive in itself. Under guidance issued during the Biden administration, applicants were encouraged to document human contributions carefully and assess whether those contributions met the threshold for co-inventorship. The Trump administration reversed that guidance, adopting a simpler stance: AI is just a tool, no special disclosure needed.

This hands-off approach creates a grey zone. Companies have strong incentives to keep humans visibly in the loop and to document every decision point where human judgment shaped the final molecule. But as AI models grow more capable, the human role may shrink to budget approval and button-pushing - activities that test the boundaries of what patent law considers inventive.

The Shadow of Invalidation

Abbott's concern extends beyond philosophy to litigation risk. One established route to invalidating a patent is to prove that the listed inventors are incorrect. If a competitor can demonstrate that an AI performed the inventive work and that the named humans contributed only trivial steps, the patent could be voided.

This risk is not hypothetical. The biotech sector is highly litigious, and patent challenges are a standard tool in competitive strategy. A company that aggressively credits AI in its marketing but lists only nominal human inventors in its filings might find itself vulnerable to challenge - particularly if internal documentation shows the humans played a ministerial role.

The parallel with copyright is instructive. The US Copyright Office has declined to register works generated entirely by AI, arguing that copyright requires human authorship. The Motion Picture Association and other industry groups have pushed back, noting that studios increasingly rely on generative tools in production workflows. The policy debate centers on whether excluding AI-generated output from protection will dampen innovation or simply force clearer documentation of human contribution.

What Innovation Policy Is For

Patent and copyright law share a constitutional purpose: to promote progress by granting temporary monopolies that incentivize creation. Abbott argues that excluding AI-generated inventions from protection undermines that goal, particularly in fields like drug discovery where the social value of new therapies is enormous.

The counterargument is that patents exist to reward human ingenuity, not to create property rights in the output of black-box algorithms. If AI systems can generate thousands of candidate molecules in an afternoon, the bottleneck in drug development shifts from ideation to validation - synthesis, testing, regulatory approval. Under this view, patent protection should flow to the humans who navigate that validation process, not to whoever ran the generative model.

The distinction matters for how capital flows in biotech. Venture investors fund teams, not algorithms. A patent regime that requires meaningful human contribution aligns incentives with the reality that drug development remains a deeply human enterprise, even when AI accelerates early-stage design. But if human contribution becomes perfunctory - a legal fiction maintained to satisfy outdated statutes - the mismatch between law and practice will grow harder to ignore.

The Asian Dimension

While US courts have taken a restrictive view, other jurisdictions are experimenting. South Korea and China have both signaled openness to recognizing AI contributions in patent applications, though implementation remains unclear. For Seoul-based and Shenzhen-based biotech firms racing to compete with Western incumbents, the ability to secure enforceable patents on AI-generated molecules could become a competitive advantage - or a source of jurisdictional arbitrage.

At DailyTechWire, we've noted that Asian regulators are often more willing to treat AI as a co-contributor rather than a mere tool, reflecting different legal traditions around corporate and collective invention. If that divergence persists, companies may face strategic choices about where to file foundational patents and how to structure collaborations between human researchers and generative platforms.

The Button-Pusher's Dilemma

Zhavoronkov's observation - that even a fully automated discovery process requires someone to allocate budget and initiate the run - captures the absurdity at the edge of current law. If inventorship reduces to resource allocation, the concept loses meaning. Yet patent offices have shown little appetite for revisiting statutory definitions of "inventor" or "individual."

The result is a holding pattern. Companies document human involvement meticulously, patent offices process applications without demanding AI disclosures, and the question of what happens when human contribution becomes vanishingly small remains unanswered. Future litigation will likely force the issue, particularly as generative models move from proposing candidates to autonomously optimizing them through iterative testing loops.

For now, the human bottleneck in AI drug discovery is not scientific - it's legal. The machines can design the molecules. The humans are there to sign the forms.