BioTech IP Review

Artificial intelligence is increasingly used to generate candidate drug molecules, proteins, and other biologically active compounds with minimal direct human design. In some cases, machine learning systems propose chemical structures or amino acid sequences that no human explicitly conceived in advance. This raises a fundamental question for patent law: when an AI system designs a molecule, who—if anyone—qualifies as the inventor?

The answer turns on how courts interpret the doctrine of conception in an era where creativity may be computational rather than cognitive.

The Legal Standard for Inventorship

Under U.S. patent law, an inventor is one who contributes to the conception of the claimed invention. Conception is defined as the formation, in the mind of the inventor, of a definite and permanent idea of the complete and operative invention, as it is to be applied in practice.

This definition presumes a human mental act. Current statutory and judicial frameworks do not recognize non-human entities as inventors. Recent cases involving AI systems named as inventors have reaffirmed that inventorship is limited to natural persons.

However, AI-assisted molecule design strains this framework. When a model generates a chemical structure satisfying predefined criteria—binding affinity, solubility, or toxicity thresholds—the human may not have envisioned the resulting structure. Instead, the human defines the objective function, and the machine performs the combinatorial search.

The question becomes whether defining the problem constitutes conception of the solution.

AI as Tool Versus AI as Originator

Inventorship disputes will likely turn on whether the AI system is characterized as a tool or as a creative agent.

In traditional drug discovery, computational methods such as molecular docking and QSAR modeling assist human chemists, who still select candidate structures and rationalize their features. In such cases, the human remains the inventor.

By contrast, modern generative models can autonomously propose novel molecules optimized for specified parameters. If the human merely inputs a target and constraints, and the AI outputs a structure never previously contemplated, the causal chain between human thought and molecular structure becomes attenuated.

Courts will likely analogize to existing jurisprudence on automation. Tools that assist but do not replace human conception do not displace inventorship. But where a machine determines the key inventive features, assigning inventorship to a human may become legally strained.

Defining the Inventive Contribution

To resolve inventorship in AI-designed molecules, courts may need to redefine what constitutes the inventive act. Possible loci of invention include:

1. The problem definition – specifying the desired molecular function or performance criteria.
2. The algorithm design – creating the model architecture and training regime.
3. The molecule selection – choosing which AI-generated output to pursue experimentally.
4. The molecule itself – the specific chemical structure or protein sequence.

If inventorship attaches to problem definition, then a researcher who defines a biological target could claim inventorship of any molecule generated against that target. This risks collapsing inventorship into mere research direction.

If inventorship attaches to algorithm design, then software developers may be inventors of downstream chemical compounds, even where they had no biological intent.

If inventorship attaches to molecule selection, then screening becomes conception, potentially converting routine filtering into inventive activity.

None of these mappings cleanly aligns with traditional doctrine, which assumes that the inventor mentally conceives the claimed structure.

Written Description and Enablement Implications

Inventorship disputes will also intersect with written description requirements. To claim a molecule, the specification must demonstrate possession of the invention. Where a molecule is generated by an opaque neural network, the applicant may struggle to explain why the molecule has the claimed properties or how it was conceived beyond stating that it was selected by a model.

This raises a subtle risk: if the applicant cannot articulate the inventive rationale behind the molecule, challengers may argue that no human ever truly conceived it. The patent system may be forced to confront whether algorithmic generation without explanatory insight satisfies the philosophical basis of patent rights.

Comparison to Natural Product Doctrine

AI-designed molecules invert traditional patent eligibility problems. Natural products are excluded because they are discovered rather than invented. AI molecules may be excluded, paradoxically, because they are too invented—generated without human creative intervention.

Where a molecule has no natural analog and arises solely from computational search, it may be patent-eligible as a composition of matter. Yet inventorship remains problematic if no human can claim mental conception of the structure.

This creates a conceptual gap: an invention may exist without an inventor.

Likely Near-Term Outcomes

In the near term, patent applicants will likely assign inventorship to humans who:

• Designed the AI system,
• Defined the optimization criteria, and
• Selected and validated the molecule experimentally.

This strategy aligns with current legal doctrine, even if it stretches the notion of conception. Patent examiners and courts are likely to tolerate this approach initially, favoring administrability over theoretical purity.

However, as AI systems become more autonomous and capable of iterative self-improvement, this fiction may become harder to sustain. If an AI proposes molecules based on training data without human-curated objectives, the human role may become too remote to justify inventorship.

Policy Implications

Recognizing AI as an inventor would require statutory reform and raises profound policy questions. Granting patents without human inventors could concentrate ownership in entities controlling AI systems, potentially reducing transparency and accountability.

Conversely, denying patent protection to AI-designed molecules could disincentivize investment in algorithmic drug discovery, pushing such inventions into trade secrecy.

The legal system must therefore balance three competing interests: preserving human-centered inventorship doctrine, incentivizing innovation, and maintaining coherent standards of contribution.

Conclusion

AI-designed molecules challenge the foundational assumption that invention is a cognitive act. As algorithms increasingly determine molecular structures, inventorship doctrine will be forced to evolve from a theory of mental conception to a theory of system authorship. This emerging conflict foreshadows a shift from disputes over who invented a molecule to disputes over who controlled the machine that generated it. The resolution of this issue will not only determine patent ownership but may redefine the meaning of invention itself in molecular science.