Can AI invent? Patents, personhood and the DABUS problem

The DABUS litigation became a cultural flashpoint because it allowed people to ask a deceptively simple question: if an artificial intelligence system helps generate something novel, why should it not be named as the inventor?

The question sounds modern. The answer, at least under current legal frameworks, is surprisingly old-fashioned.

Patent systems were not built only to identify novelty. They were built to allocate rights and responsibilities within a human legal order. Inventorship is not just a factual label attached to a useful output. It is a legal concept connected to ownership, reward, disclosure, attribution and, ultimately, the social bargain that patents represent.

That matters because a patent does two things at once. It grants a temporary monopoly, and it asks the inventor to disclose enough information for society to learn from the invention once the exclusivity ends. The system is therefore not just rewarding technical achievement in the abstract. It is directing incentives toward human actors who are presumed to care about recognition, commercial value and future innovation.

An AI system does not fit neatly inside that structure.

Even if a model or machine produces a result that appears surprising, useful or commercially significant, it does not follow that the law should treat the system itself as the inventor. The immediate difficulty is doctrinal. Patent legislation, judicial reasoning and long-standing assumptions across multiple jurisdictions treat inventorship as a human role. Courts have repeatedly focused on ordinary statutory language, legislative history and the surrounding structure of rights to conclude that inventors must be natural persons.

But the deeper issue is conceptual.

When people argue for machine inventorship, they often collapse two different questions into one. The first is whether AI systems can materially contribute to the generation of new ideas. The answer is clearly yes. The second is whether the law should interpret that contribution as inventorship in the same sense it attributes inventive activity to people. That is a different proposition entirely.

Inventorship is not just about causation.

If it were, modern innovation systems would already be much messier. We would have to ask whether the real inventor was the lab technician, the dataset compiler, the tool maker, the investor who funded development, the prompt author, the model trainer, or the organisation that provided the infrastructure. In high-complexity environments, lots of actors contribute to an outcome. The law narrows that field not because the others are irrelevant, but because legal systems need a stable and administrable account of who did the inventive work.

AI complicates that account by generating outputs through statistical and computational processes that are powerful but not self-interested. The system does not seek monopoly rights. It does not value attribution. It does not experience reward. And it does not participate in the social compromise that the patent framework is meant to mediate.

This is why analogies to human creativity can mislead.

People sometimes describe advanced systems in anthropomorphic terms. A model “discovers” an idea. A machine “comes up with” a solution. A system “creates” something new. Those phrases may be descriptively convenient, but they can smuggle in normative conclusions. The fact that a machine output is impressive does not tell us who the law should recognise, compensate or burden.

There is also a policy risk in moving too quickly from capability to legal personification. If inventorship were assigned directly to machines, what problem would that solve? In most commercial settings, human stakeholders already exist around the system. Developers, deployers, owners, operators and researchers all have plausible claims to the economic value generated through the tool. Calling the machine the inventor does not eliminate those competing claims. It simply inserts a non-human actor into a regime that still requires humans or corporations to hold, manage and enforce the resulting rights.

The more useful question is therefore not whether AI should receive the title of inventor. It is how law should allocate value when AI-assisted processes materially contribute to invention.

That conversation will likely turn on contribution, control and disclosure. Who set the problem? Who selected the inputs? Who understood the result well enough to recognise its significance? Who could explain the path from experimentation to claimed invention? These are messy questions, but they are at least questions the law knows how to ask.

There is also a philosophical tension worth acknowledging. Patent law partly reflects a belief that creativity has value because of the process that produced it, not just because of the output left behind. If invention is reframed as a brute-force optimisation exercise, generated by iterating through enormous solution spaces until something viable emerges, then the meaning of inventive step itself starts to feel unstable. That pressure applies to AI, but it also returns to humans. If invention becomes a numbers game, what exactly is being rewarded?

For now, courts have been reluctant to remake the system around that uncertainty, and I think that caution is sensible.

AI will keep generating commercially valuable outputs. Some will influence product development, scientific discovery and design in ways that feel genuinely novel. But legal systems do not need to pretend that machines are inventors in order to respond to that reality. A better approach is to preserve human accountability while refining the rules around ownership, contribution and disclosure in AI-assisted environments.

The law’s resistance to machine inventorship is not just conservatism. It is a reminder that intellectual property does more than catalogue novelty. It decides who the system is for.