IP and AI, Two Incompatible Acronyms?
Written by Jacob Forward
I was privileged to take part in the E-Lab residential course at King’s last September. During the week it became clear that a common theme among my fellow participants was an interest in weaving AI into our business designs. The next logical step for a tech-oriented start-up would seem to be securing Intellectual Property (IP), to reap the lucrative rewards of a temporary monopoly on a particular idea. However, the nuances of IP are hard to align with the messy realities of software development. This disconnect is amplified in the context of artificial intelligence. Is it possible that these two formidable acronyms are, practically speaking, incompatible? And does it even matter that it’s hard for AI-based start-ups to weave the protective cloak of intellectual property around their creations? Fortunately for us, E-Lab members received a visit from an expert from Basck Intellectual Property Services, to help us with these questions.
The Elusive Nature of Software Patenting
During the workshop many of the difficulties involved in patenting software were highlighted. While the software you write is automatically afforded some protections under copyright laws, as your own creative output, this offers very little real protection. Those seeking more security from a patent often encounter challenges to protecting their software, particularly in the UK. For instance, it is crucial, in your patent application, to demonstrate the technical effect of your invention. This is notoriously difficult to evidence concretely in software. The incremental, team-based nature of programming also means that traces of other projects often percolate into new creations, and isolating any original code stretches the limits of practicality.
Many businesses, however, have successfully applied for patent protection for the code they develop, and this is invariably key to their scaling and financial success. Basck’s expert suggested that it might be possible to circumvent the challenges of gaining a software patent in the UK by applying for a patent in the US first, and then some other countries, to improve the odds of eventually gaining a UK patent. These tips aside, software remains considerably more difficult to protect than tangible inventions. In AI it is even more complex.
The Particular Challenges of AI
AI models have a highly iterative development process. Models trained on open datasets via algorithms crafted by multiple engineers over months morph into emergent behaviours beyond individual contributions. This development process can make isolating the “inventive step” and the “novelty” necessary to pass the patent office’s non-obviousness test increasingly difficult. AI models also tend to draw heavily on open-source resources, whether code or datasets, and this further blurs the boundaries around IP. In short, the traditional understanding of patenting, which has been focused on static, tangible inventions for hundreds of years, struggles to keep up with the fluidity of AI systems. Algorithms that continuously evolve, retrain, and learn from vast datasets make it hard to pin down a static creation worthy of protection. The collaborative nature of the AI community, which is such a positive feature for innovators and start-ups, poses a serious challenge to IP protection, be that patents, copyright or other forms of IP, by continually redrawing the lines between original creation and derivative work.
To IP or not to IP?
The question of how to approach this unique character of AI innovation is where the expert’s opinion and mine may diverge. A good IP lawyer will go to pains to point out the indispensable importance of protecting your intellectual property. Owning IP can significantly improve your chances of securing investment, it is a valuable asset on the books of the company that can later be sold, and crucially, it gives you a brief but lucrative monopoly over an in-demand commodity or service. Moreover, major patent authorities, the UKIPO and EPO, have recently revised their stances on AI, recognising it as a 'technical effect.' This shift has lowered the barriers to gaining patent protection for AI-enabled start-ups.
While these points are all valid, I would suggest that there may be additional arguments we need to consider. I’m not convinced that owning IP fully ameliorates imitation risk. A large company can still copy your invention, safe in the knowledge that it will be prohibitively expensive for your start-up to defend its IP in court. Even if you have the foresight and the budget to purchase IP insurance, which gives you access to a hefty legal defence fund to take on challenges to your IP, the amount pales in comparison to the R&D expenditure of some large companies. If your idea is good enough, these companies would surely out-spend you in court with ease. Perhaps it is overly pessimistic of me to imply that the richest always win in legal disputes, particularly in light of the recent Massimo vs Apple case which demonstrates that smaller entities can win against tech giants in patent cases. Nevertheless, I’d suggest that being granted IP rights is not necessarily the protective panacea we may believe it to be.
Reconsidering the first-mover advantage
We should consider other potential alternatives well suited to the unique character of AI. The first-mover advantage, in which a company gains a competitive advantage by being the very first to bring a new product or service to market, can be quite lucrative on its own. Establishing your AI-based solution as the industry standard before copycats emerge allows you to sustain market leadership without major legal artillery, not indefinitely, but for long enough to turn a decent profit. I believe this is especially the case if you are serving a niche market which holds limited appeal to deep-pocketed tech giants. Indeed, the very size of the industry giants can be their Achilles-heel, since they aren’t as nimble or unencumbered by bureaucracy as start-ups, they cannot easily chase small markets and keep pace with the rapid tempo of development in the field of AI. The start-up’s advantages of agility, adaptability, and a willingness to take risks are ideally suited to the AI sector at the moment. I would add the caveat that these arguments apply best to an AI-based company whose service piggybacks on a model from a larger company, or from the open-source community, since developing AI models from scratch remains too computationally and technically resource intensive for most start-ups.
In an ideal world, entrepreneurs would combine robust IP protection with the first-mover advantage (and every other advantage they can possibly gain). While there are complexities and challenges specific to gaining IP protection for AI inventions, recent legislative advances have improved the odds, and even if there’s nothing patentable in your use of AI, the first mover advantage can be impactful on its own. If the big players like OpenAI, Anthropic, Google, Meta and others are dividing up the cake between themselves, then their gorging has led to a vast abundance of crumbs, and these crumbs of market share, in niche markets and specialised applications of AI, can still be worth millions. I would therefore encourage any of my fellow E-Lab members interested in harnessing the power of AI in their start-up to be as suave and swashbuckling as our chum Laurence on Survivor!
With thanks to Basck Intellectual Property Service for feedback and advice during the writing of this article.
Jacob Forward
Jacob is an E-Lab member and PhD student in the Faculty of History. He previously read for an MPhil in American History at Cambridge (King's College), and a BA in History at Oxford (Keble College). He has worked for History and Policy at the Institute for Historical Research and consulted on research projects at the School of Advanced Study.