Will the UK emerge victorious in AI

Since the conquering of natural language by artificial neural networks, and the public demonstration of that capability with the launch of ChatGPT, AI development has increasingly been framed as a two-horse race between two great powers, embodying two very different worldviews and, as we have come to understand, pursuing two very different approaches to AI.

In the US, we are seeing the ultimate expression of the Silicon Valley playbook: an unregulated bonfire of investor cash with tech bros clamouring to capture market share for downstream exploitation. Competition for talent is fierce, with know-how and trade secrets flowing between top labs in a manner that limits the possibility of meaningful IP moats, while entrenching an intellectual monoculture around transformer-based large language models (LLMs). Under those conditions, scale and capital become the key differentiators, along with access to data, which the generative AI industry is emboldened to grab by every means possible. The executive branch supports this maximalist approach while seeking to control competition abroad through export controls on hardware.

The most successful companies in the sector will develop and deploy sophisticated IP strategies

This modus operandi is codified in America’s AI Action plan, whose core strategy can be summed up by a single word: dominance.

Meanwhile, China plays for disruption, with the CCP actively encouraging the development and export of open models in an effort to hobble American dominance over the technology stack, while maintaining tight control over domestic AI rollout and supporting development in sectors it has targeted for growth. The Wild West finds its counterbalance in the East.

So what does it mean for the UK, and other “middle powers”? With a few notable exceptions, the capital-intensive nature of training LLMs means there is little appetite to do so outside of the US-China dipole. Does this need to change if Britain is to realise its ambition of being a world leader in AI? Not necessarily. While the prevailing view is to attribute most of the value of the AI industry to the model providers, it is increasingly obvious that LLMs are becoming commoditised, and that the capabilities of open models lag their closed counterparts only by a few months. Aside from the observation that many AI stocks are almost certainly overvalued on this basis, it also presents a serious challenge for the major LLM providers, who will sooner or later find themselves having to monetise their users in service of their investors. If the viability of open model solutions prevents them from doing so via traditional means (subscriptions, ads), the model providers will increasingly look for profits in the application layer, and in particular the adaptation and incorporation of AI models into enterprise-ready products, either by offering such products themselves or by demanding a revenue share from companies that do so based on their models.

This is where IP plays an invaluable role. There is a huge scope for innovation that is both highly valuable and potentially protectable in applying AI to specific domains, and/or in addressing the many problems posed by generative AI regarding reliability, security, and trust. These solutions are not typically capital intensive, but rely on one or more of domain expertise, inventiveness, and access to high-quality proprietary data. As the new-toy dazzle of generative AI wears off and makes way for the need for products that work in the real world, this is where value will primarily reside for the UK’s AI sector, already the third most valuable in the world and with every chance of maintaining and bolstering that position. The most successful companies in the sector will develop and deploy sophisticated IP strategies that jointly leverage the benefits of open source, patents, and trade secrets, taking into account the divergence of patent eligibility standards across jurisdictions including the UK and US, to lay claim to their chosen part of the AI stack in their chosen markets, while enabling them to fend off, collaborate, or bargain with other parties as necessary. To the extent that LLMs are involved, the most prudent strategies are likely to be model-agnostic, tending towards open solutions to prevent risks associated with vendor lock-in and to avoid the rent-seekers premium of closed providers.

The government has a key role to play in facilitating this success, through public sector procurement of AI solutions, galvanising the conditions for successful companies to remain in the UK, and by introducing targeted regulation which supports innovation over extraction while protecting citizens from the worst excesses of the industry. Getting it right would cement the UK’s position as an island of trust and lay the ground for our domestic AI industry to thrive. Special attention must be given to our world leading creative and media sectors, which along with our public sector bodies sit on some of the most valuable data in the world, while providing the cultural capital and identity that make Britain a great place to live. Along with the UK’s strengths in research and our trusted legal system, these assets make Britain a natural place to lead on trustworthy, responsible, and sustainable AI.