An ARIA for AI? The innovative ideas from the House of Lords that may shape tech’s new rules

The first global  AI Safety Summit held in the UK drew to a close yesterday and thoughts now turn to what new regulations may look like.

The UK is grappling with the balance between fostering AI innovation, while at the same time reducing the risks it poses, including safeguarding individual rights.

In part one of a two-part series, we heard how members of the House of Lords wanted the government to move faster in implementing new protection for rightsholders.

In this article, we hear the house’s AI activists weigh in on the future of AI regulation, with a consensus forming that favours the adoption of 'risk-based' regulation and the need for global cooperation on rules.

‘Risk-based’ regulation

Lord Clement-Jones, full name Timothy Clement-Jones, advocates for risk-based regulation, where the degree of regulation depends on the risk posed by a particular AI.

Additionally, he wants to see “ethical principles embedded in standards in terms of testing, risk assessment and management.”

He also wants the “protection of human creativity at the front”.

“It's all very well to say we mustn't impede innovation, but if innovation basically destroys the value in human creativity, that doesn't seem to be a very sensible way forward.”

Clement-Jones, who is also a consultant at DLA Piper, stresses the need for AI to be treated as a tool for human benefit, with regulation serving to ensure that AI technologies serve society's best interests.

The Earl of Devon, Charles Courtenay, emphasises the significance of observing how “established legal remedies”, like copyright laws, are used to address concerns before deciding if further regulation is necessary.

“One of the strengths of our common law system is its ability to apply existing precedent to new scenarios, and I would encourage regulators and governments to take their time, remembering that hard facts can make bad law,” adds Courtenay.

Furthermore, Lord Ravensdale, full name Daniel Mosley, suggests that we may reach a point where AI will have the capability of creating new inventions, which would necessitate a substantial transformation of IP laws.

Courtenay is interested to see the extent to which IP offices around the world deal with the registration of AI-generated works—both in patent and in copyright.

There's a “fine balance to be struck” in terms of regulation because too much of a “heavy handed approach” will “stifle innovation,” says Mosley.

Mosley explains the government's White Paper, A pro-innovation approach to AI regulation, “struck a good balance” in terms of how to begin to regulate AI, but “the conversation has moved on from there in terms of the emerging risks of the technology.”

“The UK is wrestling with the balance between AI innovation and the protection of individual rights,” says Courtenay.

Time for an ARIA for AI?

A lack of comprehensive understanding of AI tech may hinder effective regulation, says Mosley.

This stretches to public dialogue, which Courtenay believes would benefit from education to engender a more nuanced analysis.

There is a need to embrace AI and harness its potential from a position of “informed confidence”, he says.

“Research and regulation go hand in hand here and they're different sides of the same coin,” agrees Mosley, who proposes that any initial action should be focused on establishing governance frameworks and creating a good understanding of AI, rather than simply considering regulations.

He emphasises the importance of setting up an agency for AI research and safety, similar to the Advanced Research and Invention Agency (ARIA), to gain a deep understanding of AI technology.

UK's approach vs international standards

When comparing AI regulation internationally, Mosley notes that the EU’s AI Act adopts a more stringent regulatory stance, while the UK's White Paper favours a lighter touch.

However, Mosley predicts that this is likely to change as AI research and our comprehension of it evolve.

Clement-Jones advocates for legislation that is more closely aligned with the EU, due to the need for trade with Europe, the desire to market products to European customers, and the necessity for compatibility with EU interoperability regulations.

“The EU started off on exactly the right basis, they had a pyramid of risk, they looked at what needed to be done in order to mitigate the risk and they came up with a form of regulation designed to do that,” says Clement-Jones.

EU regulation with regard to AI “demonstrates the right kind of flexibility” according to Clement-Jones.

Mosley highlights the importance of international collaboration, as pursuing separate national approaches is “not going to get us anywhere”.

“International collaboration is absolutely crucial here because if we're not joined up across countries you could get a situation where corporations will set up in the areas with the lightest touch regulation,” he adds.

Courtenay insists that jurisdictions collaborate “to ensure the collective protection of creative works and a level playing field of regulation and enforcement.”

Clement-Jones says he is “optimistic” that such collaboration will occur. Specifically through a convergence of AI standards at the international level, such as via organisations like the Organisation for Economic Co-operation and Development (OECD).

Breaking from the safety ‘straight-jacket’

Clement-Jones also expresses concerns about the limited scope of the AI Safety Summit.

Regulation can effectively come in place “if we manage to break out of the current straight jacket” of the safety summit’s agenda, he says.

“Characterising it as purely about safety seems to be misguided, it's about the ethical aspects, explainability, transparency, bias, and the impact on society,” he adds.

Clement-Jones wants to see proposals for regulations that are compatible with the EU, a full inquiry or review by the Intellectual Property Office on AI and IP, and the UK pushing convergence on standards as part of the OECD effort.

Courtenay warns “While we need to act swiftly in the regulation of AI, we must not regulate precipitously.

“Incorrect regulation that is rushed may be more damaging than no regulation at all. Policy needs to be informed and, if legislation is proposed, Parliament must educate itself as to the significance of AI and its implications,” he says.

Mosley provides three key recommendations for policymakers.

“What I would like to see first is for the government to commit to some form of agency,” for example setting up an AI research and safety branch.

The second focus would see a greater ambition around the UK’s computing capacity. Currently, the government has an aspiration to construct a supercomputer with ‘exascale’ levels of computing by 2026.

This is a type of ultra-powerful supercomputing, with systems performing billions of computations per second.

The government wants to have a compute resource available that would give the country the ability to train one GPT for scale, four years behind OpenAI’s release.

“We would already be behind the curve if that's our aspiration for 2026,” says Mosley. “I think expanding the ambition for that is needed in order for us to become leaders in AI.”

Finally, Mosley suggests placing the principles outlined in the White Paper on a statutory footing, requiring regulators to consider them more closely in industry practices.