2 February 2024
By Jakub Krupa
Companies developing generative artificial intelligence systems need more “steer” from regulators as their technology evolves, a report by a committee of UK lawmakers said today. The government should avoid any catastrophizing and focus on promoting innovation, evading regulatory capture, and resolving emerging competition and copyright problems, they said.
In a detailed 95-page report, Parliament’s Communications and Digital Committee expressed fears that if the UK does not adjust its regulatory approach, it could miss out on the opportunities brought by AI, particularly systems built on large-language models — algorithms based on huge data sets that allow AI applications to generate original content.
The lawmakers’ report said companies need better-equipped regulators and agile regulatory frameworks capable of dealing with practical problems here and now — from copyright to disinformation, cyber security, online safety and digital competition issues — without distractions provided by dire warnings about things going wrong further down the line.
Balancing act
The report highlighted a particular worry that the UK government was “narrowly focused on catastrophic risks” and that it “is not striking the right balance between innovation and risk,” with its attention “shifting too far towards a narrow view of high-stakes AI safety” and not enough on opportunities.
Focus on “catastrophic” outcomes — as outlined in a paper before the inaugural AI Safety Summit hosted in the UK last year — could make Britain miss more urgent issues and put it at risk of “falling behind international competitors and becoming strategically dependent on a small number of overseas tech firms,” mainly in the US and China.
It was “almost certain,” they said, that existential risks “will not manifest within three years and highly likely not within the next decade,” so the government should focus on getting the most out of AI safely and not restricting it by red tape and overly prescriptive compliance demands.
While praising different elements of the regulatory environments in the US, the EU and China, the committee said that “wholesale replication of their regulatory approaches appeared unwise” given that the UK “lacks the distinctive features that shape their positions — such as the EU’s customer base and appetite for regulatory heft; American market power; and China’s political objectives.”
The UK should try to pave its own way, seeking not to diverge too far from other partners but also not using fear of doing so as an excuse to delay, they said. Developments so far have been “slow,” with a principles-based policy paper published last March yet to be translated into any practical regulation. An update promised in December has not yet been published.
While the committee agreed that “extensive primary legislation aimed solely at large language models is not currently appropriate,” ministers should still focus on developing “an enforceable, pro-innovation” framework.
Separate worries were expressed about the “inadequate” delivery of the government’s “central function” intended to support key AI regulators such as the Information Commissioner’s Office, the Competition and Markets Authority and the Financial Conduct Authority.
“Relying on existing regulators to ensure good outcomes from AI will only work if they are properly resourced and empowered,” the lawmakers said. They called for “standardized powers” across sectors to help regulators and a sanctions regime to “provide a credible deterrent against egregious wrongdoing.”
Existing regulators had “significant variation in technical expertise,” with some having no AI governance specialists or funding in place.
Regulatory capture, competition
Lawmakers further warned that the industry was showing “mounting concern” about regulatory capture as a result of lobbying and a situation where “officials lack technical know-how and come to rely on a narrow pool of private-sector expertise to inform policy and standard.”
They also warned that overreliance on external AI expertise could give rise to conflicts of interest, calling for more transparency in how ministerial aides and sherpas for international negotiations are picked.
The committee called for administrative safeguards to ensure “decisions are subject to systematic challenge and review” to “mitigate the risks of inadvertent regulatory capture and groupthink.”
The committee also pointed to competition concerns given that businesses dominating the market for large-language models “will have unprecedented powers to shape access to information and commercial practices across the world.”
The lawmakers heard from experts that “the exploitation of first mover advantage among large developers could lead to entrenched market power,” with similar effects to search engines and social media platforms.
They recommended that the government make market competition in large language models “an explicit policy objective,” with “ensuring regulatory interventions do not stifle low-risk open-access model providers.”
They also urged ministers to work closely with the Competition and Markets Authority, building on its first foundation models review late last year.
Copyright
Unresolved questions around copyright regulations were highlighted in the report, with government-backed talks among market participants having struggled to get to an agreement on any voluntary code of practice.
The ambition is to find a way to reconcile the interests of rights holders with those of tech developers who warn it would be impossible to train models without copyrighted materials.
The committee said it was “disappointed” that the government could not pro-actively articulate its view on applying current laws in the AI context and rejected ministerial suggestions that it would be up to courts to interpret the existing provisions.
Siding with the rights holders, the lawmakers said that it was not “fair for tech firms to use rightsholder data for commercial purposes without permission or compensation.” They warned that “the current legal framework is failing to ensure [fair] outcomes occur.”
“The government has a duty to act. It cannot sit on its hands for the next decade until sufficient case law has emerged,” they warned. Ministers should clarify how they understand the current legal limbo. If they conclude that existing safeguards are insufficient, they should “set out options for updating legislation.”
With voluntary talks repeatedly failing to reach any consensus, the committee suggested a deadline of spring 2024, after which ministers “must set out options and prepare to resolve the dispute definitively” if there is no agreement.
In November, AI minister Jonathan Berry said the government would not “get into an endless talking shop about this” as he sought to turn up the heat on negotiators to move or face regulatory intervention, but it is understood the talks are not progressing.
