Artificial Intelligence (Regulation) Bill [HL]

2nd reading
Friday 22nd March 2024 Lords Chamber

Second Reading

10:06

Moved by

[Lord Holmes of Richmond]

That the Bill be now read a second time.

[Lord Holmes of Richmond (Con)]

My Lords, I declare my technology interests as adviser to Boston Ltd. I thank all noble Lords who have signed up to speak; I eagerly anticipate all their contributions and, indeed, hearing from my noble friend the Minister. I also thank all the organisations that got in contact with me and other noble Lords for their briefings, as well as those that took time to meet me ahead of this Second Reading debate. Noble Lords and others who would like to follow this on social media can use #AIBill #AIFutures.

If we are to secure the opportunities and control the challenges of artificial intelligence, it is time to legislate and to lead. We need something that is principles-based and outcomes-focused, with input transparent, permissioned and wherever applicable paid for and understood.

There are at last three reasons why we should legislate on this: social, democratic and economic. On reason one, the social reason, some of the greatest benefits we could secure from AI come in this space, including truly personalised education for all, and healthcare. We saw only yesterday the exciting early results from the NHS Grampian breast-screening AI programme. Then there is mobility and net zero sustainability.

Reason two is about our democracy and jurisdiction. With 40% of the world’s democracies going to the polls this year, with deepfakes, cheap fakes, misinformation and disinformation, we are in a high-threat environment for our democracy. As our 2020 Democracy and Digital Technologies Select Committee report put it, with a proliferation of misinformation and disinformation, trust will evaporate and, without trust, democracy as we know it will simply disappear.

On our jurisdiction and system of law, the UK has a unique opportunity at this moment in time. We do not have to fear being in the first mover spotlight—the EU has taken that with its Act, in all its 892 pages. The US has had the executive order but is still yet to commit fully to this phase. The UK, with our common-law tradition, respected right around the world, has such an opportunity to legislate in a way that will be adaptive, versatile and able to develop through precedent and case law.

On reason three, our economy, PwC’s AI tracker says that by 2030, there will be a 14% increase in global GDP worth $15.7 trillion. The UK must act to ensure our share of that AI boom. To take just one technology, the chatbot global market grew tenfold in just four years. The Alan Turing Institute report on AI in the public sector, which came out just this week, says that 84% of government services could benefit from AI automation in over 200 different services. Regulated markets perform better. Right-sized regulation is good for innovation and good for inward investment.

Those are the three reasons. What about three individual impacts of AI right now? What if we find ourselves on the wrong end of an AI decision in a recruitment shortlisting, the wrong end of an AI decision in being turned down for a loan, or, even worse, the wrong end of an AI decision when awaiting a liver transplant? All these are illustrations of AI impacting individuals, often when they would not even know that AI was involved. We need to put paid to the myth, the false dichotomy, that you must have heavy, rules-based regulation or a free hand—that we have to pay tribute to the cry of the frontierists in every epoque: “Don’t fence me in”. Right-sized regulation is good socially, democratically and economically. Here is the thing: AI is to human intellect what steam was to human strength. You get the picture. Steam literally changed time. It is our time to act, and that is why I bring this Bill to your Lordships’ House today.

In constructing the Bill, I have sought to consult widely, to be very cognisant of the Government’s pro-innovation White Paper, of all the great work of BCS, technology, industry, civil society and more. I wanted the Bill to be threaded through with the principles of transparency and trustworthiness; inclusion and innovation; interoperability and international focus; accountability and assurance.

Turning to the clauses, Clause 1 sets up an AI authority. Lest any noble Lord suddenly feels that I am proposing a do-it-all, huge, cumbersome regulator, I am most certainly not. In many ways, it would not be much bigger in scope than what the DSIT unit is proposing: an agile, right-sized regulator, horizontally focused to look across all existing regulators, not least the economic regulators, to assess their competency to address the opportunities and challenges presented by AI and to highlight the gaps. And there are gaps, as rightly identified by the excellent Ada Lovelace Institute report. For example, where do you go if you are on the wrong end of that AI recruitment shortlisting decision? It must have the authority, similarly, to look across all relevant legislation—consumer protection and product safety, to name but two—to assess its competency to address the challenges and opportunities presented by AI.

The AI authority must have at its heart the principles set out in Clause 2: it must be not just the custodian of those principles, but a very lighthouse for them, and it must have an educational function and a pro-innovation purpose. Many of those principles will be very recognisable; they are taken from the Government’s White Paper but put on a statutory footing. If they are good enough to be in the White Paper, we should commit to them, believe in them and know that they will be our greatest guides for the positive path forward, when put in a statutory framework. We must have everything inclusive by design, and with a proportionality thread running through all the principles, so none of them can be deployed in a burdensome way.

Clause 3 concerns sandboxes, so brilliantly developed in the UK in 2016 with the fintech regulatory sandbox. If you want a measure of its success, it is replicated in well over 50 jurisdictions around the world. It enables innovation in a safe, regulated, supported environment: real customers, real market, real innovations, but in a splendid sandbox concept.

Clause 4 sets up the AI responsible officer, to be conceived of not as a person but as a role, to ensure the safe, ethical and unbiased deployment of AI in her or his organisation. It does not have to be burdensome, or a whole person in a start-up; but that function needs to be performed, with reporting requirements under the Companies Act that are well understood by any business. Again, crucially, it must be subject to that proportionality principle.

Clause 5 concerns labelling and IP, which is such a critical part of how we will get this right with AI. Labelling: so that if anybody is subject to a service or a good where AI is in the mix, it will be clearly labelled. AI can be part of the solution to providing this labelling approach. Where IP or third-party data is used, that has to be reported to the AI authority. Again, this can be done efficiently and effectively using the very technology itself. On the critical question of IP, I met with 25 organisations representing tens of thousands of our great creatives: the people that make us laugh, make us smile, challenge us, push us to places we never even knew existed; those who make music, such sweet music, where otherwise there may be silence. It is critical to understand that they want to be part of this AI transformation, but in a consented, negotiated, paid-for manner. As Dan Guthrie, director-general of the Alliance for Intellectual Property, put it, it is extraordinary that businesses together worth trillions take creatives’ IP without consent and without payment, while fiercely defending their own intellectual property. This Bill will change that.

Clause 6 concerns public engagement. For me, this is probably the most important clause in the Bill, because without public engagement, how can we have trustworthiness? People need to be able to ask, “What is in this for me? Why should I care? How is this impacting my life? How can I get involved?” We need to look at innovative ways to consult and engage. A good example, in Taiwan, is the Alignment Assemblies, but there are hundreds of novel approaches. Government consultations should have millions of responses, because this is both desirable and now, with the technology, analysable.

Clause 7 concerns interpretation. At this stage, I have drawn the definitions of AI deliberately broadly. We should certainly debate this, but as set out in Clause 7, much would and should be included in those definitions.

Clause 8 sets out the potential for regulating for offences and fines thereunder, to give teeth to so much of what I have already set out and, rightly, to pay the correct respect to all the devolved nations. So, such regulations would have to go through the Scottish Parliament, Senedd Cymru and the Northern Ireland Assembly.

That brings us to Clause 9, the final clause, which makes this a UK-wide Bill.

So, that is the Bill. We know how to do this. Just last year, the Electronic Trade Documents Act showed that we know how to legislate for the possibilities of these new technologies; and, my word, we know how to innovate in the UK—Turing, Lovelace, Berners-Lee, Demis at DeepMind, and so many more.

If we know how to do this, why are we not legislating? What will we know in, say, 12 months’ time that we do not know now about citizens’ rights, consumer protection, IP rights, being pro-innovation, labelling and the opportunity to transform public engagement? We need to act now, because we know what we need to know—if not now, when? The Bletchley summit last year was a success. Understandably, it focused on safety, but having done that it is imperative that we stand up all the other elements of AI already impacting people’s lives in so many ways, often without their knowledge.

Perhaps the greatest and finest learning from Bletchley is not so much the safety summit but what happened there two generations before, when a diverse team of talent gathered and deployed the technology of their day to defeat the greatest threat to our civilisation. Talent and technology brought forth light in one of the darkest hours of human history. As it was in Bletchley in the 1940s, so it is in the United Kingdom in the 2020s. It is time for human-led, human-in-the-loop, principle-based artificial intelligence. It is time to legislate and to lead; for transparency and trustworthiness, inclusion and innovation, interoperability and international focus, accountability and assurance; for AI developers, deployers and democracy itself; for citizens, creatives and our country—our data, our decisions, #ourAIfutures. That is what this Bill is all about. I beg to move.

10:22

[Viscount Chandos (Lab)]

My Lords, there can have been few Private Members’ Bills that have set out to address such towering issues as this Bill from the noble Lord, Lord Holmes of Richmond. He has been an important voice on the opportunities and challenges arising from generative AI in this House and outside it. This Bill and his powerful introduction to it are only the latest contributions to the vital issue of regulating AI to ensure that the social and financial interests and security of consumers are protected as a first priority.

The noble Lord also contributed to a wide-ranging discussion on the regulation of AI in relation to misinformation and disinformation convened by the Thomson Foundation, of which, as recorded in the register, I am chair. Disinformation in news has existed and grown as a problem since long before the emergence of generative AI, but each iteration of AI makes the disinformation that human bad actors promote even harder to detect and counteract.

This year a record number of people in the world will go to the polls in general elections, as the noble Lord said. The Thomson Foundation has commissioned research into the incidence of AI-fuelled disinformation in the Taiwanese presidential elections in mid-January, conducted by Professor Chen-Ling Hung of National Taiwan University. Your Lordships may not be surprised that the preliminary conclusions of the work, which will be continued in relation to other elections, confirms the concerns that the noble Lord voiced in his introduction. Generative AI’s role in exacerbating misinformation and disinformation in news and the impact that can have on the democratic process are hugely important, but this is only one of a large number of areas where generative AI is at the same time an opportunity and a threat.

I strongly support this well-judged and balanced Bill, which recognises the fast-changing, dynamic nature of this technology—Moore’s law on steroids, as I have previously suggested—and sets out a logical and coherent role for the proposed AI authority, bringing a transparency and clarity to the regulation of AI for its developers and users that is currently lacking.

I look forward to the Minister’s winding up, but with my expectations firmly under control. The Prime Minister’s position seems incoherent. On the one hand he says that generative AI poses an existential threat and on the other that no new regulatory body is needed and the technology is too fast-moving for a comprehensive regulatory framework to be established. That is a guarantee that we will be heaving to close a creaking stable door as the thoroughbred horse disappears over the horizon. I will not be surprised to hear the Minister extol the steps taken in recent months, such as the establishment of the AI unit, as a demonstration that everything is under control. Even if these small initiatives are welcome, they fall well short of establishing the transparency and clarity of regulation needed to engender confidence in all parties—consumers, employers, workers and civil society.

If evidence is needed to make the case for a transparent, well-defined regulatory regime rather than ad hoc, fragmented departmental action, the Industry and Regulators Committee, of which I am privileged to be a member, today published a letter to the Secretary of State for Levelling Up about the regulation of property agents. Five years ago, a working group chaired by the noble Lord, Lord Best, recommended that the sector should be regulated, yet despite positive initial noises from the Government, nothing has happened. Even making allowance for the impact of the pandemic during this time, this does not engender confidence in their willingness and ability to grasp regulatory nettles in a low-tech industry, let alone in a high-tech one.

It is hard not to suspect that this reflects an ideological suspicion within the Conservative Party that regulation is the enemy of innovation and economic success rather than a necessary condition, which I believe it is. Evidence to the Industry and Regulators Committee from a representative of Merck confirmed that the life sciences industry thrives in countries where there is strong regulation.

I urge the Government to give parliamentary time to this Bill to allow it to go forward. I look forward to addressing its detailed issues then.

10:28

[Lord Young of Cookham (Con)]

My Lords, one of the advantages of sitting every day between my noble friends Lord Holmes and Lord Kirkhope is that their enthusiasm for a subject on which they have a lot of knowledge and I have very little passes by a process of osmosis along the Bench. I commend my noble friend on his Bill and his speech. I will add a footnote to it.

My noble friend’s Bill is timely, coming after the Government published their consultation outcome last month, shortly after the European Commission published its Artificial Intelligence Act and as we see how other countries, such as the USA, are responding to the AI challenge. Ideally, there should be some global architecture to deal with a phenomenon that knows no boundaries. The Prime Minister said as much in October:

“My vision, and our ultimate goal, should be to work towards a more international approach to safety where we collaborate with partners to ensure AI systems are safe”.


However, we only have to look at the pressures on existing international organisations, like the United Nations and the WTO, to see that that is a big ask. There is a headwind of protectionism, and at times nationalism, making collaboration difficult. It is not helped by the world being increasingly divided between democracies and autocracies, with the latter using AI as a substitute for conventional warfare.

The most pragmatic approach, therefore, is to go for some lowest common denominators, building on the Bletchley Declaration which talks about sharing responsibility and collaboration. We want to avoid regulatory regimes that are incompatible, which would lead to regulatory arbitrage and difficulties with compliance.

The response to the consultation refers to this in paragraphs 71 and 72, stating:

“the intense competition between companies to release ever-more-capable systems means we will need to remain highly vigilant to meaningful compliance, accountability, and effective risk mitigation. It may be the case that commercial incentives are not always aligned with the public good”.

It concludes:

“the challenges posed by AI technologies will ultimately require legislative action in every country once understanding of risk has matured”.

My noble friend’s Private Member’s Bill is a heroic first shot at what that legislation might look like. To simplify, there is a debate between top-down, as set out in the Bill, and bottom-up, as set out in the Government’s response, delegating regulation to individual regulators with a control function in DSIT. At some point, there will have to be convergence between the two approaches.

There is one particular clause in my noble friend’s Bill that I think is important: Clause 1(2)(c), which states that the function of the AI authority is to,

“undertake a gap analysis of regulatory responsibilities in respect of AI”.

The White Paper and the consultation outcome have numerous references to regulators. What I was looking for and never found was a list of all our regulators, and what they regulate. I confess I may have missed it, but without such a comprehensive list of regulators and what they regulate, any strategy risks being incomplete because we do not have a full picture.

My noble friend mentioned education. We have a shortage of teachers in many disciplines, and many complain about paperwork and are thinking of leaving. There is a huge contribution to be made by AI. But who is in charge? If you put the question into Google, it says,

“the DFE is responsible for children’s services and education”.

Then there is Ofsted, which inspects schools; there is Ofqual, which deals with exams; and then there is the Office for Students. The Russell group of universities have signed up to a set of principles ensuring that pupils would be taught to become AI literate.

Who is looking at the huge volume of material which AI companies are drowning schools and teachers with, as new and more accessible chatbots are developed? Who is looking at AI for marking homework? What about AI for adaptive testing? Who is looking at AI being used for home tuition, as increasingly used by parents? Who is looking at AI for marking papers? As my noble friend said, what happens if they get it wrong?

The education sector is trying to get a handle on this technological maelstrom and there may be some bad actors in there. However, the same may be happening elsewhere because the regulatory regimes lack clarity. Hence, should by any chance my noble friend’s Bill not survive in full, Clause 1(2)(c) should.

10:33

[Lord Thomas of Cwmgiedd (CB)]

My Lords, I also warmly support the Bill introduced by the noble Lord, Lord Holmes of Richmond. I support it because it has the right balance of radicalism to fit the revolution in which we are living. I will look at it through eight points—that may be ambitious in five minutes, but I think I can do it.

There is a degree of serious common ground. First, we need fair standards to protect the public. We need to protect privacy, security, human rights, fraud and intellectual property. We also need to protect, however, rights to access, like data and the processes by which artificial intelligence makes decisions in respect of you. An enforcement system is needed to make that work. If we have that, we do not need the elaborate mechanism of the EU by regulating individual products.

Secondly, it is clear there has to be a consistency of standards. We cannot have one rule for one market, and one rule for another market. If you look back at the 19th century, when we underwent the last massive technological revolution, the courts sometimes made the mistake of fashioning rules to fit individual markets. That was an error, and that is why we need to look at it comprehensively.

Thirdly, we have got to protect innovation. I believe that is common ground, but the points to which I shall come in a moment show the difficulties.

Fourthly, we have got to produce a system that is interoperable. The noble Lord, Lord Holmes, referred to the trade documents Bill, which was the product of international development. We have adapted the common law to fit it and other countries’ systems will do it. That is a sine qua non.

I believe all those points are common ground, but I now come to four points that I do not think are common ground. The first is simplicity. When you look at Bills in this House, I sometimes feel we are making the law unworkable by its complexity. There can be absolutely no doubt that regulation is becoming unworkable because of the complexity. I can quite understand why innovators are horrified at the prospect of regulation, but they have got the wrong kind of regulation. They have got what we have created, unfortunately; it is a huge burden and is not based on simplicity and principles. If we are to persuade people to regulate, we need a radically different approach, and this Bill brings it about.

Secondly, there needs to be transparency and accountability. I do not believe that doing this through a small body within a ministry is the right way; it has to be done openly.

Thirdly—and this is probably highly controversial—when you look at regulation, our idea is of the statutory regulator with its vast empire created. Do we need that? Look back at the 19th century: the way in which the country developed was through self-regulation supported by the courts, Parliament and government. We need to look at that again. I see nothing wrong with self-regulation. It has itself a shocking name, as a result of what happened in the financial markets at the turn of the century, but I believe that we should look at it again. Effective self-regulation can be good regulation.

Finally, the regulator must be independent. There is nothing inconsistent with self-regulation and independence.

We need a radical approach, and the Bill gives us that. No one will come here if we pretend we are going to set up a regulator—like the financial markets regulator, the pensions regulator and so on—because people will recoil in horror. If we have this Bill, however, with its simplicity and emphasis on comprehensiveness, we can do it. By saying that, it seems to me that the fundamental flaw in what the Government are doing is leaving the current regulatory system in place. We cannot afford to do that. We need to face the new industrial revolution with a new form of regulation.

10:38

[Baroness Stowell of Beeston (Con)]

My Lords, it is a great pleasure to follow the noble and learned Lord, Lord Thomas, and his interesting speech. I remind noble Lords that the Communications and Digital Committee, which I have the privilege to chair, published our report Large Language Models and Generative AI only last month. For anyone who has not read it, I wholeheartedly recommend it, and I am going to draw heavily on it in my speech.

It is a pleasure to speak in a debate led by my noble friend Lord Holmes, and I congratulate him on all that he does in the digital and technology space. As he knows, I cannot support his Bill because I do not agree with the concept of an AI authority, but I have listened carefully to the arguments put forward by the noble and learned Lord, Lord Thomas, a moment ago. But neither would I encourage the Government to follow the Europeans and rush to develop overly specific legislation for this general-purpose technology.

That said, there is much common ground on which my noble friend and I can stand when it comes to our ambitions for AI, so I will say a little about that and where I see danger with the Government’s current approach to this massively important technological development.

As we have heard, AI is reshaping our world. Some of these changes are modest, and some are hype, but others are genuinely profound. Large language models in particular have the potential to fundamentally reshape our relationship with machines. In the right hands, they could drive huge benefits to our economy, supporting ground-breaking scientific research and much more.

I agree with my noble friend Lord Holmes about how we should approach AI. It must be developed and used to benefit people and society, not just big tech giants. Existing regulators must be equipped and empowered to hold tech firms to account as this technology operates in their own respective sectors, and we must ensure that there are proper safety tests for the riskiest models.

That said, we must maintain an open market for AI, and so any testing must not create barriers to entry. Indeed, one of my biggest fears is an even greater concentration of power among the big tech firms and repeating the same mistakes which led to a single firm dominating search, no UK-based cloud service, and a couple of firms controlling social media. Instead, we must ensure that generative AI creates new markets and, if possible, use it to address the existing market distortions.

Our large language model report looked in detail at what needs to happen over the next three years to catalyse AI innovation responsibly and mitigate risks proportionately. The UK is well-placed to be among the world leaders of this technology, but we can only achieve that by being positive and ambitious. The recent focus on existential sci-fi scenarios has shifted attention towards too narrow a view of AI safety. On its own, a concentration on safety will not deliver the broader capabilities and commercial heft that the UK needs to shape international norms. However, we cannot keep up with international competitors without more focus on supporting commercial opportunities and academic excellence. A rebalance in government strategy and a more positive vision is therefore needed. The Government should improve access to computing power, increase support for digital, and do more to help start-ups grow out of university research.

I do not wish to downplay the risks of AI. Many need to be addressed quickly—for example, cyberattacks and synthetic child sexual abuse, as well as bias and discrimination, which we have already heard about. The Government should scale up existing mitigations, and ensure industry improves its own guard-rails. However, the overall point is about balance. Regulation should be thoughtful and proportionate, to catalyse rather than stifle responsible innovation, otherwise we risk creating extensive rules that end up entrenching incumbents’ market power, and we throttle domestic industry in the process. Regulatory capture is a real danger that our inquiry highlighted.

Copyright is another danger, and this is where there is a clear case for government action now. The point of copyright is to reward innovation, yet tech firms have been exploiting rights holders by using works without permission or payment. Some of that is starting to change, and I am pleased to see some firms now striking deals with publishers. However, these remain small steps, and the fundamental question about respecting copyright in the first place remains unresolved.

The role for government here is clear: it should endorse the principles behind copyright and uphold fair play, and should then update legislation. Unfortunately, the current approach remains unclear and inadequate. It has abandoned the IPO-led process, but apparently without anything more ambitious in its place. I hope for better news in the Government’s response to our report, expected next month, and it would be better still if my noble friend the Minister could say something reassuring today.

In the meantime, I am grateful to my noble friend Lord Holmes for providing the opportunity to debate such an important topic.

10:44

[Baroness Kidron (CB)]

My Lords, I too congratulate the noble Lord, Lord Holmes, on his wonderful speech. I declare my interests as an adviser to the Oxford Institute for Ethics in AI and the UN Secretary-General’s AI Advisory Body.

When I read the Bill, I asked myself three questions. Do we need an AI regulation Bill? Is this the Bill we need? What happens if we do not have a Bill? It is arguable that it would be better to deal with AI sector by sector—in education, the delivery of public services, defence, media, justice and so on—but that would require an enormous legislative push. Like others, I note that we are in the middle of a legislative push, with digital markets legislation, media legislation, data protection legislation and online harms legislation, all of which resolutely ignore both existing and future risk.

The taxpayer has been asked to make a £100 million investment in launching the world’s first AI safety institute, but as the Ada Lovelace Institute says:

“We are concerned that the Government’s approach to AI regulation is ‘all eyes, no hands’”,


with plenty of “horizon scanning” but no

“powers and resources to prevent those risks or even to react to them effectively after the fact”.

So yes, we need an AI regulation Bill.

Is this the Bill we need? Perhaps I should say to the House that I am a fan of the Bill. It covers testing and sandboxes, it considers what the public want, and it deals with a very important specific issue that I have raised a number of times in the House, in the form of creating AI-responsible officers. On that point, the CEO of the International Association of Privacy Professionals came to see me recently and made an enormously compelling case that, globally, we need hundreds of thousands of AI professionals, as the systems become smarter and more ubiquitous, and that those professionals will need standards and norms within which to work. He also made the case that the UK would be very well-placed to create those professionals at scale.

I have a couple of additions. Unless the Minister is going to make a surprise announcement, I think we are allowed to consider that he is going to take the Bill on in full. In addition, under Clause 2, which sets out regulatory principles, I would like to see consideration of children’s rights and development needs; employment rights, concerning both management by AI and job displacement; a public interest case; and more clarity that material that is an offence—such as creating viruses, CSAM or inciting violence—is also an offence, whether created by AI or not, with specific responsibilities that accrue to users, developers and distributors.

The Stanford Internet Observatory recently identified hundreds of known images of child sexual abuse material in an open dataset used to train popular AI text-to-image models, saying:

“It is challenging to clean or stop the distribution of publicly distributed datasets as it has been widely disseminated. Future datasets could use freely available detection tools to prevent the collection of known CSAM”.


The report illustrates that it is very possible to remove such images, but that it did not bother, and now those images are proliferating at scale.

We need to have rules upon which AI is developed. It is poised to transform healthcare, both diagnosis and treatment. It will take the weight out of some of the public services we can no longer afford, and it will release money to make life better for many. However, it brings forward a range of dangers, from fake images to lethal autonomous weapons and deliberate pandemics. AI is not a case of good or bad; it is a question of uses and abuses.

I recently hosted Geoffrey Hinton, whom many will know as the “godfather of AI”. His address to parliamentarians was as chilling as it was compelling, and he put timescales on the outcomes that leave no time to wait. I will not stray into his points about the nature of human intelligence, but he was utterly clear that the concentration of power, the asymmetry of benefit and the control over resources—energy, water and hardware—needed to run these powerful systems would be, if left until later, in so few hands that they, and not we, would be doing the rule setting.

My final question is: if we have no AI Bill, can the Government please consider putting the content of the AI regulation Bill into the data Bill currently passing through Parliament and deal with it in that way?

10:50

[Lord Vaizey of Didcot (Con)]

My Lords, I thought that this would be one of the rare debates where I did not have an interest to declare, but then I heard the noble Lord, Lord Young, talking about AI and education and realised that I am a paid adviser to Common Sense Media, a large US not-for-profit that campaigns for internet safety and has published the first ever ratings of AI applications used in schools. I refer the noble Lord to its excellent work in this area.

It is a pleasure to speak in the debate on this Bill, so ably put forward by the noble Lord, Lord Holmes. It is pretty clear from the reaction to his speech how much he is admired in this House for his work on this issue and so many others to do with media and technology, where he is one of the leading voices in public affairs. Let me say how humiliating it is for me to follow the noble Baronesses, Lady Stowell and Lady Kidron, both of whom are experts in this area and have done so much to advance public policy.

I am a regulator and in favour of regulation. I strongly supported the Online Safety Act, despite the narrow byways and culs-de-sac it ended up in, because I believe that platforms and technology need to be accountable in some way. I do not support people who say that the job is too big to be attempted—we must attempt it. What I always say about the Online Safety Act is that the legislation itself is irrelevant; what is relevant is the number of staff and amount of expertise that Ofcom now has, which will make it one of the world’s leaders in this space.

We talk about AI now because it has come to the forefront of consumers’ minds through applications such as ChatGPT, but large language models and the use of AI have been around for many years. As AI becomes ubiquitous, it is right that we now consider how we could or should regulate it. Indeed, with the approaching elections, not just here in the UK but in the United States and other areas around the world, we will see the abuse of artificial intelligence, and many people will wring their hands about how on earth to cope with the plethora of disinformation that is likely to emerge.

I am often asked at technology events, which I attend assiduously, what the Government’s policy is on artificial intelligence. To a certain extent I have to make it up, but to a certain extent I think that, broadly speaking, I have it right. On the one hand, there is an important focus on safety for artificial intelligence to make it as safe as possible for consumers, which in itself begs the question of whether that is possible; on the other, there is a need to ensure that the UK remains a wonderful place for AI innovation. We are rightly proud that DeepMind, although owned by Google, wishes to stay in the UK. Indeed, in a tweet yesterday the Chancellor himself bigged up Mustafa Suleyman for taking on the role of leading AI at Microsoft. It is true that the UK remains a second-tier nation in AI after China and the US, but it is the leading second-tier nation.

The question now is: what do we mean by regulation? I do not necessarily believe that now is the moment to create an AI safety regulator. I was interested to hear the contribution of the noble and learned Lord, Lord Thomas, who referred to the 19th century. I refer him to the late 20th century and the early 21st century: the internet itself has long been self-regulated, at least in terms of the technology and the global standards that exist, so it is possible for AI to proceed largely on the basis of self-regulation.

The Government’s approach to regulation is the right one. We have, for example, the Digital Regulation Cooperation Forum, which brings together all the regulators that either obviously, such as Ofcom, or indirectly, such as the FCA, have skin the game when it comes to digital. My specific request to the Minister is to bring the House up to date on the work of that forum and how he sees it developing.

I was surprised by the creation of the AI Safety Institute as a stand-alone body with such generous funding. It seems to me that the Government do not need legislation to do an examination of the plethora of bodies that have sprung up over the last 10 or 15 years. Many of them do excellent work, but where their responsibilities begin and end is confusing. They include the Ada Lovelace Institute, the Alan Turing Institute, the AI Safety Institute, Ofcom and DSIT, but how do they all fit together into a clear narrative? That is the essential task that the Government must now undertake.

I will pick up on one remark that the noble Baroness, Lady Stowell, made. While we look at the flashy stuff, if you like, such as disinformation and copyright, she is quite right to say that we have to look at the picks and shovels as AI becomes more prevalent and as the UK seeks to maintain our lead. Boring but absolutely essential things such as power networks for data centres will be important, so they must also be part of the Government’s task.

10:56

[Lord Empey (UUP)]

My Lords, like other Members, I congratulate the noble Lord, Lord Holmes, on what he has been doing.

The general public have become more aware of AI in very recent times, but it is nothing new; people have been working on it for decades. Because it is reaching such a critical mass and getting into all the different areas of our lives, it is now in the public mind. While I do not want to get into the minutiae of the Bill—that is for Committee—speaking as a non-expert, I think that the general public are now at a stage where they have a right to know what legislators think. Given the way things have developed in recent years, the Government cannot stand back and do nothing.

Like gunpowder, AI cannot be uninvented. The growing capacity of chips and other developments, and the power of a limited number of companies around the world, ensure that such a powerful tool will now be in the hands of a very small number of corporations. The Prime Minister took the lead last year and indicated that he wished to see the United Kingdom as a world leader in this field, and he went to the United States and other locations. I rather feel that we have lost momentum and that nothing is currently happening that ought to be happening.

As with all developments, there are opportunities and threats. In the last 24 hours, we have seen both. As the noble Lord, Lord Holmes, pointed out, studies on breast cancer were published yesterday, showing that X-rays, CT scans, et cetera were interpreted more accurately by AI than by humans. How many cases have we had in recent years of tests having to be recalled by various trusts, causing anxiety and stress for thousands upon thousands of patients? It is perfectly clear that, in the field of medicine alone, AI could not only improve treatment rates but relieve people of a lot of the anxieties that such inaccuracies cause. We also saw the threats on our television screens last night. As the noble Lord referred to, a well-known newscaster showed that she was presented by AI in a porn movie—she had it there on the screens for us to see last night. So you can see the threats as well as the opportunities.

So the question is: can Parliament, can government, stand by and just let things happen? I believe that the Government cannot idly stand by. We have an opportunity to lead. Yes, we do not want to create circumstances where we suffocate innovation. There is an argument over regulation between what the noble Viscount, Lord Chandos, said, what the noble and learned Lord, Lord Thomas, said, and what I think the Government’s response will be. However, bolting bits on to existing regulators is not necessarily the best way of doing business. You need a laser focus on this and you need the people with the capacity and the expertise. They are not going to be widely available and, if you have a regulator with too much on its agenda, the outcome will be fairly dilute and feeble.

In advance of this, I said to the Minister that we have all seen the “Terminator” movies, and I am sure that the general public have seen them over the years. The fact is that it is no longer as far-fetched as it once was. I have to ask the Minister: what is our capacity to deal with hacking? If it gets into weapons systems, never mind utilities, one can see straight away a huge potential for danger.

So, once again, we are delighted that the Bill has been brought forward. I would like to think that ultimately the Government will take this over, because that is the only way that it will become law, and it does need refinement. A response from the Minister, particularly on the last point, which creates huge anxiety, would be most beneficial.

11:01

[Lord Freyberg (CB)]

My Lords, I too am very grateful to the noble Lord, Lord Holmes of Richmond, for introducing this important Artificial Intelligence (Regulation) Bill. In my contribution today, I will speak directly to the challenges and threats posed to visual artists by generative AI and to the need for regulatory clarity to enable artists to explore the creative potential of AI. I declare my interest as having a background in the visual arts.

Visual artists have expressed worries, as have their counterparts in other industries and disciplines, about their intellectual property being used to train AI models without their consent, credit or payment. In January 2024, lists containing the names of more than 16,000 non-consenting artists whose works were allegedly used to train the Midjourney generative AI platform were accidentally leaked online, intensifying the debate on copyright and consent in AI image creation even further.

The legality of using human artists’ work to train generative AI programmes remains unclear, but disputes over documents such as the Midjourney style list, as it became known, provide insight into the real procedures involved in turning copyrighted artwork into AI reference material. These popular AI image-generator models are extremely profitable for their owners, the majority of whom are situated in the United States. Midjourney was valued at around $10.5 billion in 2022. It stands to reason that, if artists’ IP is being used to train these models, it is only fair that they be compensated, credited and given the option to opt out.

DACS, the UK’s leading copyright society for artists, of which I am a member, conducted a survey that received responses from 1,000 artists and their representatives, 74% of whom were concerned about their own work being used to train AI models. Two-thirds of artists cited ethical and legal concerns as a barrier to using such technology in their creative practices. DACS also heard first-hand accounts of artists who found that working creatively with AI has its own set of difficulties, such as the artist who made a work that included generative AI and wanted to distribute it on a well-known global platform. The platform did not want the liabilities associated with an unregistered product, so it asked for the AI component to be removed. If artists are deterred from using AI or face legal consequences for doing so, creativity will suffer. There is a real danger that artists will miss out on these opportunities, which would worsen their already precarious financial situation and challenging working conditions.

In the same survey, artists expressed fear that human-made artworks will have no distinctive or unique value in the marketplace in which they operate, and that AI may thereby render them obsolete. One commercial photographer said, “What’s the point of training professionally to create works for clients if a model can be trained on your own work to replace you?” Artists rely on IP royalties to sustain a living and invest in their practice. UK artists are already low-paid and two-thirds are considering abandoning the profession. Another artist remarked in the survey, “Copyright makes it possible for artists to dedicate time and education to become a professional artist. Once copyright has no meaning any more, there will be no more possibility to make a living. This will be detrimental to society as a whole”.

It is therefore imperative that we protect their copyright and provide fair compensation to artists whose works are used to train artificial intelligence. While the Bill references IP, artists would have welcomed a specific clause on remuneration and an obligation for owners of copyright material used in AI training to be paid. To that end, it is therefore critical to maintain a record of every work that AI applications use, particularly to validate the original artist’s permission. It is currently not required by law to reveal the content that AI systems are trained on. Record-keeping requirements are starting to appear in regulatory proposals related to AI worldwide, including those from China and the EU.

The UK ought to adopt a similar mandate requiring companies using material in their AI systems to keep track of the works that they have learned and ingested. To differentiate AI-generated images from human-composed compositions, the Government should make sure that any commercially accessible AI-generated works are branded as such. As the noble Lord, Lord Holmes, has already mentioned, labelling shields consumers from false claims about what is and is not AI-generated. Furthermore, given that many creators work alone, every individual must have access to clear, appropriate redress mechanisms so that they can meaningfully challenge situations where their rights have been misused. Having said that, I welcome the inclusion in the Bill that any training data must be preceded by informed consent. This measure will go some way to safeguarding artists’ copyright and providing them with the necessary agency to determine how their work is used in training, and on what terms.

In conclusion, I commend the noble Lord, Lord Holmes, for introducing this Bill, which will provide much-needed regulation. Artists themselves support these measures, with 89% of respondents to the DACS survey expressing a desire for more regulation around AI. If we want artists to use AI and be creative with new technology, we need to make it ethical and viable.

11:07

[Baroness Moyo (Con)]

My Lords, I join other noble Lords in commending the noble Lord, Lord Holmes, for bringing forward this Bill.

I come to this debate with the fundamental belief that supporting innovation and investment must be embedded in all regulation, but even more so in the regulation of artificial intelligence. After all, this wave of artificial intelligence is being billed as a catalyst that could propel economic growth and human progress for decades to come. The United Kingdom should not miss this supercycle and the promise of a lengthy period of economic expansion—the first of its kind since deglobalisation and deregulation 40 years ago.

With this in mind, in reading the AI regulation Bill I am struck by the weight of emphasis on risk mitigation, as opposed to innovation and investment. I must say that right at this moment, notwithstanding the fact that I realise that the Government, through other routes, including the pro-innovation stance that we talked about, are looking into innovation in investment. Even so, I feel that, on balance, the weight here is more on risk mitigation than innovation. I am keen that, in the drafting and execution of the artificial intelligence authority’s mandate in particular, and in the evolution of this Bill in general, the management of risk does not deter investment in this game-changing innovation.

I am of course reassured that innovation or opportunity are mentioned at least two times in the Bill. For example, Clause 6(a) signals that the public engagement exercise will consider

“the opportunities and risks presented by AI”.

Perhaps more pointedly, Clause 1(2)(e) states that the list of functions of the Al Authority are to include support for innovation. However, this mandate is at best left open to interpretation and at worst downgrades the importance and centrality of innovation.

My concern is that the new AI authority could see support for innovation as a distant or secondary objective, and that risk-aversion and mitigation become the cultural bedrock of the organisation. If we were to take too heavy-handed a risk-mitigation approach to AI, what opportunities could be missed? In terms of economic growth, as my noble friend Lord Holmes mentioned, PricewaterhouseCoopers estimates that AI could contribute more than $15 trillion to the world economy by 2030. In this prevailing era of slow economic growth, AI could meaningfully alter the growth trajectory.

In terms of business, AI could spur a new start-up ecosystem, creating a new generation of small and medium-sized enterprises. Furthermore, to underscore this point, AI promises to boost productivity gains, which could help generate an additional $4.4 trillion in annual profits, according to a 2023 report by McKinsey. To place this in context, this annual gain is nearly one and a half times larger than the UK’s annual GDP.

On public goods such as education and healthcare, the Chancellor in his Spring Budget a few weeks ago indicated the substantial role that a technology upgrade, including the use of AI, could play in improving delivery and access and in unlocking up to £35 billion of savings.

Clearly, a lot is at stake. This is why it is imperative that this AI Bill, and the way it is interpreted, strikes the right balance between mitigating risk and supporting investment and innovation.

I am very much aware of the perennial risks of malevolent state actors and errant new technologies, and thus, the need for effective regulation is clear, as the noble and learned Lord, Lord Thomas, stressed. This is unambiguous, and I support the Bill. However, we must be alert to the danger of regulation becoming a synonym for risk-management. This would overshadow the critical regulatory responsibility of ensuring a competitive environment in which innovation can thrive and thereby attract investment.

11:12

[The Lord Bishop of Worcester]

My Lords, I guarantee that this is not an AI-generated speech. Indeed, Members of the House might decide after five minutes that there is not much intelligence of any kind involved in its creation. Be that as it may, we on these Benches have engaged extensively with the impacts and implications of new technologies for years—from contributions to the Warnock committee in the 1980s through to the passage of the Online Safety Bill through this House last year. I am grateful to the noble Lord, Lord Holmes, for this timely and thoughtful Bill and for his brilliant introduction to it. Innovation must be enthusiastically encouraged, as the noble Baroness, Lady Moyo, has just reminded us. It is a pleasure to follow her.

That said, I will take us back to first principles for a moment: to Christian principles, which I hope all of good will would want to support. From these principles arise two imperatives for regulation and governance, whatever breakthroughs new technologies enable. The first is that a flourishing society depends on respecting human dignity and agency. The more any new tool threatens such innate dignity, the more carefully it should be evaluated and regulated. The second imperative is a duty of government, and all of us, to defend and promote the needs of the nation’s weak and marginalised —those who cannot always help themselves. I am not convinced that the current pro-innovation and “observe first, intervene later” approach to AI get this perennial balance quite right. For that reason, I support the ambitions outlined in the Bill.

There are certainly aspects of last year’s AI White Paper that get things in the right order: I warmly commend the Government for including fairness, accountability and redress among the five guiding principles going forward. Establishing an AI authority would formalise the hub-and-spoke structure the Government are already putting in place, with the added benefit of shifting from a voluntary to a compulsory basis, and an industry-funded regulatory model of the kind the Online Safety Act is beginning to implement.

The voluntary code of practice on which the Government’s approach currently depends is surely inadequate. The track record of the big tech companies that developed the AI economy and are now training the most powerful AI models shows that profit trumps users’ safety and well-being time and again. “Move fast and break things” and “act first, apologise later” remains the lodestar. Sam Altman’s qualities of character and conduct while at the helm of OpenAI have come under considerable scrutiny over the last few months. At Davos in January this year, the Secretary-General of the United Nations complained:

“Powerful tech companies are already pursuing profits with a reckless disregard for human rights, personal privacy, and social impact.”


How can it be right that the richest companies in history have no mandatory duties to financially support a robust safety framework? Surely, it should not be for the taxpayer alone to shoulder the costs of an AI digital hub to find and fix gaps that lead to risks or harm. Why should the taxpayer shoulder the cost of providing appropriate regulatory sandboxes for testing new product safety?

The Government’s five guiding principles are a good guide for AI, but they need legal powers underpinning them and the sharpened teeth of financial penalties for corporations that intentionally flout best practice, to the clear and obvious harm of consumers.

I commend the ambitions of the Bill. A whole-system, proportional and legally enforceable approach to regulating AI is urgently needed. Balancing industry’s need to innovate with its duty to respect human dignity and the vulnerable in society is vital if we are safely to navigate the many changes and challenges not just over the horizon but already in plain sight.

11:17

[Lord Davies of Brixton (Lab)]

My Lords, I speak not as an expert in AI but as a user, and I make no apology for the fact that I use it to do my work here in this Chamber. Your Lordships can form your own judgment as to which bits of my following remarks were written by me, and which are from ChatGPT.

I very much welcome the Bill. The Noble Lord, Lord Holmes of Richmond, gave us an inspirational speech which was totally convincing on the need for legislation. The Bill is obviously the first step on that way. The promise of artificial intelligence is undeniable. There is a large degree of hype from those with vested interests, and there is, to a significant extent, a bubble. Nevertheless, even if that is true, we still need an appropriate level of regulation.

AI provides the opportunity to revolutionise industries, enhance our daily lives and solve some of the most pressing problems we face today—from healthcare to climate change—and solutions that are not available in other ways. However, with greater power comes greater responsibility. The rapid advance of AI technology has outpaced our regulatory frameworks, leading to innovation without adequate oversight, ethical consideration or accountability, so we undoubtedly need a regulator. I take the point that it has to be focused and simple. We need rigorous ethical standards and transparency in AI development to ensure that these technologies serve the good of all, not just commercial interests. We cannot wait for these forces to play out before deciding what needs to be done. I very much support the remarks of the previous speaker, the right reverend Prelate the Bishop of Worcester, who set out the position very clearly.

We need to have a full understanding of the implications of AI for employment and the workforce. These technologies will automate tasks previously performed by humans, and we face significant impacts on the labour market. The prevailing model for AI is to seek the advantage for the developers and not so much for the workers. This is an issue we will need to confront. We will have to debate the extent to which that is the job of the regulator.

As I indicate, I favour a cautious approach to AI development. We should be focusing on meaningful applications that prioritise human well-being and benefits to society over corporate profit. Again, how this fits in with the role of the regulator is for discussion, but a particular point that needs to be made here is that we need to understand the massive amounts of energy that even simple forms of AI consume. This needs to be borne in mind in any approach to developing this industry.

In the Bill, my attention was caught by the use of the undefined term “relevant regulators”. Perhaps the noble Lord, Lord Holmes, could fill that in a bit more; it is a bit of a catch-all at the moment. My particular concern is the finance industry, which will use this technology massively, not necessarily to the benefit of consumers. The noble and learned Lord, Lord Thomas of Cwmgiedd, emphasised the problem of regulatory arbitrage. We need a consistent layer of regulation. Another concern is mental health: there will be AI systems that claim to offer benefits to those with mental health problems. Again, this will need severe regulation.

To conclude, I agree with my noble friend Lord Chandos that regulation is necessarily the enemy of economic success. There is a balance to be drawn between gaining all the benefits of technology and the potential downsides. I welcome the opportunity to discuss how this should be regulated.

11:22

[Lord Fairfax of Cameron (Con)]

My Lords, I too congratulate my noble friend Lord Holmes on bringing forward this AI regulation Bill, in the context of the continuing failure of the Government to do so. At the same time, I declare my interest as a long-term investor in at least one fund that invests in AI and tech companies.

A year ago, one of the so-called godfathers of AI, Geoffrey Hinton, cried “fire” about where AI was going and, more importantly, when. Just last week, following the International Dialogue on AI Safety in Beijing, a joint statement was issued by leading western and Chinese figures in the field, including Chinese Turing award winner Andrew Yao, Yoshua Bengio and Stuart Russell. Among other things, that statement said:

“Unsafe development, deployment, or use of AI systems may pose catastrophic or even existential risks to humanity within our lifetimes … We should immediately implement domestic registration for AI models and training runs above certain compute or capability thresholds”.


Of course, we are talking about not only extinction risks but other very concerning risks, some of which have been mentioned by my noble friend Lord Holmes: extreme concentration of power, deepfakes and disinformation, wholesale copyright infringement and data-scraping, military abuse of AI in the nuclear area, the risk of bioterrorism, and the opacity and unreliability of some AI decision-making, to say nothing of the risk of mass unemployment. Ian Hogarth, the head of the UK AI Safety Institute, has written in the past about some of these concerns and risks.

Nevertheless, despite signing the Center for AI Safety statement and publicly admitting many of these serious concerns, the leading tech companies continue to race against each other towards the holy grail of artificial general intelligence. Why is this? Well, as they say, “It’s the money, stupid”. It is estimated that, between 2020 and 2022, $600 billion in total was invested in AI development, and much more has been since. This is to be compared with the pitifully small sums invested by the AI industry in AI safety. We have £10 million from this Government now. These factors have led many people in the world to ask how it is that they have accidentally outsourced their entire futures to a few tech companies and their leaders. Ordinary people have a pervading sense of powerlessness in the face of AI development.

These facts also raise the question of why the Government continue to delay putting in place proper and properly funded regulatory frameworks. Others, such as the EU, US, Italy, Canada and Brazil, are taking steps towards regulation, while, as noble Lords have heard, China has already regulated and India plans to regulate this summer. Here, the shadow IT Minister has indicated that, if elected, a new Labour Government would regulate AI. Given that a Government’s primary duty is to keep their country safe, as we so often heard recently in relation to the defence budget, this is both strange and concerning.

Why is this? There is a strong suspicion in some quarters that the Prime Minister, having told the public immediately before the Bletchley conference that AI brings national security risks that could end our way of life, and that AI could pose an extinction risk to humanity, has since succumbed to regulatory capture. Some also think that the Government do not want to jeopardise relations with leading tech companies while the AI Safety Institute is gaining access to their frontier models. Indeed, the Government proudly state that they

“will not rush to legislate”,

reinforcing the concern that the Prime Minister may have gone native on this issue. In my view, this deliberate delay on the part of the Government is seriously misconceived and very dangerous.

What have the Government done to date? To their credit, they organised and hosted Bletchley, and importantly got China to attend too. Since then, they have narrowed the gap between themselves and the tech companies—but the big issues remain, particularly the critical issue of regulation versus self-regulation. Importantly, and to their credit, the Government have also set up the UK AI Safety Institute, with some impressive senior hires. However, no one should be in any doubt that this body is not a regulator. On the critical issue of the continuing absence of a dedicated unitary AI regulator, it is simply not good enough for the Government to say that the various relevant government bodies will co-operate together on oversight of AI. It is obvious to almost everyone, apart from the Government themselves, that a dedicated, unitary, high-expertise and very well-funded UK AI regulator is required now.

The recent Gladstone AI report, commissioned by the US Government, has highlighted similar risks to US national security from advanced AI development. Against this concerning background, I strongly applaud my noble friend Lord Holmes for bringing forward the Bill. It may of course be able to be improved, but its overall intention and thrust are absolutely right.

11:28

[The Earl of Erroll (CB)]

My Lords, I entirely agree with those last sentiments, which will get us thinking about what on earth we do about this. An awful lot of nonsense is talked, and a lot of great wisdom is talked. The contributions to the debate have been useful in getting people thinking along the right lines.

I will say something about artificial general intelligence, which is very different, because it may well aim to control people or the environment in which we live, rather than generative AI or large language models, which I think people are thinking of: ChatGPT, Llama, Google Gemini, and all those bits and pieces. They are trawling through large amounts of information incredibly usefully and producing a good formatted epitome of what is in there. Because you do not have time to read, for instance, large research datasets, they can find things in them that you have not had time to trawl through and find. They can be incredibly useful for development there.

AI could start to do other things: it could control things and we could make it take decisions. Some people suggest that it could replace the law courts and a lot of those sorts of things. But the problem with that is that we live in a complex world and complex systems are not deterministic, to use a mathematical thing. You cannot control them with rules. Rules have unintended consequences, as is well known—the famous butterfly effect. You cannot be certain about what will happen when you change one little bit. AI will not necessarily be able to predict that because, if you look at how it trains itself, you do not know what it has learned—it is not done by algorithm, and some AI systems can modify their own code. So you do not know what it is doing and you cannot regulate for the algorithms or any of that.

I think we have to end up regulating, or passing laws on, the outcomes. We always did this in common law: we said, “Thou shalt not kill”, and then we developed it a bit further, but the principle of not going around killing people was established. The same is true of other simple things like “You shan’t nick things”. It is what comes out of it that matters. This applies when you want to establish liability, which we will have to do in the case of self-driving cars, for instance, which will take over more and more as other things get clogged up. They will crash less, kill fewer people and cause fewer accidents. But, because it is a machine doing it, it will be highly psychologically unacceptable—with human drivers, there will be more accidents. There will have to be changes in thought on that.

Regulation or legislation has to be around the outcomes rather than the method, because we cannot control where these things go. A computer does not have an innate sense of right and wrong or empathy, which comes into human decisions a lot. We may be able to mimic it, and we could probably train computers up on models to try to do that. One lot of AI might try to say whether another lot of AI is producing okay outcomes. It will be very interesting. I have no idea how we will get there.

Another thing that will be quite fun is when the net-zero people get on to these self-training models. An LLM trawling through data uses huge amounts of energy, which will not help us towards our net-zero capabilities. However, AI might help if we put it in charge of planning how to get electricity from point A to point B in an acceptable fashion. But on the other hand people will not trust it, including planners. I am sorry—I am trying to illustrate a complex system. How on earth can you translate that into something that you can put on paper and try to control? You cannot, and that is what people have to realise. It is an interesting world.

I am glad that the Bill is coming along, because it is high time we started thinking about this and what we expect we can do about it. It is also transnational—it goes right across all borders—so we cannot regulate in isolation. In this new interconnected and networked world, we cannot have a little isolated island in the middle of it all where we can control it—that is just not going to happen. Anyway, we live in very interesting times.

11:33

[Lord Kirkhope of Harrogate (Con)]

My Lords, as has been illustrated this morning, we stand on the cusp of a technological revolution. We find ourselves at the crossroads between innovation and responsibility. Artificial intelligence, a marvel of modern science, promises to reshape the world. Yet with great power comes great responsibility, and it is therefore imperative that we approach this with caution. Regulation in the realm of AI is not an adversary to innovation; rather, it is the very framework within which responsible and sustainable innovation must occur. Our goal should not be to stifle the creative spirit but to channel it, ensuring that it serves the common good while safeguarding our societal values and ethical standards.

However, we must not do this in isolation. In the digital domain, where boundaries blur, international collaboration becomes not just beneficial but essential. The challenges and opportunities presented by AI do not recognise national borders, and our responses too must be global in perspective. The quest for balance in regulation must be undertaken with a keen eye on international agreements, ensuring that the UK remains in step with the global community, not at odds with it. In our pursuit of this regulatory framework suitable for the UK, we must consider others. The European Union’s AI Act, authored by German MEP Axel Voss, offers valuable insights and, by examining what works within the EU’s and other approaches, as well as identifying areas for improvement, we can learn from the experiences of our neighbours to forge a path that is distinctly British, yet globally resonant.

Accountability stands as a cornerstone in the responsible deployment of AI technologies. Every algorithm and every application that is released into the world must have a clearly identifiable human or corporate entity behind it. This is where the regulatory approach must differ to that inherent in the general data protection regulations, which I had the pleasure of helping to formulate in Brussels. This accountability is crucial for ethical, legal and social reasons, ensuring that there is always a recourse and a responsible party when AI systems interact with our world.

Yet, as we delve into the mechanics of regulation and oversight, we must also pause to reflect on the quintessentially human aspect of our existence that AI can never replicate: emotion. The depth and complexity of emotions that define our humanity remain beyond the realm of AI and always will. These elements, intrinsic to our being, highlight the irreplaceable value of the human touch. While AI can augment, it can never replace human experience. The challenge before us is to foster an environment where innovation thrives within a framework of ethical and responsible governance. We must be vigilant not to become global enforcers of compliance at the expense of being pioneers of innovation.

The journey we embark on with the regulation of AI is not one that ends with the enactment of laws; that is merely the beginning. The dynamic nature of AI demands that our regulatory frameworks be agile and capable of adapting to rapid advancements and unforeseen challenges. So, as I have suggested on a number of occasions, we need smart legislation—a third tier of legislation behind the present primary and secondary structures—to keep up with these things.

In the dynamic landscape of AI, the concept of sandboxes stands out as a forward-thinking approach to innovation in this field. This was referred to by my noble friend in introducing his Bill. They offer a controlled environment where new technologies can be tested and refined without the immediate pressures and risks associated with full-scale deployment.

I emphasise that support for small and medium-sized enterprises in navigating the regulatory landscape is of paramount importance. These entities, often the cradles of innovation, must be equipped with the tools and knowledge to flourish within the bounds of regulation. The personnel in our regulatory authorities must also be of the highest calibre—individuals who not only comprehend the technicalities of AI but appreciate its broader implications for society and the economy.

At this threshold of a new era shaped by AI, we should proceed with caution but also with optimism. Let us never lose sight of the fact that at the heart of all technological advancement lies the indomitable spirit of human actions and emotions, which no machine or electronic device can create alone. I warmly welcome my noble friend Lord Holmes’s Bill, which I will fully support throughout its process in this House.

11:38

[Baroness Finlay of Llandaff (CB)]

My Lords, I am most grateful to the noble Lord, Lord Holmes, for the meeting he arranged to explain his Bill in detail and to answer some of the more naive questions from some Members of this House. Having gone through the Bill, I cannot see how we can ignore the importance of this, going forwards. I am also grateful to my noble friend Lady Kidron for the meeting that she established, which I think educated many of us on the realities of AI.

I want to focus on the use of AI in medicine because that is my field. The New England Journal of Medicine has just launched NEJM AI as a new journal to collate what is happening. AI use is becoming widespread but across the NHS tends to be small-scale. People hope that AI will streamline administrative tasks which are burdensome, improve communication with patients and do even simple things such as making out-patient appointments more suitable for people and arranging transport better.

For any innovations to be used, however, the infrastructure needs to be set up. I was struck yesterday at a meeting on emergency medicine where the consultant explained that it now takes longer to enter patient details in the computer system than it used to using old-fashioned pen and paper—the reason being that the computer terminals are not in the room where the consultation is happening so it takes people away.

A lot of people in medicine are tremendously enthusiastic—we see the benefits in diagnostics for images of cells and X-rays and so on—but there has to be reliability. Patients and people in the general population are buying different apps to diagnose things such as skin cancers, but the reliability of these apps is unproven. What we need is the use of AI to improve diagnostic accuracy. Currently, post-mortems show about a 5% error in what is written on the death certificate; in other words, at post-mortem, people are found to have died of something different from the disease or condition they were being treated for. So we have to improve diagnostics right across the piece.

But the problem is that we have to put that in a whole system. The information that goes in to train and teach these diagnostic systems has to be of very high quality, and we need audit in there to make sure that high quality is maintained. Although we are seeing fantastic advances in images such as mammograms, supporting radiologists, and in rapid diagnosis of strokes and so on, there is a need to ensure quality control in the system, so that it does not go wild on its own, that the input is being monitored and, as things change in the population, that that change is also being detected.

Thinking about this Bill, I reflected on the Covid experience, when the ground-glass appearance on X-rays was noted to be somewhat unique and new in advanced Covid lung disease. We have a fantastic opportunity, if we could use all of that data properly, to follow up people in the long term, to see how many have got clotting problems and how many later go on to develop difficulties or other conditions of which we have been unaware. There could be a fantastic public health benefit if we use the technology properly.

The problem is that, if it is not used properly, it will lose public trust. I noted that, in his speech introducing the Bill, the noble Lord, Lord Holmes, used the word “trust” very often. It seems that a light-touch regulator that goes across many domains and areas is what we will need. It will protect copyright, protect the intellectual property rights of the people who are developing systems, keep those investments in the UK and foster innovation in the UK. Unless we do that, and unless we develop trust across the board, we will fail in our developments in the longer term. The real world that we live in today has to be safe, and the world that AI takes us into has to be safe.

I finish with a phrase that I have often heard resonate in my ears:

“Trust arrives on foot and leaves on horseback”.


We must not let AI be the horses that take all the trust in the developments away.

11:44

[Lord Ranger of Northwood (Con)]

My Lords, are we ready for the power of artificial intelligence? With each leap in human ability to invent and change what we can achieve, we have utilised a new power, a new energy that has redefined the boundaries of imagination: steam and the Industrial Revolution; electricity and the age of light; and so, again, we stand on the precipice of another seismic leap.

However, the future of AI is not just about what we can do with it but about who will have access to control its power. So I welcome the attempt made by my noble friend Lord Holmes via this Bill to encourage an open public debate on democratic oversight of AI, but I do have some concerns. Our view of AI at this early stage is heavily coloured by how this power will deliver automation and the potential reduction of process-reliant jobs and how those who hold the pen on writing the algorithms behind AI could exert vast power and influence on the masses via media manipulation. We fear that the AI genie is out of the bottle and we may not be able to control it. The sheer, limitless potential of AI is intimidating.

If, like me, you are from a certain generation, these seeds of fear and fascination at the power of artificial intelligence have long been planted by numerous Hollywood movies picking on our hopes, dreams and fears of what AI could do to us. Think of the unnerving subservience of HAL in Stanley Kubrick’s “2001: A Space Odyssey” made in 1968, the menacing and semi-obedient robot Maximilian from the 1979 Disney production “The Black Hole”, the fantasy woman called Lisa created by the power of 80s home computing in “Weird Science” from 1985, and, of course, the ultimate hellish future of machine intelligence taking over the world in the form of Skynet in “The Terminator” made in 1984. These and many other futuristic interpretations of AI helped to fan the flames in the minds of engineers, computer scientists and super-geeks, many of whom created and now run the biggest tech firms in the world.

But where are we now? The advancement in processing power, coupled with vast amounts of big data and developments such as large language models, have led to the era of commercialisation of AI. Dollops of AI are available in everyday software programmes via chatbots and automated services. Obviously, the emergence of ChatGPT turbocharged the public awareness and usage of the technology. We have poured algorithms into machines and made them “think”. We have stopped prioritising trying to get robots to look and feel like us, and focused instead on the automation of systems and processes, enabling them to do more activities. We have moved from the pioneering to the application era of AI.

With all this innovation, with so many opportunities and benefits to be derived by its application, what should we fear? My answer is not from the world of Hollywood science fiction; it relates not to individuals losing control to machines but, rather, to how we will ensure that this power remains democratic and accessible and benefits the many. How will we ensure that control does not fall into the hands of the few, that wealth does not determine the ability to benefit from innovation and that a small set of organisations do not gain ultimate global control or influence over our lives? How, also, will we ensure that Governments and bureaucracies do not end up ever furthering the power and control of the state through well-intentioned regulatory control? This is why we must appreciate the size of this opportunity, think about the long-term future, and start to design the policy frameworks and new public bodies that will work in tandem with those who will design and deliver our future world.

But here is the rub: I do not believe we can control, manage or regulate this technology through a single authority. I am extremely supportive of the ambitions of my noble friend Lord Holmes to drive this debate. However, I humbly suggest that the question we need to focus on will be how we can ensure that the innovations, outcomes and quality services that AI delivers are beneficial and well understood. The Bill as it stands may be overambitious for the scope of this AI authority: to act as oversight across other regulators; to assess safety, risks and opportunities; to monitor risks across the economy; to promote interoperability and regulatory frameworks; and to act as an incubator to innovation. To achieve this and more, the AIA would need vast cross-cutting capability and resources. Again, I appreciate what my noble friend Lord Holmes is trying to achieve and, as such, I would say that we need to consider with more focus the questions that we are trying to answer.

I wholeheartedly believe and agree that the critical role will be to drive public education, engagement and awareness of AI, and where and how it is used, and to clearly identify the risks and benefits to the end-users, consumers, customers and the broader public. However, I strongly suggest that we do not begin this journey by requiring labelling, under Clause 5(1)(a)(iii), using “unambiguous health warnings” on AI products or services. That would not help us to work hand in hand with industry and trade bodies to build trust and confidence in the technology.

I believe there will eventually be a need for some form of future government body to help provide guidance to both industry and the public about how AI outcomes, especially those in delivering public sector services, are transparent, fair in design and ethical in approach. Such a body will need to take note of the approach of other nations and will need to engage with local and global businesses to test and formulate the best way forward. So, although I am sceptical of many of the specifics of the Bill, I welcome and support the journey that it, my noble friend Lord Holmes and this debate are taking us on.

11:50

[Lord Clement-Jones (LD)]

My Lords, I congratulate the noble Lord, Lord Holmes, on his inspiring introduction and on stimulating such an extraordinarily good and interesting debate.

The excellent House of Lords Library guide to the Bill warns us early on:

“The bill would represent a departure from the UK government’s current approach to the regulation of AI”.


Given the timidity of the Government’s pro-innovation AI White Paper and their response, I would have thought that was very much a “#StepInTheRightDirection”, as the noble Lord, Lord Holmes, might say.

There is clearly a fair wind around the House for the Bill, and I very much hope it progresses and we see the Government adopt it, although I am somewhat pessimistic about that. As we have heard in the debate, there are so many areas where AI is and can potentially be hugely beneficial, despite the rather dystopian narratives that the noble Lord, Lord Ranger, so graphically outlined. However, as many noble Lords have emphasised, it also carries risks, not just of the existential kind, which the Bletchley Park summit seemed to address, but others mentioned by noble Lords today, such as misinformation, disinformation, child sexual abuse, and so on, as well as the whole area of competition, mentioned by the noble Lord, Lord Fairfax, and the noble Baroness, Lady Stowell—the issue of the power and the asymmetry of these big tech AI systems and the danger of regulatory capture.

It is disappointing that, after a long gestation of national AI policy-making, which started so well back in 2017 with the Hall-Pesenti review, contributed to by our own House of Lords Artificial Intelligence Committee, the Government have ended up by producing a minimalist approach to AI regulation. I liked the phrase used by the noble Lord, Lord Empey, “lost momentum”, because it certainly feels like that after this period of time.

The UK’s National AI Strategy, a 10-year plan for UK investment in and support of AI, was published in September 2021 and accepted that in the UK we needed to prepare for artificial general intelligence. We needed to establish public trust and trustworthy AI, so often mentioned by noble Lords today. The Government had to set an example in their use of AI and to adopt international standards for AI development and use. So far, so good. Then, in the subsequent AI policy paper, AI Action Plan, published in 2022, the Government set out their emerging proposals for regulating AI, in which they committed to develop

“a pro-innovation national position on governing and regulating AI”,

to be set out in a subsequent governance White Paper. The Government proposed several early cross-sectoral and overarching principles that built on the OECD principles on artificial intelligence: ensuring safety, security, transparency, fairness, accountability and the ability to obtain redress.

Again, that is all good, but the subsequent AI governance White Paper in 2023 opted for a “context-specific approach” that distributes responsibility for embedding ethical principles into the regulation of AI systems across several UK sector regulators without giving them any new regulatory powers. I thought the analysis of this by the noble Lord, Lord Young, was interesting. There seemed to be no appreciation that there were gaps between regulators. That approach was confirmed this February in the response to the White Paper consultation.

Although there is an intention to set up a central body of some kind, there is no stated lead regulator, and the various regulators are expected to interpret and apply the principles in their individual sectors in the expectation that they will somehow join the dots between them. There is no recognition that the different forms of AI are technologies that need a comprehensive cross-sectoral approach to ensure that they are transparent, explainable, accurate and free of bias, whether they are in an existing regulated or unregulated sector. As noble Lords have mentioned, discussing existential risk is one thing, but going on not to regulate is quite another.

Under the current Data Protection and Digital Information Bill, data subject rights regarding automated decision-making—in practice, by AI systems—are being watered down, while our creatives and the creative industries are up in arms about the lack of support from government in asserting their intellectual property rights in the face of the ingestion of their material by generative AI developers. It was a pleasure to hear what the noble Lord, Lord Freyberg, had to say on that.

For me, the cardinal rules are that business needs clarity, certainty and consistency in the regulatory system if it is to develop and adopt AI systems, and we need regulation to mitigate risk to ensure that we have public trust in AI technology. As the noble Viscount, Lord Chandos, said, regulation is not necessarily the enemy of innovation; it can be a stimulus. That is something that we need to take away from this discussion. I was also very taken with the idea of public trust leaving on horseback.

This is where the Bill of the noble Lord, Lord Holmes, is an important stake in the ground, as he has described. It provides for a central AI authority that has a duty of looking for gaps in regulation; it sets out extremely well out the safety and ethical principles to be followed; it provides for regulatory sandboxes, which we should not forget are an innovation invented in the UK; and it provides for AI responsible officers and for public engagement. Importantly, it builds in a duty of transparency regarding data and IP-protected material where they are used for training purposes, and for labelling AI-generated material, as the noble Baroness, Lady Stowell, and her committee have advocated. By itself, that would be a major step forward, so, as the noble Lord knows, we on these Benches wish the Bill very well, as do all those with an interest in protecting intellectual property, as we heard the other day at the round table that he convened.

However, in my view what is needed at the end of the day is the approach that the interim report of the Science, Innovation and Technology Committee recommended towards the end of last year in its inquiry into AI governance: a combination of risk-based cross-sectoral regulation and specific regulation in sectors such as financial services, applying to both developers and adopters, underpinned by common trustworthy standards of risk assessment, audit and monitoring. That should also provide recourse and redress, as the Ada Lovelace Institute, which has done so much work in the area, asserts, and as the noble Lord, Lord Kirkhope, mentioned.

That should include the private sector, where there is no effective regulator for the workplace, as the noble Lord, Lord Davies, mentioned, and the public sector, where there is no central or local government compliance mechanism; no transparency yet in the form of a public register of use of automated decision-making, despite the promised adoption of the algorithmic recording standard; and no recognition by the Government that explicit legislation and/or regulation for intrusive AI technologies used in the public sector, such as live facial recognition and other biometric capture, is needed. Then, of course, we need to meet the IP challenge. We need to introduce personality rights to protect our artists, writers and performers. We need the labelling of AI-generated material alongside the kinds of transparency duties contained in the noble Lord’s Bill.

Then there is another challenge, which is more international. This was mentioned by the noble Lords, Lord Kirkhope and Lord Young, the noble and learned Lord, Lord Thomas of Cwmgiedd, and the noble Earl, Lord Erroll. We have world-beating AI researchers and developers. How can we ensure that, despite differing regulatory regimes—for instance, between ourselves and the EU or the US—developers are able to commercialise their products on a global basis and adopters can have the necessary confidence that the AI product meets ethical standards?

The answer, in my view, lies in international agreement on common standards such as those of risk and impact assessment, testing, audit, ethical design for AI systems, and consumer assurance, which incorporate what have become common internationally accepted AI ethics. Having a harmonised approach to standards would help provide the certainty that business needs to develop and invest in the UK more readily, irrespective of the level of obligation to adopt them in different jurisdictions and the necessary public trust. In this respect, the UK has the opportunity to play a much more positive role with the Alan Turing Institute’s AI Standards Hub and the British Standards Institution. The OECD.AI group of experts is heavily involved in a project to find common ground between the various standards.

We need a combination of proportionate but effective regulation in the UK and the development of international standards, so, in the words of the noble Lord, Lord Holmes, why are we not legislating? His Bill is a really good start; let us build on it.

12:01

[Baroness Twycross (Lab)]

My Lords, like others, I congratulate the noble Lord, Lord Holmes of Richmond, on his Private Member’s Bill, the Artificial Intelligence (Regulation) Bill. It has been a fascinating debate and one that is pivotal to our future. My noble friend Lord Leong apologises for his absence and I am grateful to the Government Benches for allowing me, in the absence of an AI-generated hologram of my noble friend, to take part in this debate. If the tone of my comments is at times his, that is because my noble friend is supremely organised and I will be using much of what he prepared for this debate. Like the noble Lord, Lord Young, I am relying heavily on osmosis; I am much more knowledgeable on this subject now than two hours ago.

My first jobs were reliant on some of the now-defunct technologies, although I still think that one of the most useful skills I learned was touch-typing. I learned that on a typewriter, complete with carbon paper and absolutely no use of Tipp-Ex allowed. However, automation and our continued and growing reliance on computers have improved many jobs rather than simply replacing them. AI can help businesses save money and increase productivity by adopting new technologies; it can also release people from repetitive data-entry tasks, enabling them to focus on creative and value-added tasks. New jobs requiring different skills can be created and, while this is not the whole focus of the debate, how we achieve people being able to take up new jobs also needs to be a focus of government policy in this area.

As many noble Lords have observed, we stand on the brink of an AI revolution, one that has already started. It is already changing the way we live, the way we work and the way we relate to one another. I count myself in the same generation of film viewers as the noble Lord, Lord Ranger. The rapidly approaching tech transformation is unlike anything that humankind has experienced in its speed, scale and scope: 20th-century science fiction is becoming commonplace in our 21st-century lives.

As the noble Baroness, Lady Moyo, said, it is estimated that AI technology could contribute up to £15 trillion to the world economy by 2030. As many noble Lords mentioned, AI also presents government with huge opportunities to transform public services, potentially delivering billions of pounds in savings and increasing the service to the public. For example, it could help with the workforce crisis in health, particularly in critical health diagnostics, as highlighted by the noble Lord, Lord Empey. The noble Baroness, Lady Finlay, highlighted the example of how diagnosis of Covid lung has benefited through the use of AI, but, as she said, that introduces requirements for additional infrastructure. My noble friend Lord Davies also noted that AI can help to contribute to how we tackle climate change.

The use of AI by government underpins Labour’s missions to revive our country’s fortunes and ensure that the UK thrives and is at the forefront of the coming technological revolution. However, we should not and must not overlook the risks that may arise from its use, nor the unease around AI and the lack of confidence among the public around its use. Speaking as someone who generally focuses on education from these Benches, this is not least in the protection of children, as the noble Baroness, Lady Kidron, pointed out. AI can help education in a range of ways, but these also need regulation. As the noble Baroness said, we need rules to defend against the potential abuses.

Goldman Sachs predicts that the equivalent of 300 million full-time jobs globally will be replaced; this includes around a quarter of current work tasks in the US and Europe. Furthermore, as has been noted, AI can damage our physical and mental health. It can infringe upon individual privacy and, if not protected against, undermine human rights. Our collective response to these concerns must be as integrated and comprehensive as our embracing of the potential benefits. It should involve all stakeholders, from the public and private sectors to academia and civil society. Permission should and must be sought by AI developers for the use of copyright-protected work, with remuneration and attribution provided to creators and rights holders, an issue highlighted by the noble Lord, Lord Freyberg. Most importantly, transparency needs to be delivered on what content is used to train generative AI models. I found the speech of the noble Earl, Lord Erroll, focusing on outcomes, of particular interest.

Around the world, countries and regions are already beginning to draft rules for AI. As the noble Lord, Lord Kirkhope, said, this does not need to stifle innovation. The Government’s White Paper on AI regulation adopted a cross-sector and outcome-based framework, underpinned by its five core principles. Unfortunately, there are no proposals in the current White Paper for introducing a new AI regulator to oversee the implementation of the framework. Existing regulators, such as the Information Commissioner’s Office, Ofcom and the FCA have instead been asked to implement the five principles from within their respective domains. As a number of noble Lords referred to, the Ada Lovelace Institute has expressed concern about the Government’s approach, which it has described as “all eyes, no hands”. The institute says that, despite

“significant horizon-scanning capabilities to anticipate and monitor AI risks … it has not given itself the powers and resources to prevent those risks or even react to them effectively after the fact”.

The Bill introduced by the noble Lord, Lord Holmes, seeks to address these shortcomings and, as he said in his opening remarks: if not now, when? Until such time as an independent AI regulator is established, the challenge lies in ensuring its effective implementation across various regulatory domains. This includes data protection, competition, communications and financial services. A number of noble Lords mentioned the multitude of regulatory bodies involved. This means that effective governance between them will be paramount. Regulatory clarity, which enables business to adopt and scale investment in AI, will bolster the UK’s competitive edge. The UK has so far been focusing on voluntary measures for general-purpose AI systems. As the right reverend Prelate the Bishop of Worcester said, this is not adequate: human rights and privacy must also be protected.

The noble Lord, Lord Kirkhope, noted that AI does not respect national borders. A range of international approaches to AI safety and governance are developing, some of which were mentioned by the noble Lord, Lord Fairfax. The EU has opted for a comprehensive and prescriptive legislative approach; the US is introducing some mandatory reporting requirements; for example, for foundation models that pose serious national or economic security risks.

Moreover, a joint US-EU initiative is drafting a set of voluntary rules for AI businesses—the AI code of conduct. In the short term, these may serve as de facto international standards for global firms. Can the Minister tell your Lordships’ House whether the Government are engaging with this drafting? The noble Lord, Lord Empey, suggested that the Government have lost momentum. Can the Minister explain why the Government are allowing the UK to lose influence over the development of international AI regulation?

The noble Lord, Lord Clement-Jones, noted that the Library briefing states that this Bill marks a departure from government approach. The Government have argued that introducing legislation now would be premature and that the risks and challenges associated with AI, the regulatory gaps and the best way to address them must be better understood. This cannot be the case. Using the horse analogy adopted by the noble Baroness earlier, we need to make sure that we do not act after the horse has bolted.

I pay tribute, as others have done, to the work of the House of Lords Communications and Digital Committee. I found the points highlighted by its chair and her comments very helpful. We are facing an inflection point with AI. It is regrettable that the government response is not keeping up with the change. Why are the Government procrastinating while all other G7 members are adopting a different, more proactive approach? A Labour Government would act decisively and not delay. Self-regulation is simply not enough.

The honourable Member for Hove, the shadow Secretary of State for Science, Innovation and Technology, outlined Labour’s plans recently at techUK’s annual conference. He said:

“Businesses need fast, clear and consistent regulation … that … does not unnecessarily slow down innovation”—


a point reflected in comments by the noble and learned Lord, Lord Thomas. We also need regulation that encourages risk taking and finding new ways of working. We need regulation that addresses the concerns and protects the privacy of the public.

As my noble friend Lord Chandos said, the UK also needs to address concerns about misinformation and disinformation, not least in instances where these are democratic threats. This point was also reflected by the noble Lords, Lord Vaizey and Lord Fairfax.

Labour’s regulatory innovation office would give strategic steers aligned with our industrial strategy. It would set and monitor targets on regulatory approval timelines, benchmark against international comparators and strengthen the work done by the Regulatory Horizons Council. The public need to know that safety will be baked into how AI is used by both the public and the private sectors. A Labour Government would ensure that the UK public sector is a leader in responsibly and transparently applying AI. We will require safety reports from the companies developing frontier AI. We are developing plans to make sure that AI works for everyone.

Without clear regulation, widespread business adoption and public trust, the UK’s adoption of AI will be too slow. It is the Government’s responsibility to acknowledge and address how AI affects people’s jobs, lives, data and privacy, and the rapidly changing world in which they live. The Government are veering haphazardly between extreme risk, extreme optimism and extreme delay on this issue. Labour is developing a practical, well-informed and long-term approach to regulation.

In the meantime, we support and welcome the principles behind the Private Member’s Bill from the noble Lord, Lord Holmes, but remain open-minded on the current situation and solution, while acknowledging that there is still much more to be done.

12:14

[The Parliamentary Under-Secretary of State, Department for Science, Innovation and Technology (Viscount Camrose) (Con)]

I join my thanks to those of others to my noble friend Lord Holmes for bringing forward this Bill. I thank all noble Lords who have taken part in this absolutely fascinating debate of the highest standard. We have covered a wide range of topics today. I will do my best to respond, hopefully directly, to as many points as possible, given the time available.

The Government recognise the intent of the Bill and the differing views on how we should go about regulating artificial intelligence. For reasons I will now set out, the Government would like to express reservations about my noble friend’s Bill.

First, with the publication of our AI White Paper in March 2023, we set out proposals for a regulatory framework that is proportionate, adaptable and pro-innovation. Rather than designing a new regulatory system from scratch, the White Paper proposed five cross-sectoral principles, which include safety, transparency and fairness, for our existing regulators to apply within their remits. The principles-based approach will enable regulators to keep pace with the rapid technological change of AI.

The strength of this approach is that regulators can act now on AI within their own remits. This common-sense, pragmatic approach has won endorsement from leading voices across civil society, academia and business, as well as many of the companies right at the cutting edge of frontier AI development. Last month we published an update through the Government’s response to the consultation on the AI White Paper. The White Paper response outlines a range of measures to support existing regulators to deliver against the AI regulatory framework. This includes providing further support to regulators to deliver the regulatory framework through a boost of more than £100 million to upskill regulators and help unlock new AI research and innovation.

As part of this, we announced a £10 million package to jump-start regulators’ AI capabilities, preparing and upskilling regulators to address the risks and to harness the opportunities of this defining technology. It also includes publishing new guidance to support the coherent implementation of the principles. To ensure robust implementation of the framework, we will continue our work to establish the central function.

Let me reassure noble Lords that the Government take mitigating AI risks extremely seriously. That is why several aspects of the central function have already been established, such as the central AI risk function, which will shortly be consulting on its cross-economy AI risk register. Let me reassure the noble Lord, Lord Empey, that the AI risk function will maintain a holistic view of risks across the AI ecosystem, including misuse risks, such as where AI capabilities may be leveraged to undermine cybersecurity.

Specifically on criminality, the Government recognise that the use of AI in criminal activity is a very important issue. We are working with a range of stakeholders, including regulators, and a range of legal experts to explore ways in which liability, including criminal liability, is currently allocated through the AI value chain.

In the coming months we will set up a new steering committee, which will support and guide the activities of a formal regulator co-ordination structure within government. We also wrote to key regulators, requesting that they publish their AI plans by 30 April, setting out how they are considering, preparing for and addressing AI risks and opportunities in their domain.

As for the next steps for ongoing policy development, we are developing our thinking on the regulation of highly capable general-purpose models. Our White Paper consultation response sets out key policy questions related to possible future binding measures, which we are exploring with experts and our international partners. We plan to publish findings from this expert engagement and an update on our thinking later this year.

We also confirmed in the White Paper response that we believe legislative action will be required in every country once the understanding of risks from the most capable AI systems has matured. However, legislating too soon could easily result in measures that are ineffective against the risks, are disproportionate or quickly become out of date.

Finally, we make clear that our approach is adaptable and iterative. We will continue to work collaboratively with the US, the EU and others across the international landscape to both influence and learn from international development.

I turn to key proposals in the Bill that the noble Lord has tabled. On the proposal to establish a new AI authority, it is crucial that we put in place agile and effective mechanisms that will support the coherent and consistent implementation of the AI regulatory framework and principles. We believe that a non-statutory central function is the most appropriate and proportionate mechanism for delivering this at present, as we observe a period of non-statutory implementation across our regulators and conduct our review of regulator powers and remits.

In the longer term, we recognise that there may be a case for reviewing how and where the central function has delivered, once its functions have become more clearly defined and established, including whether the function is housed within central government or in a different form. However, the Government feel that this would not be appropriate for the first stage of implementation. To that end, as I mentioned earlier, we are delivering the central function within DSIT, to bring coherence to the regulatory framework. The work of the central function will provide clarity and ensure that the framework is working as intended and that joined-up and proportionate action can be taken if there are gaps in our approach.

We recognise the need to assess the existing powers and remits of the UK’s regulators to ensure they are equipped to address AI risks and opportunities in their domains and to implement the principles consistently and comprehensively. We anticipate having to introduce a statutory duty on regulators requiring them to have due regard to the principles after an initial period of non-statutory implementation. For now, however, we want to test and iterate our approach. We believe this approach offers critical adaptability, but we will keep it under review; for example, by assessing the updates on strategic approaches to AI that several key regulators will publish by the end of April. We will also work with government departments and regulators to analyse and review potential gaps in existing regulatory powers and remits.

Like many noble Lords, we see approaches such as regulatory sandboxes as a crucial way of helping businesses navigate the AI regulatory landscape. That is why we have funded the four regulators in the Digital Regulation Cooperation Forum to pilot a new, multiagency advisory service known as the AI and digital hub. We expect the hub to launch in mid-May and will provide further details in the coming weeks on when this service will be open for applications from innovators.

One of the principles at the heart of the AI regulatory framework is accountability and governance. We said in the White Paper that a key part of implementation of this principle is to ensure effective oversight of the design and use of AI systems. We have recognised that additional binding measures may be required for developers of the most capable AI systems and that such measures could include requirements related to accountability. However, it would be too soon to mandate measures such as AI-responsible officers, even for these most capable systems, until we understand more about the risks and the effectiveness of potential mitigations. This could quickly become burdensome in a way that is disproportionate to risk for most uses of AI.

Let me reassure my noble friend Lord Holmes that we continue to work across government to ensure that we are ready to respond to the risks to democracy posed by deep fakes; for example, through the Defending Democracy Taskforce, as well as through existing criminal offences that protect our democratic processes. However, we should remember that AI labelling and identification technology is still at an early stage. No specific technology has yet been proven to be both technically and organisationally feasible at scale. It would not be right to mandate labelling in law until the potential benefits and risks are better understood.

Noble Lords raised the importance of protecting intellectual property, a profoundly important subject. In the AI White Paper consultation response, the Government committed to provide an update on their approach to AI and copyright issues soon. I am confident that, when we do so, it will address many of the issues that noble Lords have raised today.

In summary, our approach, combining a principles-based framework, international leadership and voluntary measures on developers, is right for today, as it allows us to keep pace with rapid and uncertain advances in AI. The UK has successfully positioned itself as a global leader on AI, in recognition of the fact that AI knows no borders and that its complexity demands nuanced international governance. In addition to spearheading thought leadership through the AI Safety Summit, the UK has supported effective action through the G7, the Council of Europe, the OECD, the G5, the G20 and the UN, among other bodies. We look forward to continuing to engage with all noble Lords on these critical issues as we continue to develop our regulatory approach.

12:25

[Lord Holmes of Richmond (Con)]

My Lords, I thank all noble Lords who have contributed to this excellent debate. It is pretty clear that the issues are very much with us today and we have what we need to act today. To respond to a question kindly asked by the noble Lord, Lord Davies of Brixton, in my drafting I am probably allowing “relevant” regulators to do some quite heavy lifting, but what I envisage within that is certainly all the economic regulators, and indeed all regulators who are in a sector where AI is being developed, deployed and in use. Everybody who has taken part in this debate and beyond may benefit from having a comprehensive list of all the regulators across government. Perhaps I could ask that of the Minister. I think it would be illuminating for all of us.

At the autumn FT conference, my noble friend the Minister said that heavy-handed regulation could stifle innovation. Certainly, it could. Heavy-handed regulation would not only stifle innovation but would be a singular failure of that creation of the regulatory process. History tells us that right-size regulation is pro-citizen, pro-consumer and pro-innovation; it drives innovation and inward investment. I was taken by so much of what the Ada Lovelace Institute put in its report. The Government really have given themselves all the eyes and not the hands to act. It reminds me very much of a Yorkshire saying: see all, hear all, do nowt. What is required is for these technologies to be human led, in our human hands, and human in the loop throughout. Right-size regulation, because it is principles-based, is necessarily agile, adaptive and can move as the technology moves. It should be principles-based and outcomes focused, with inputs that are transparent, understood, permissioned and, wherever and whenever applicable, paid for.

My noble friend the Minister has said on many occasions that there will come a time when we will legislate on AI. Let 22 March 2024 be that time. It is time to legislate; it is time to lead.

Bill read a second time and committed to a Committee of the Whole House.