That this House takes note of the ongoing development of advanced artificial intelligence, associated risks and potential approaches to regulation within the UK and internationally.
My Lords, I first declare my interest as a project director working for Atkins and by noting that this is not just my debate: a number of noble Lords right across the Cross Benches put forward submissions on this topic, including the noble Baroness, Lady Kidron, the noble and right reverend Lord, Lord Harries, and the noble Lord, Lord Patel.
There are a couple of reasons why I was keen to put this forward. First, as we have seen recently, the rapid advancement of AI has been brought into sharp relief by the ongoing development of large language models, some of which are likely to be able to pass the famous Turing test for machine intelligence, which has been something of a benchmark for it over the past 50 years. The questions of existential risk of that technology have also resurfaced. AI is the most important technology of our generation. It will cause significant upheaval right across the economy, good and bad. We in Parliament need to be thinking about this area and talking about it a lot more than we are. It should be right at the top of our agenda. Secondly, there is the matter of timing. The Government released their White Paper earlier this year but, in some respects, this has been overtaken by events and the Government appear to be rethinking aspects of their strategy. Therefore, this is the perfect time for this House to express its views on the issues before us, to help inform forthcoming government responses.
In my work as an engineering consultant, I have worked on and seen the continued advancement of these technologies over the years. Several years ago, one of my projects involved the transfer of large amounts of engineering data—the complete design definition of a nuclear reactor, hundreds of thousands of documents —from an old to a new computer system. A proposal was developed to get eight graduate engineers sitting at desks and manually transferring the data, which would have been a terrible waste of talent. We sat down with our brightest young engineers, and they investigated and developed a smart algorithm which worked in the graphical user interface, just as a human would. It was effectively a software robot to undertake this work and replace human workers. This was able to crunch through the entire task in minimal time—a matter of months—saving hundreds of thousands of pounds and thousands of hours of engineering effort.
Across the industry, we are starting to see automation coupled with AI and machine learning to learn how to resolve discrepancies and data from past experience, continuing that process of freeing up humans in clerical jobs for more value-added work. This is one example of the huge benefits that the AI revolution is having and will continue to have on society. Anyone who has read Adam Smith’s The Wealth of Nations and his description of the pin factory sees the logic and economic benefits of increasing specialisation—but also the drudgery of work that can result from that. Among many other benefits, AI will continue that process of freeing people up from repetitive, inane tasks and on to more value-added work, increasing human happiness along with it.
My Lords, I first congratulate the noble Lord, Lord Ravensdale, on securing this debate and on the comprehensive and interesting way he has introduced it. I signed up to speak for two reasons: first, because I thought I might learn something; and secondly, because I thought it would be helpful for me to highlight that the Communications and Digital Select Committee of your Lordships’ House, which I have the great privilege to chair, has recently launched an inquiry into large language models, focusing on how we can capitalise on the opportunities while managing the risks.
I am under no illusion: the latest advances in generative AI are significant, but we must not allow scaremongering about the future to be a distraction from today’s opportunities and risks. In the committee’s view, what is most important at the moment is to separate hype from reality and make a considered assessment of what guardrails and controls are needed now.
When we come back in September, we will take a detailed look at how large language models are expected to develop over the next three years, how well those changes are accounted for by the Government’s White Paper and our existing regulators, and what needs to happen to capitalise on the benefits and address the most pressing risks. That will include close examination of the structure, work and capacity of the regulators and government teams and their ability to deliver on the White Paper’s expectations. We are open for written submissions and are currently inviting witnesses. We intend to hear from a wide range of key players—from the big tech platforms and fast-moving start-ups to academics, industry experts, regulators, government advisers and institutions abroad.
A key part of our work will be to demystify some of the issues and make sure we are not blinded by the rosy outlook that tech firms are proposing or by doom-saying about the imminent collapse of civilisation. I do not know about noble Lords, but I cannot help thinking how convenient it is to the big tech bros that so few people understand what is going on, so we are going to try to change that through our inquiry. This is not just to mitigate anything bad happening that we do not know about, but to make sure that all the power is not concentrated in a few people’s hands and that the many exciting, potential opportunities of this technology are available not only to them.
My Lords, it is a distinct pleasure to follow the noble Baroness, Lady Stowell of Beeston. I associate myself with her words of commendation and congratulation to the noble Lord, Lord Ravensdale; it is entirely appropriate that this debate be led by someone with the lived experience of an engineer, and in the noble Lord we have found that person.
Mindful of time, I will limit myself to asking a few diagnostic questions with which I hope the Minister will engage. I believe that they raise issues which are essential to harnessing the benefits of AI, if it is to be done in a manner which is both sustainable and enjoys public consent and trust. Using the incremental process of legislation to govern a technology characterised by chronic exponential technological leaps is not easy. Though tempting, the answer is not to oscillate between the poles of a false dichotomy, with regulatory rigour on one side and innovation on the other. Like climate change, AI is a potentially existential risk that is chartered by ever deepening scientific understanding, emerging opportunity and emerging demonstrable risks.
It is not always true that an absence of government means liberation for business or innovation, especially when business and innovation know that more comprehensive regulation is on the horizon. Clear signals from the Government of regulation, even in advance of that legislation, will not inhibit the AI sector but give it greater confidence in planning, resourcing and pursuing technological advances. My first question is: given that the Prime Minister last month announced his intention for the UK to play a role in leading the world in AI regulation, how does he plan to shape an international legal framework when our own is still largely hypothetical? When do the Government plan to devote parliamentary time to bringing forward some instrument or statement which will deal squarely with the future direction of domestic AI regulation? The President of the United States seems to be doing this these days to ensure, in his own words, that
My Lords, it is a distinct pleasure to follow the noble Lord, Lord Browne of Ladyton, and to join other noble Lords in congratulating my noble friend Lord Ravensdale on the very thoughtful way in which he introduced this important debate. I declare my interests as chairman of King’s Health Partners, chairman of the King’s Fund and chairman of the Office for Strategic Coordination of Health Research.
There are few areas in national life, the conduct of government and the delivery of public services that will become as dependent on artificial intelligence as that of healthcare. It is that particular area to which I will confine my remarks. We all recognise that there are increasing demands on the delivery of healthcare services through a changing population demographic, a subsequent increased demand on clinical services and a substantial workforce shortage. Of course, with all that increasing demand, there will be the need either for the economy to grow at a substantial rate to be able to provide funding for those services or for us to adopt innovation to deliver those services. One of the important innovations is of course the application of artificial intelligence. We have seen that already in healthcare in the areas of diagnostics, imaging and pathology. It is helping us to deliver high-throughput analysis of scans of pathological samples, and, through the application of algorithms, it is helping us to determine better the risk of poor outcomes in patients, to improve diagnosis and therefore to improve the efficiency of our service.
However, there are also substantial challenges. The development of artificial intelligence modalities requires access to high-quality data, and in healthcare we know that data are fragmented across the system through the use of different methods for the collection and creation of them. As a result, unless we have a single approach to the collection of data, we run a substantial risk that the data used to develop and to train AI systems will be inaccurate, and, as a consequence, inaccuracy will be translated into the provision of clinical services. That will potentially drive discrimination in those services, whereby the data on underrepresented populations are not sufficiently incorporated into the development of such technologies and tools.
It is pleasure to follow the noble Lord, Lord Kakkar, and I thank the noble Lord, Lord Ravensdale, for scheduling this most timely debate. I draw attention to my relevant interests in the register, specifically my advisory roles with various tech companies—notably Tadaweb, Whitespace and Thales—and my membership of the House of Lords AI in Weapon Systems Committee.
It is as a result of my membership of that committee that I am prompted to speak, but I emphasise that the committee, although it has collected a huge amount of evidence, is still some way off reaching its final conclusions and recommendations. So today I speak from a purely personal and military perspective, not as a representative of the committee’s considered views. I want to make a few broad points in the context of the regulation of artificial intelligence in weapon systems.
First, it is clear to me at least, from the wide range of specialist evidence given to our committee, that much of that evidence is conflicted, lacks consensus or is short of factual support. This is especially true of the technology, the capability of which is mostly concerned with future risk rather than current reality. Secondly, it is reasonably clear that, although there is no perfect equilibrium, there are as many benefits to modern warfare from artificial intelligence as there are risks. I think of such things as greater precision, less collateral damage, speed of action, fewer human casualties, less human frailty and a greater deterrent effect—but this is not to deny that there are significant risks. My third general point is that to deny ourselves the potential benefit of AI for national military advantage in Armed Forces increasingly lacking scale would surely not be appropriate. It will most certainly not be the course of action that our enemies will pursue, though they may well impel us to do so through devious means.
My Lords, I declare my interest as a member of two recent select and scrutiny committees on AI, and as a founding board member of the Centre for Data Ethics and Innovation.
Together with others, I congratulate the noble Lord, Lord Ravensdale, on this debate, and it is a pleasure to follow the noble and gallant Lord, Lord Houghton.
We are at a pivotal moment in the development of AI. As others have said, there is immense potential for good and immense potential for harm in the new technologies. The question before us is not primarily one of assessing risk and developing regulation. Risk and regulation must both rest on the foundation of ethics. My fundamental question is: what is the Government’s view on the place of ethics within these debates, and the place of the humanities and civil society in the development and translation of ethics?
In 1979, Pope John Paul II published the first public document of his papacy, the encyclical Redemptor Hominis. He drew attention to humanity’s growing fear of what humanity itself produces—a fear revealed in much recent coverage of AI. Humanity is rightly afraid that technology can become the means and instrument for self-destruction and harm, compared with which all the cataclysms and catastrophes of history known to us seem to fade away. Pope John Paul II goes on to argue that the development of technology demands a proportional development of morals and ethics. He argues that this last development seems, unfortunately, to be always left behind.
Professor Shoshana Zuboff made a similar point on this time lag more recently, in her book The Age of Surveillance Capitalism. She wrote:
“We have yet to invent the politics and new forms of collaborative action … that effectively assert the people’s right to a human future”.
These new structures must be developed not by engineers alone but in rich dialogue across society. Society together must ask the big ethical questions. Will these new technologies lead us into a more deeply humane future and towards greater equality, dignity of the person and the creative flourishing of all? Or will they lead us instead to a future of human enslavement to algorithms, unchallenged bias, still greater inequalities, concentration of wealth and power, less fulfilling work and a passive consumerism?
My Lords, it is a great pleasure to follow the right reverend Prelate. I declare my interest as a member of the AI in Weapon Systems Committee. I very much thank the noble Lord, Lord Ravensdale, for choosing for debate a subject that arguably now trumps all others in the world in importance. I also owe a debt of gratitude to a brilliant young AI researcher at Cambridge who is researching AI risk and impacts.
I could, but do not have the time to, discuss the enormous benefits that AI may bring and some of the well-known risks: possible extreme concentration of power; mass surveillance; disinformation and manipulation, for example of elections; and the military misuse of AI—to say nothing of the possible loss, as estimated by Goldman Sachs, of 300 million jobs globally to AI. Rather, in my five minutes I will focus on the existential risks that may flow from humans creating an entity that is more intelligent than we are. Five minutes is not long to discuss the possible extinction of humanity, but I will do my best.
Forty years ago, if you said some of the things I am about to say, you were called a fruitcake and a Luddite, but no longer. What has changed in that time? The changes result mainly from the enormous development in the last 10 years of machine learning and its very broad applicability—for example, distinguishing images and audio, learning patterns in language, and simulating the folding of proteins—as long as you have the enormous financial resources necessary to do it.
Where is all this going? Richard Ngo, a researcher at OpenAI and previously at DeepMind, has publicly said that there is a 50:50 chance that by 2025 neural nets will, among other things, be able to understand that they are machines and how their actions interface with the world, and to autonomously design, code and distribute all but the most complex apps. Of course, the world knows all about ChatGPT.
My Lords, let me join the queue in congratulating the noble Lord, Lord Ravensdale, on his stellar career to date and his excellent speech. He will bring an enormous amount to your Lordships’ House, especially his depth of knowledge in the area we are discussing.
ChatGPT and other forms of generative AI have taken the world by storm. I am a social scientist, but I have spent several years studying AI and we have never seen anything like this: every day around the world, the media has stories about the evolution of AI. This is a quite extraordinary phenomenon.
As we all know, the regulatory problems are huge, and Governments everywhere are scrambling to keep up. Extremely distinguished scientific figures, such as Geoffrey Hinton, just mentioned by the noble Lord, Lord Fairfax, have declared generative AI to be an existential threat to humanity. I do not wholly agree with this; it is awesome that human consciousness might be replicated and even improved on, in some sense, but the whole structure of science depends on social institutions. We cannot simply replace that by machines. I think there is a progressive merging of human beings and intelligent machines, and I do not see much of a way back.
In the meantime, a host of other organisations such as schools and universities, where I work, is struggling to cope with much more mundane problems such as how to ensure that students are actually the authors of the academic work they produce. That does not sound like much, but it is a big issue on the ground.
From its earliest origins, AI has been linked to geopolitics and war. It was created not by Silicon Valley but by ARPA and DARPA, the huge research programmes set up by the US Government in response to Russia’s Sputnik 2, which carried Yuri Gagarin into space. Regulatory efforts today, as in previous eras, are likely to be dislocated on a global level by current geopolitical tensions, especially those involving Russia, China and the West.
20 of 63 shown
However, then we have the flip side, the concerns around risks, all the way up to existential risks. We live in a radically uncertain world, a terminology coined by the noble Lord, Lord King of Lothbury, and John Kay. There has been much hyperbole around AI risk in recent months, but we need to take those risks seriously. Just as Martin Weitzman put forward his very elegant argument around the rationale for investing large amounts of money today on climate change, based on the tail risks of worst-case scenarios, so too we should consider the tail risks of where massive increases in digital compute and the potential emergence of a superintelligence —something that far exceeds human intellectual capabilities in every area—will take us, and invest appropriately based on that.
There are no historical parallels for the technological singularity that AI could unleash. Perhaps one instructive episode would be the fate of the Aztec civilisation. The Aztecs had existed in the world as they knew it for many thousands of years, with no significant change from one year to the next. Then one day in 1519, the white sails of the fleet of Cortés appeared on the horizon, and nothing was ever the same again. Within months, Cortés and his few hundred men had conquered the vast Aztec empire, with its population of millions, one of the most remarkable and tragic feats in human history. To avoid perhaps one day in the coming decades seeing a version of the white sails of Cortés on our own horizon, we must carefully consider our approaches now to this rapidly developing technology and manage the risks. That means regulation. It just will not do to let the private sector get on with it and hope for the best.
What should this mean for regulation and legislation development? The key point for me is that the Government cannot effectively regulate something that they do not adequately understand. I may be wrong, but I do not think that the Government, or any noble Lord here today, will have a fully thought-through plan of action for the regulation of AI. We are in a highly unpredictable situation. To this end, the first thing that we need to think about is how we can implement a sovereign research capability in AI which will develop regulation in parallel.
Research and regulation are different sides of the same coin in this instance. We need to learn by doing, we need agencies that can attract top-class people and we need new models of governance that enable the agility and flexibility that will be required for public investment into AI research and regulation. Any attempt to fold this effort into a government department or a traditional public research organisation is simply not going to work.
So how should we go about this? It was a great privilege a few years back to help shape the Advanced Research and Invention Agency Act, and I am very pleased to see ARIA moving forward at this albeit early stage. There are a number of things we can draw from it regarding how we approach AI capability. AI capability is exactly the sort of high-risk, high-reward technology we would expect ARIA to be investing in. But if we agree that AI needs a research focus, we could perhaps set up a different organisation in the same way as ARIA, but with a specific focus on AI, and call it ARIA-AI; or we could even increase funding and provide that focus to an existing part of ARIA’s organisational set-up.
In Committee on the ARIA Bill, we debated extensively the potential to give ARIA a focus or aim similar to that of the United States’ Defence Advanced Research Projects Agency, and ARPA-E. The Government wanted ARIA to maintain the freedom to choose its own goals, but there is an opportunity now to look at this again and use the strengths of the ARIA concept—its set-up, governance structures and freedom of action—to help the UK move forward in this area.
This is similar in some respects to the “national laboratory” for AI proposed by Tony Blair and the noble Lord, Lord Hague, in their recent report. This research organisation, along with the AI task force, if set up in the right way, would advance research alongside regulation, enable a unique competitive advantage for the UK in this area and begin the process of solving AI safety problems.
This will all need to be backed up by the right levels of government support. This is one of those areas where we should fully commit as a nation to this effort, or not press on with it at all. I can think of a number of such examples. The Government’s aspiration to build up to exascale levels of computing by 2026 is very welcome but would give the entire British state the ability to train only one GPT-4 scale model four years after OpenAl did. In addition, before DeepMind was acquired by Google, it had an annual budget of approximately £l billion a year, which gives a view of the scale of investment required. Can the Minister in summing up say what plans there are to scale up the Government’s ambitions in this area?
Finally, the Government’s recent White Paper outlines a pretty sensible approach in balancing management of the risks and opportunities of the technology, but as I said at the start, there are areas where it has perhaps been overtaken by events, in a field that is moving at breakneck speed—and that speed is the problem here. Unlike climate change, the full effects of which will not manifest over decades, or even centuries, AI is developing at an incredible pace. We therefore need to start thinking immediately about the initial regulatory frameworks. The Government could consider as a minimum putting their five principles in the White Paper on a statutory footing in the near term to provide regulators with enhanced powers to address the risks of AI.
Here is the bones of an AI Bill, perhaps legislating to set up a new research organisation, providing regulators with the right initial powers and the funding to sit behind all of this, which would at the same time build upon the world-leading AI development capabilities we now have in the UK. I beg to move.
Some industries are already seriously concerned, and with good reason. Those in the creative sector, particularly news publishers, are worried about intellectual property. The Minister covers IP policy as well as AI and will be aware just how important this issue is. I would be grateful if he updated us on the Intellectual Property Office working group, which is developing government policy so that news organisations, publishers, writers, artists, musicians and everyone else whose creations are being used by the tech firms to develop LLMs can be properly compensated, and commercial terms established that are fair to all.
Content creators are already seeing their work being used to train generative AI models. If studio businesses can get movie scripts, images or computer-generated background artists for free, why would they pay? The strikes in Hollywood are probably just the beginning of the disruption. In my committee’s creative industries report in January, we predicted looming disruption in the sector and called on DCMS to pay more attention. Sadly, we were right, although changes have come much faster than expected.
At the same time, we cannot wish these technologies away, and nor should we—they present massive opportunities too. We may now be at a critical juncture, both in securing UK competitive advantage in the AI race, and in preventing the risk of overmighty tech firms releasing technologies they cannot control. We need to get this right, and fast. I hope my committee’s work will play a role in shaping this debate and informing government policy.
I look forward to hearing much more on AI regulation in the coming months, and I hope the Minister and his colleagues will respond enthusiastically when we invite them to give evidence to our committee.
“innovation doesn’t come at the expense of Americans’ rights and safety”.
I am mindful too of machinery of government issues. Like climate change, AI cuts across apparently discrete areas and will have consequences for all areas of government policy-making. Of course, as a member of the AI in Weapons Systems Select Committee of your Lordships’ House, I am conscious that the ethical implications of AI for national defence are sparking great concern. But, as the Government’s White Paper made clear, we envisage a role for AI in everything, from health and energy policy to law enforcement and intelligence gathering. It is therefore imperative that the Government establish clear lines of accountability within Whitehall so that these intersections between discrete areas of policy-making are monitored and only appropriate innovation is encouraged.
Briefings we all received in anticipation of this debate highlight growing concern over the lack of transparency and accountability about the existing use of AI in areas such as policing and justice, with particular emphasis on pursuing alleged benefit fraud. The Dutch example should be a lesson to us all.
I should be grateful if the Minister would describe how the current formal structures interact, as well as the degree to which No. 10 provides a central co-ordinating role. As the AI Council recedes from view and as the Centre for Data Ethics and Innovation’s newly appointed executive director and apparently refreshed board get to grips with how to support the delivery of priorities set out in the Government’s National Data Strategy, my second question to the Minister is whether is he feels that the recommendation in the recent joint Blair-Hague report should be under active consideration—especially having the Foundation Model Taskforce report directly to the Prime Minister. That may be a useful step to achieving better co-ordination on AI across government. If not, why not?
In preparing for today, I had the pleasure of tracking the Government’s publications on this issue for the past three years or so. In each of those, they quite rightly emphasise the importance of public trust and consent. From my experience as a member of the AI in Weapons Systems Select Committee, I note that in the first section of the executive summary of the Defence Artificial Intelligence Strategy, the Government’s vision is to be “the world’s most … trusted” organisation for AI in defence. An essential element of that trust, we are told, is that the use of AI-enabled weapons systems will be restricted to the extent of the tolerance of the UK public. The maintenance of public trust and support will be a constant qualification of the principles that will inform the use of AI-enabled systems. As there has never been any public consultation on the defence AI strategy, how will the Government on our behalf determine the limits of the tolerance of the public? My own research has revealed that this is a very difficult thing to measure if it is not informed tolerance or opinion. The Centre for Data Ethics and Innovation’s polling corroborates that. What steps are the Government taking to educate the public so they can have an informed base to decide their individual or collective tolerance or level or trust in the use of AI for any purpose, never mind defence?
Successive Governments have had great difficulty in establishing the social licence that will allow for the broad collection and use of those data to drive research, technology and innovation opportunities in healthcare. The whole data area will be one of the most important regulatory challenges in the safe and effective development of AI technologies in healthcare. Does the Minister believe that His Majesty’s Government are at a place now where they can secure access to those data to drive these important opportunities? If not, how will they drive the data revolution in such a way that the public, more generally, are confident that those data will be used broadly for this purpose and other research and development purposes?
The MHRA in 2020 defined the regulatory pathway for the adoption of AI technologies as one very much mirroring those for medical devices. Clearly, some years have passed since that important approach to the regulation of AI was first established. The rigour with which the development of devices, or the regulatory supervision of the development of devices, is applied is slightly different to that for other therapeutic innovations. Is the Minister content that, in pursuing a pathway of regulatory development that is based on the device pathway—which is predominantly risk-based; that is reasonable—and looks at the safety and performance of these applications, there will be sufficient regulatory rigour to provide public confidence?
Regulatory elements of that pathway must not only include an understanding of the source of the data used to develop these technologies but provide a requirement for transparency in terms of an appropriate understanding of what forms the basis of the AI application. That is so that there can be a proper clinical understanding of the appropriateness of that application and of how it can be applied in the broader context of what must be understood about patients and their broader circumstances to reach an appropriate clinical decision.
Beyond that, there are also substantial concerns about ethical considerations in terms of both data privacy and the questions asked to train a clinical application being properly grounded in the modern ethics of healthcare delivery. Is the Minister content that the current regulatory pathway is sufficient and what steps are proposed by His Majesty’s Government to continue to develop these regulatory pathways so that they keep pace with the important advances and therefore benefits that will be derived from AI in healthcare?
My own view, therefore, is that the sensible policy approach to the future development of AI in weapon systems is to proceed but to do so with caution, the challenge being how to satisfactorily mitigate the risks. To some extent, the answer is regulation. The form that might take is up for further debate and refinement but, from my perspective, it should embrace at least three main categories.
The first would be the continued pursuit of international agreement or enhancements to international law or treaty obligations to prevent the misuse of artificial intelligence in lethal weapon systems. The second would be a refined regulatory framework which controlled the research, development, trials, testing and, ultimately—when passed—the authorisation of AI-assisted weapon systems prior to operational employment. This could be part of a national framework initiative.
As an aside, I think I can say without fear of contradiction that no military commander—certainly no British one—would wish to have the responsibility for a fielded weapon system that made autonomous judgments through artificial intelligence, the technology and reliability of which was beyond human control or comprehension.
The third area of regulation is on the battlefield itself. This is not to bureaucratise the battlefield. I think I have just about managed to convince a number of my fellow committee members that the use of lethal force on operations is already a highly regulated affair. But there would need to be specific enhancements to the interplay between levels of autonomy and the retention of meaningful human control. This is needed both to retain human accountability and to ensure compliance with international humanitarian law. This will involve a quite sophisticated training burden, but none of this is insurmountable.
I want to finish with two general points of concern. There are two dangerous and interlinked dynamics regarding artificial intelligence and the nature of future warfare. Together, they need us to reimagine the way future warfare may be, and arguably already is being, conducted. Future warfare may not be defined by the outcome of military engagement in set-piece battles that test the relative military quality of weapons, humans and tactics. The desirability of risking the unpredictability of crossing the threshold of formalised warfare may cause many people, including political leaders, to think of alternate means of gaining international competitive advantage.
The true dangers of artificial intelligence, in a reimagined form of warfare that is below the threshold of formalised war, lie in its ability to exploit social media and the internet of things to radicalise, fake, misinform, disrupt national life, create new dependencies and, ultimately, create alternate truths and destroy the democratic process. By comparison with the task of regulating this reimagined form of warfare, the regulation of autonomous weapon systems is relatively straightforward.
Five years ago the Government established the Centre for Data Ethics and Innovation to explore these questions. The centre began well but was never established on an independent legal footing and seems to have slipped ever further from the centre of the Government’s thinking and reflection. One of the hopes of the CDEI was that it would provide an authoritative overview of sector-led innovation, have a co-ordinating and oversight role, and be a place for bringing together public engagement, civil society, good governance and technology. I therefore ask the Minister: what are the Government’s plans for the CDEI in the current landscape? What are the plans for the engagement of civil society with AI regulation and ethics, including with the international conference planned for the autumn? Will the Government underline their commitment to the precautionary principle as a counterweight to the unrestrained development of technology, because of the risks of harm? Will we mind the widening gap between technology and ethics for the sake of human flourishing into the future?
At the extreme, artificial systems could solve strategic-level problems better than human institutions, disempower humanity and lead to catastrophic loss of life and value. Godfathers of AI, such as Geoffrey Hinton and Yoshua Bengio, now predict that such things may become possible within the next five to 20 years. Despite two decades of concentrated effort, there has been no significant progress on, nor consensus among AI researchers about, credible proposals on the problems of alignment and control. This led many senior AI academics—including some prominent Chinese ones, I emphasise—as well as the leaderships of Microsoft, Google, OpenAI, DeepMind and Anthropic, among others, recently to sign a short public statement, hosted by the Center for AI Safety:
“Mitigating the risk of extinction from AI should be a global priority alongside other societal-scale risks such as pandemics and nuclear war”.
In other words, they are shouting “Fire!” and the escape window may be closing fast.
As a result, many Governments and international institutions, such as the UN and the EU, are suddenly waking up to the threats posed by advanced AI. The Government here are to host a global AI safety summit this autumn, apparently, but Governments, as some have said, are starting miles behind the start line. It will be critical for that summit to get the right people in the room and in particular not to allow the tech giants to regulate themselves. As Nick Bostrom wrote:
“The best path towards the development of beneficial superintelligence is one where AI developers and AI safety researchers are on the same side”.
What might the shape of AI regulation look like? Among other things, as the noble Lord, Lord Ravensdale, said, Governments need to significantly increase the information they have about the technological frontiers. The public interest far outweighs commercial secrecy. This challenge is international and global; AI, like a pandemic, knows no boundaries.
Regulation should document the well-known and anticipated harms of societal-scale AI and incentivise developers to address these harms. Best practice for the trustworthy development of advanced AI systems should include regular risk assessments, red teaming, third-party audits, mandatory public consultation, post-deployment monitoring, incident reporting and redress.
There are those who say that this is all unwarranted scaremongering, as some have touched on this afternoon, and that “there is nothing to see here”. But that is not convincing because those people—and they know who they are—are transparently just talking their own commercial and corporate book. I also pray in aid the following well-known question: would you board an aeroplane if the engineers who designed it said it had a 5% chance of crashing? Some, such as Eliezer Yudkowsky, say that we are already too late and that, as with Oppenheimer, the genie is already out of the bottle; all humanity can do is to die with dignity in the face of superhuman AI.
Nevertheless, there are some very recent possible causes for hope, such as the just-announced White House voluntary commitments by the large tech companies and the Prime Minister’s appointment of Ian Hogarth as the chair of the UK Government’s AI Foundation Model Taskforce.
For the sake of humanity, I end with the words of Dylan Thomas:
“Do not go gentle into that good night …
Rage, rage against the dying of the light”.
Given that we are starting miles behind the start line, I refer to Churchill’s well-known exhortation: “Action this day”.
The most extensive plan for regulation that I have seen is that developed by the EU. As with our Government’s programme, its aim is to balance opportunity and risk, although there are lots of huge issues still unresolved in all this. I support that ambition but it will be very difficult to achieve, given that the pace and scope of innovation has accelerated off the wall. This time, much of it is driven by digital start-ups rather than the huge digital corporations, and it is genuinely global.
Is there anyone in the Chamber who does not have their smartphone close at hand? Even I have mine in my pocket, although I sort of hate it. The smartphone in your pocket has much more intelligence than many machines that we have created, including those that took human beings to the moon. This is quite extraordinary stuff.
As was announced only about two days ago in the US, ChatGPT will become available to smartphone users across the world. Imagine the impact of that, given that the proportion of the world’s population with access to a smartphone has risen to 84%. It will be absolutely awesome. I make a strong plea to the Government Front Bench to have a far more extensive debate about these issues, because they are so fundamental to a whole range of political, social and economic problems that we face.