The risk posed by superintelligent AI is on a par with the nuclear threat - Alex Sobel

Every day seems to bring a new revelation about the risks posed by artificial intelligence (AI). Reports of AI behaving deceptively, pursuing goals in unexpected ways, and even collaborating autonomously have moved concerns about AI from the realm of science fiction into mainstream debate.

Governments around the world are now scrambling to catch up, with growing momentum behind international regulation.

The UK Parliament’s Joint Committee on Human Rights (JCHR) has just published its report, ‘Human Rights and the Regulation of AI’. I proposed the inquiry several months ago and now have the privilege of chairing the committee as its conclusions are released at a moment of growing concern about the risks posed by advanced AI.

The report found that AI is already riddled with bias and human rights violations. AI systems pose particular risks to privacy and personal data, and there are a number of problems with existing protections. Human rights principles must inform all aspects of the government’s policies and practices relating to AI systems, including its approach to the regulation of AI.

An unassailable problem arises when it comes to ‘Superintelligent Artificial Intelligence’: AI that humans cannot control. We can never be fully prepared to reign this kind of technology back in once it’s unleashed.

It is important to understand why this happens. The general-purpose AI systems used today are not coded line-by-line, like traditional software. Using vast amounts of data, they are grown, not built. This results in systems we don’t fully understand and can’t predict.

Top AI companies continue a reckless race towards superintelligent AI, whilst already releasing products whose behaviour they cannot fully explain or control. The primary risk of superintelligent AI is that humans permanently lose control.

The JCHR report explains how Professor Roman Yampolskiy, Director of the Cybersecurity Laboratory at the University of Louisville, recommended a ban on the creation of uncontrolled general superintelligence, which he characterised as a national security threat.

The development of the most advanced AI systems happens on a spectrum, so it matters where we draw the line. That line must recognise the capabilities of superintelligent AI. The clearest example of such a capability is recursive self-improvement: the full automation of AI development with no humans in the loop. That is the point where we lose control, as development towards superintelligent AI proceeds too fast for humans to intervene.

The risk posed by superintelligent AI is on a par with nuclear threat. AI is the crucial technology of the 21st century, promising vast benefits for science, medicine and the UK’s prosperity, but superintelligent AI is different. It would not be a tool that we can leverage but an entity in its own right, capable of improving itself, replicating itself and resisting attempts to shut it down. It would amount to a rogue power, stationed inside our data centres, under no one’s command, capable of seizing control of our critical infrastructure.

The computing power required to develop increasingly advanced AI systems is also driving demand for hyperscale data centres, as we have seen in Leeds. The two issues are intrinsically linked, which is why effective regulation and robust safeguards are so important.

We have acted before to build resilient systems when faced with existential threats, such as nuclear and chemical warfare. Britain has led the way when it comes to AI, let's continue to lead from the front.

With AI policy now at the Cabinet table, Britain has an opportunity, and I would argue a responsibility, to shape the next stage of international AI governance. I am therefore delighted to see the Prime Minister has taken on board the recommendations of the JCHR and announced on Tuesday that the UK will use its upcoming G20 presidency to help forge a global agreement on AI.

When talking about existential threats, the human mind is inclined to push it into the distant future, but we are much closer than we might think to the dangers of superintelligent AI.

Progress in AI development is rapidly picking up speed. Yodkowsky and Soares note in their book, ‘If Anyone Builds It, Everyone Dies’, that, “most computer scientists in 2015 would have told you that ChatGPT level artificial conversation wouldn’t be in reach for another 30 or 50 years.” For perspective, ChatGPT became publicly available only three years ago, yet AI companies could reach superintelligent AI in the next two to five years.

Looking closer to home, the government should introduce a new AI Bill, as promised in the 2024 King’s Speech, to place AI regulation on a sound basis and give full effect to the Framework Convention. In my capacity as an MP, my private members’ bill on artificial superintelligence seeks to regulate and prohibit development now before it's too late.