Faustian Bargain 2.0. Life Under the US AI (Artificial Intelligence) Umbrella
The US National Security Memo says it must lead the world in AI: What this means for allies in general and the UK specifically.
[A short, 500 word version of some of this post first appeared as part of Just Security’s expert commentary on the US National Security Memo on AI, you can read it, and the views of many experts, here].
The US has woken up. The US National Security Memorandum on Artificial Intelligence published last week (24 Oct 24) describes AI as “An Era Defining Technology” and offers a comprehensive plan for US leadership in the field. Its scope and ambition, its clarity of action, are remarkable. Of course, this does not mean that US institutions will be able to respond to such clarity with coherence and efficient implementation. Nor can there be any guarantee that the next White House Administration will implement the Memo. But we should assume that they will, and plan accordingly given the risks it poses to the UK. We might also note that as direction, the document is probably as good as it could be to ensure US leadership.
Read from across the pond, it should be a wake-up call. It is a document that makes clear the intention that ‘The United States must lead the world in the… …application of AI to…national security functions.” This should be as much a cause for alarm among allies as reassurance.
First the reassurance. Russian President Vladimir Putin was right to say that ‘Whoever leads in AI will Rule the World’, even if it is vanishingly unlikely it will be Russia that leads. The Memorandum on AI should reassure because by far the worst outcome for democratic nations across the world is that the Chinese Communist Party (or other autocracy) wins the race to Artificial General Intelligence and Superintelligence (AGI, ASI), and ‘rules the world’.
Winning would give the CCP unprecedented economic growth, unhindered by demographic constraints from China’s ageing society. It would allow innovation without education, reducing or removing the need to educate the population to increase its inventiveness and productivity, side-stepping a constraint usually encountered by autocracies: educating people can also make them more aware of the tyranny in which they live, better able and more willing to challenge it. With AI ideas become a dime a dozen, as one AI paper recently described – and this when we are still well short of AGI.
A CCP AGI/ASI would be able to devise and implement strategies, material and persuasive, that would be constantly ahead of and unanticipated by the CCP’s opponents domestic and foreign. If this same CCP AGI or ASI were capable of recursive self-improvement, it would be something close to a doomsday scenario for democracies – we might never be able to match the intelligence that could be deployed by the CCP. As citizens and countries, we would watch an autocratic intelligence explosion – AGI/ASI deployed everywhere, and intelligence runaways – innovation escape velocity – AGI/ASI getting better and better leaving us further and further behind. The CCP would have unprecedented and unmatchable power over its own population, but also over people everywhere, things we know already it seeks.[1]
The US having as clear a plan as that articulated in the Memo will likely mean democracies and other US allies and partners are shielded from such an outcome. Much like the US nuclear umbrella shields much of the world from autocratic nuclear blackmail.
Why alarm? Like the nuclear umbrella, access to US AGI/ASI protection (the US AI Umbrella?) will come at a cost to those that accept it and rely on it. And it is likely to be much higher. A Faustian bargain too far.
Power is the ability to get things done. Power in the international system is relative. Even in alliance, how powerful you are, relative to allies, determines your level of influence, your ability to secure outcomes as favourably aligned with your interests as possible. If the US plan succeeds, its power would be such that the relative power of the UK and all other allies would diminish dramatically. Our alliances would be less equal. We would be much more dependent on the US. The US memo talks of promoting ’…equitable access to AI’s benefits’ for allies and partners, not ‘equal access’. What might this mean?
To answer, consider US military equipment sales to allies. Users of US technology don’t get the fully capable versions the US military operates itself. The UK, for example, helped build the F-35 as a Tier 1 partner and gets ‘…more access to critical information than lower-tier partners.’ But that is not the same as ‘equal access’ to our US allies.
Access to the unprecedented economic and military advantage that AGI and ASI would give the US will come at a cost. Negotiating with the US will be like playing AlphaGoZero at chess or Go. Nations will be constantly outmanoeuvred. Similarly, US companies, likely given privileged access to US models, will be more innovative and effective in design and execution of strategies and plans than their non-US rivals. The US will likely use its leadership in AGI/ASI to ensure it dominates as the race to secure interplanetary, interstellar and intergalactic resources accelerates exponentially, further magnifying US dominance here on Earth. In short, the US’ relative power in the international system will so out-strip everyone else’s that our future prosperity and security, will be determined in Washington. Nothing like as worrying as it being determined by the CCP in Beijing, but alarming all the same. No nations have constantly aligned interests, no matter how close the alliance. While US allies should be reassured that the threat of an adversary getting to AGI first has receded, they must also understand that the costs of free-riding on the US are about to increase hugely.
What does the US memo say that we need to take notice of in the UK?
The US White House memo talks of ‘gaining the decisive edge’ through AI, and sets urgent timelines – 30-days, 3-months, for the US system to respond. This does not give the UK time to muddle through and hope this goes away. We must act with similar urgency and decisiveness.
The memo is an order not just for modernisation, but systemic transformation stating that:
‘Establishing national security leadership will require deliberate and meaningful changes to aspects of United States strategies, capabilities, infrastructure, governance, and organisation. AI is likely to affect almost all domains with national security significance…integration will only succeed if paired with appropriately redesigned United States Government organisational and informational infrastructure.’[2]
As part of redesigning organisations, processes, and informational infrastructure around AI, all US agencies are ordered to have a Chief AI Officer and an AI governance board.[3] All major departments are to update or issue guidance on AI.
The memo is a whole of Government plan. To pick out some of the more arresting instructions, the Memo:
· orders acquisition and procurement reform in Government within 30 days to enable rapid application of AI to National Security Strategy[4]
· requires that the US build energy, powerlines, computational resources, to ensure AI leadership, accelerating planning approvals processes and incentivising the building of AI-Infrastructure from fibre links to data centres.[5]
· plans strategic investment in overseas companies, and infrastructures – informed by intelligence estimates – to ensure US leadership in AI.[6]
· requires the US intelligence and other agencies to identify and map global AI talent within 6-months,[7] listing these people and writing a plan to bring them and their talent to the US[8] - and to make this so it requires the development and implementation of a plan to accelerate immigration and work visa processes for AI talent.
Further, it instructs the President’s Economic Advisors and Economic Council to complete an assessment of AI economic impact and US ‘relative competitive advantage’.[9]
The memo is intent on preventing strategic surprise – making science and technology intelligence a, perhaps the, priority. It provides just 90-days for the US to rewrite its intelligence priorities both foreign and domestic around the prioritising of AI leadership. It offers only 180-days (6-months) to map global supply chains in AI and to provide a plan to monitor these supply chains on an ongoing basis, with plans to ensure their resilience. It orders more rigorous efforts to stop espionage in academia and corporate/commercial collaborations, espionage via employment etc. And similarly requires more economic protections of US companies working in AI, through action by the Committee on Foreign Investment in the United States (CIFIUS).
The intelligence agencies in the US are also asked to reconsider what they release to the public, ‘…given that AI systems have demonstrated the capacity to extract previously inaccessible insight from redacted and anonymised data.’ While these agencies are also told to prioritise research and adoption of AI capabilities[10], transforming themselves to enable the technology’s application to intelligence tasks. They are instructed to develop talent plans, and expedited security clearance for AI experts. To offer scholarships and AI training.
In 6-months the US intends to have a plan, with testing already underway, to test frontier ‘AI models’ capacity to generate or exacerbate nuclear and radiological risks.’ There are to be classified evaluations on both general but also some specific risks in biological and chemical weapons.
It is also interesting to note that in the AI Safety at Section 3.3, the US is planning, to develop, within 6-months: ’the establishment of technical and legal tooling necessary for facilitating the rapid and secure transfer of United States Government, open-weight, and proprietary models to…[US Government secure]… facilities.’ In extremis, this seems to threaten something close to nationalisation. And whether it is for the protection of humanity from AI-risks, or the US state, from the risk of other nations gaining AGI and ASI access via US-based but global technology companies, may be irrelevant. Nationalisation, in some form, will be a possibility, with the legislative tools in place to do it quickly. The risk to allies should be clear.
On alliances, there is a deadline of 150 days to figure out co-development and co-deployment plans with allies[11], with the aim to ‘…promote equitable access to AI’s benefits’,[12] and an instruction to have a strategy for the advancement of ‘international AI governance norms’ in 120 days. But working with allies and partners is only a third objective in the report, after leading the world in AI, and harnessing it to achieve the US’ objectives. And even here, in the alliances section, the need for US pre-eminence is, to use a US military-term, ‘foot-stomped’ (emphasised). Allies and partners are there to help manage the US roll-out of AI, and ensure no-one else leads.
This is not to decry the US’ commitment to making sure AI is deployed to ‘promote worldwide benefits’ and develop ‘effective global norms’ to ‘manage AI risks’, ‘promote democratic values..’ etc. We would see no such commitment from, for example, the PRC, and there may be no other nation the UK could or should trust to exercise leadership in AI in a way that is so largely aligned to the UK’s interests.
In short, the memo is far-reaching, bold, rigorous, and proportionate to the risk and opportunity in AI.
The US is the first nation to try to really seize the initiative in AI, and was already best placed to take this initiative. We must not lose sight of the fact that, given neither the UK nor other allied nation is taking AI so seriously*, the most likely outcome as things stand is that will muddle our way into this Faustian bargain, and there is unlikely to be a way to muddle our way out. We must ask what will this mean for us as citizens, our businesses, our prosperity and security.
[*at least, not in the sense of proportionate investment and effort – whereas rhetorically, many offer grand statements showing they understand AI’s potential and current development trajectory - a path to embarrassment if they turn out to be right].
What should we do in the UK?
What we must not do is more of the same – muddling through. There is so much here that a great many people inside and outside of Government have tried to get the UK to do since perhaps as early as 2014, and certainly by 2018. In my own direct experience, in 2020 in No10 much of this was pursued with vigour, rigour and energy, only to unravel with the departure of Dominic Cummings from No10, and with it the work’s foremost champion.
But for all the frustration, we cannot allow ourselves to give up or counsel despair. We must act. Free-riding on the US will likely have enormous costs that we will not want to pay.
That being the case, the first thing the Prime Minister’s office should be doing is requesting of the US access to the classified Annex of the report.[13] If access is denied, the UK’s National Security Secretariat & intelligence analysis centres in the UK should consider deeply what is likely to be in it. It is very rare that the uncertainty within a known unknown such as the contents of this Annex cannot be reasonably bounded and estimated. The task is urgent. The UK has developed if not a complete dependence, then certainly something close to it, on the US for both our prosperity and security. We must have at least a working hypothesis for what the US is planning in this transformative technology, which threatens to deepen our dependence and perhaps extend it in perpetuity.
Once the UK has a clear picture of its US ally’s aims and intentions, the National Security Council should be discussing the memo, and the rapid progress towards AGI that seems most likely to have given rise to it.
In this context, it is noteworthy, I think, that the memo says nothing on economic growth until Section 3a, some 14 paragraphs in. It is in the economy that AI’s effects are likely to be most transformative and far-reaching. And it is AI’s capacity to hugely accelerate US economic power, that should be most alarming to the leadership of allied nations: the EU’s competitiveness report was written because of concerns at how far and how fast the EU was falling behind the US. The UK is in at least as poor a position. The weaker we get economically, scientifically, and technologically, the less equal our relationship with the US, the more reliant we become on US generosity and favour – which can be withdrawn, if ever we pursue our interests in ways the US does not want. The UK National Security Council must consider these economic aspects of relative power every bit as seriously as it does those that apply in the more direct national security sense – defence, intelligence, critical national infrastructure etc.
We must also be clear-eyed and hard-nosed here: the US wants leadership in AI to further enhance its power over allies as well as with them. As I wrote in ‘In Athena’s Arms’ our collaboration must be on a Reaganite ‘trust but verify’ basis. Like the US, we must be led by national interest first.
I won’t rehash my recommendations on what the UK should do to secure leadership in AI, nor the arguments for why it should, covered in both ‘Don’t Blink’ and ‘In Athena’s Arms’ in different ways, and in countless talks and articles previously.
But I can’t help feeling that we continue to treat AI as somebody else’s problem, assuming US beneficence will be extended to us such that we have equal access. This is naïve. Nations are allowed to free-load on the US only when the US judges it is in its interests for them to do so. It reduces allies power and influence – their ability to protect their interests globally - in return for domestic prosperity, and politically, the ability to spend on domestic priorities.
We do have a special relationship with the US, and it does often (with many, but not all Americans) go deeper than the transactional relationships that characterise other international relationships. But it has never been an unconditional relationship. Unless we want the conditions dictated rather than negotiated, we will need to ensure US leadership in AI is matched by the UK. And despite how much this will cost, and the effort it will take, in truth, we cannot afford to do otherwise.
[1] e.g. https://nationalpost.com/opinion/samantha-hoffman-for-china-technological-superiority-is-about-power-and-control;https://nationalpost.com/opinion/samantha-hoffman-for-china-technological-superiority-is-about-power-and-control;https://www.journalofdemocracy.org/articles/chinas-tech-enhanced-authoritarianism/
[2] Ibid. Section 1. Policy. Para 1g.
[3] 4.1. e.ii, A & B.
[4] 4.11 e.1.
[5] 3.1 e. i-iv
[6] 3.1 e. para v
[7] 3.1.b. i & ii
[8] 3.1. c. iv
[9] 3.1. c. iii
[10] 4.1
[11] 4.1. h.i, ii, A-E.
[12] 4.1 g. i. C. b.
[13] The White House, 2024. Memorandum on Advancing the United States Leadership in Artificial Intelligence: Harnessing Artificial Intelligence to Fulfill National Security Objectives and Fostering the Safety, Security, and Trustworthy Development and Use of AI. Available at: https://www.whitehouse.gov/briefing-room/presidential-actions/2024/10/24/memorandum-on-advancing-the-united-states-leadership-in-artificial-intelligence-harnessing-artificial-intelligence-to-fulfill-national-security-objectives-and-fostering-the-safety-security/ [Accessed 27 October 2024]. Section 1. Policy. Para 1a, 1b.