AI clarity reduces existential nuclear risk- multilateral military coop is key
Boulanin et al. 20 (vincent, PhD in Political Science from École des Hautes en Sciences Sociales, lora saalman, petr topychkanov, fei su, moa peldán carlsson, “ARTIFICIAL INTELLIGENCE, STRATEGIC STABILITY AND NUCLEAR RISK”) June 2020 https://www.sipri.org/sites/default/files/2020-06/artificial_intelligence_strategic_stability_and_nuclear_risk.pdf
Existing nuclear risk-reduction measures The fact that many of the risks and strategic challenges posed by AI are not fundamentally new means that policy options to mitigate the impact of AI on nuclear risk already exist. In fact, a number of existing policy options for nuclear risk reduction could have a positive effect on the strategic relations of nuclear armed states and could help to reduce nuclear risk. These include no-first-use policies, commitments to lower the alert status of nuclear arsenals, transparency and information sharing, and cooperation. No-first-use policy One of the most effective means to prevent the nuclear escalation scenarios described in chapter 4 would be if all nuclear-armed states adopted a clear NFU policy. Currently, China and India are the only nuclear-armed states that have such a policy. Although universal adoption of NFU policies would not necessarily alleviate all the signalling problems generated by the introduction of AI in nuclear weapon systems, from the authors’ perspective it would be positive for strategic stability if all nuclear-armed states, particularly Russia and the USA, were to make such a commitment. Some participants in the SIPRI workshops—notably from Russia and the USA—did not share this view. Indeed, in the current geopolitical context, it is highly unlikely that these two states, as well as the USA’s NATO partners France and the UK, would seriously consider that possibility. The USA in particular is constrained from official adoption of NFU by its alliance structures and the views of its allies in both East Asia and Europe.491 The likelihood of Russia and the USA adopting NFU policies is further compromised by both official and non-official statements from China and India about the longer-term status and nature of their NFU commitments.492 A commitment to lower the alert status of nuclear arsenals Removing strategic weapons from a launch-on-warning or launch-ready alert status would allow more time for decision makers to make appropriate assessments. India, Pakistan and China have the lowest level of alert as their nuclear warhead and delivery vehicles are reportedly kept separate.493 It would take at least a few days for them to launch a nuclear strike (although this is changing as they develop the sea legs of their nuclear triads).494 In contrast, Russia and the USA have their systems on high alert, which means that they are much more likely to use nuclear weapons by accident or deliberately in a crisis or from the outset of a conflict. It would be highly positive for international security if Russia and the USA were to de-alert their nuclear weapons––although this would not be a panacea without proper training, data integrity and cybersecurity systems in place. In the current context, this also seems to be an unrealistic possibility. Nevertheless, there is reason to hope that this possibility could be part of future bilateral nuclear riskreduction measures between these two countries, as well as between them and other nuclear-armed states. Transparency and information sharing A traditional approach to transparency and information sharing could also help in risk reduction. AI is a kind of technology that is prone to misunderstanding for two principal reasons. First, there remains widespread misconceptions about what AI is and what it can or could do. Second, it is difficult for states to assess in a tangible way each other’s progress in this area. It is, for instance, impossible for a state to assess whether an opponent’s air defence systems rely on traditional hard-coded programming or an algorithm trained by machine learning if that information has not already been disclosed by the opponent itself. There are a number of transparency-centred CBMs that states could implement voluntarily and unilaterally or as a result of bilateral or multilateral dialogue. These can be grouped in three categories: (a) AI-specific measures, (b) measures related to the military use of AI, and (c) measures related to the use of AI in connection with nuclear weapons and deterrence. AI-specific measures. One AI-specific measure would be for a state to make publicly available official documents that outline its general strategy and policy on AI. If a state has not yet adopted a national policy or strategy on AI, then it could adopt one to clarify its intention and views. Another AI-specific measure would be to make publicly available governmental recommendations (e.g. guidelines, procedures and techniques) about testing and verification and about cybersecurity in the field of AI.