Autonomous Weapons Systems: UN Expert Talks Facing Failure

Source Link

The first of two GGE meetings on AWS planned for 2022 was held in March in Geneva. Russia used the forum to justify its illegal invasion of Ukraine, which nu­merous states including Germany sharply condemn. When the Russian delegation made its closing remarks many of the delegates demonstratively left the room.

The same geopolitical tensions that cul­minated in Vladimir Putin’s war in Ukraine have already caused the de facto failure of the Geneva talks, even if the group will meet again for five days in July. Without Russian buy-in there can be no regulation of AWS through the GGE, which makes its decisions by consensus. All 125 high con­tract­ing parties to the Convention on Cer­tain Conventional Weapons are entitled to participate in the GGE, while signatory states such as Egypt also have the right to speak. In reality, only about eighty states actually attend.

Fault lines within the GGE

Even before the Russian invasion it was clear that differences of substance within the Group of Governmental Experts pre­cluded rapid agreement.

First of all, the GGE has failed to agree on a common definition of AWS. Most states support the proposal from the International Committee of the Red Cross (ICRC), under which an AWS is a “weapon system with autonomy in its critical functions” that is capable of selecting and attacking targets without human intervention. China, how­ever, would only include systems possessing the capacity to autonomously modify and/ or expand their strategic mission. Another group of states sees no need to define AWS at all, preferring instead to concentrate on the appropriate level, type, degree and form of man/machine interaction.

France took the initiative in 2021 and presented its own definition, which is also sup­ported by Germany. It distinguishes between fully and partially autonomous systems: Fully autonomous lethal weapon systems are capable of selecting their tar­get and initiating an attack without human inter­vention, as well as modifying their stra­tegic mission. Germany and France believe that this category of weapons sys­tems should be entirely prohibited. Partially autonomous systems, on the other hand, select and attack targets within a framework defined by human operators, but can­not make more far-reaching decisions on their own. This latter category, France and Germany argue, needs to be regulated in order to ensure that they are used only in accordance with legal and ethical princi­ples. Other countries including Japan have indicated their interest in deepening the discussion on this proposed definition.

There is also disagreement over the termi­nology used to define the required level of human control. While many participating states prefer the term “human control”, the United States prefers the concept of “human judgment”. The Americans under­stand human control as meaning direct manu­al intervention in the weapons system itself, whereas they prefer to control the effects of the weapons. The advocates of human control want to see control over the weapons systems and not merely their effects.

Another fault line is the arms race be­tween the United States, Russia and China, which is especially pronounced in the sphere of new technologies. China in par­ticu­lar has massively increased its military spending to modernise its armed forces, and intends to become “the world AI leader” by 2030. While China does support regula­tion of AWS in the GGE, its narrow defini­tion raises doubts as to whether it really wants to submit to regulation (presumably in light of its geopolitical ambitions).

The greatest tensions at the moment are obviously those between the United States (or NATO) and Russia. Even if Russia regards “meaningful human control” over AWS as indispensable, it opposes expanding the exist­ing international legal framework. As long as Russia insists on that point it is hard to see a path to agreement in the GGE. This raises the question whether other forums might offer better prospects of progress. Ger­many should prepare for such scenarios. For example the German Federal Ministry of Defence could prepare a national posi­tion on AWS in conjunction with the Ger­man Foreign Office.

Defining a national position on AWS

Unlike France, Germany does not yet have a national position on AWS. The German government’s AI Strategy of 2018 touches only superficially on military uses of arti­ficial intelligence. The only real marker is a commitment in the current coalition agree­ment to actively promote the out­lawing of “lethal autonomous weapons sys­tems that are completely beyond the con­trol of humans”. One reason for the lack of a national position might be found in the differences on AWS between the Foreign Office and Defence Ministry. Also Germany prefers reach agreement with important part­ners, such as the United States, France and the Netherlands, before taking action. In fact both the United States and France have already adopted national positions on AWS; it appears that Germany wishes to wait and see what the coming months bring, and espe­cially how the war in Ukraine progresses.

A national position could include the dis­tinction – made at the 2021 Geneva talks by France and Germany – between fully and partially autonomous systems. It must be noted here, however, that both France and Germany are still only talking about “lethal” weapons systems. This restriction is problematic for two reasons. Firstly, non-lethal weapons can also cause great suf­fer­ing to combatants and civilian populations. Secondly it is difficult to draw a clear line between lethal and non-lethal weapons.

A national position should also lay out the parameters within which the use of par­tially autonomous weapons systems is to be permitted. It must therefore necessarily also address the degree of human/machine inter­action and the concept of human control. Preparatory studies by the International Panel on the Regulation of Autonomous Weap­ons (iPRAW), in which the present authors play a leading role, could supply valuable pointers points in this respect. iPRAW understands human control to require “situational understanding and op­tions of intervention during attack”. Both elements must be ensured by the design of the weapons system (control by design) and implementable during its deployment (con­trol in use). The concrete extent of human control will depend on the specific context of utilisation.

In other words, a national position will need to address the various scenarios and identify the associated red lines. Deployment of AWS in urban environments could be subject to tighter restrictions than in purely military settings, for example by making it obligatory for a human to moni­tor their use and if necessary intervene. Human intervention would be less relevant with autonomous submarines for example.

In light of current geopolitical developments, however, Germany should remain open towards technological innovation and take account of the military benefits of autonomy in its national position. Speed, targeting precision and force protection are some of the examples that mitigate for the use of AWS.

The new geopolitical situation and its influence on Germany’s position on AWS

Calls for Germany to adopt a national posi­tion must also be seen in the context of the transformed security environment: In his landmark speech on 27 February 2022, Chancellor Olaf Scholz declared: “We will now – year after year – invest more than two percent of our gross domestic product in our defence.” He also announced that the government would put €100 billion into a special fund for the Bundeswehr, which has now received the Bundestag’s approval. These developments have two principal implications for a national posi­tion on AWS:

Firstly, they will boost existing defence projects, such as the Future Combat Air Sys­tem (FCAS). FCAS is a system of systems bring­ing together existing and new ele­ments; parts of the FCAS are also to be equipped with autonomous functions. An independent panel of experts on the responsible use of new technologies in the FCAS will ensure that there is no contradiction with the gov­ernment’s commitment to outlaw weapons systems operating outside human control. One of the central points of discussion within the panel is the concept of human control and its concrete operationalisation.

Secondly, in light of the large increase in defence spending announced by Olaf Scholz, there will need to for carefully scrutiny of whether procurements are actually neces­sary and sensible. New technologies might offer worthwhile savings on personnel, for example in the form of AI-based data analysis. Although that particular case does not actually involve AWS, the questions thrown up by increasing use of AI are simi­lar, however it is employed.

So the government’s commitment to out­lawing AWS does not imply a general rejec­tion of technological innovation in the military sphere. On the contrary, a national position on AWS should include a clear com­mitment to the necessity for research and development on military technologies, and name the benefits. Only once such a dif­ferentiated national position has been defined will Germany be able to participate actively in future negotiations in coordination with its partners.

Alternative forums for regulating AWS

NATO, the EU and the UN are the principal alternative forums for regulating AWS (see figure). It is also conceivable that other states might initiate a process outside the institutions altogether.

NATO – finding a common transatlantic line

Even if regulating AWS falls outside NATO’s remit, it could nevertheless serve to bring the individual NATO partners closer together on the issue and thus strengthening their posi­tion in other forums, for example at the UN.

In autumn 2021 NATO published an Arti­ficial Intelligence Strategy whose six principles for responsible use are also relevant for AWS: lawfulness; responsibility and accountability; explainability and trace­ability; reliability; governability; and bias mitigation. But is says very little on the con­crete question of human/machine inter­action. In autumn 2022 NATO will publish its Autonomy Implementation Plan, which can be expected to go into greater detail on the human/machine interaction question. It is unclear whether the Autonomy Imple­mentation Plan will principally reflect the American position or also account for the views of the other NATO states. Fundamen­tally the United States argues for a broader understanding of the degree of autonomy to be permitted in weapons systems. This is also underlined by a paper on good prac­tices in the area of lethal AWS, which the United States submitted to the GGE in March jointly with NATO partners Canada and United Kingdom and close allies Aus­tralia, Japan and South Korea.

In 2020 NATO established a working party on human systems integration for meaningful human control over AI-based systems. The working party’s members are researchers who advise on the operationalisation and implementation of human con­trol over AWS on the basis of their experi­ence (rather than representing state posi­tions). One difficulty is that the working party uses the term “meaningful human control”, which in particular the United States rejects.

The working party could make a positive contribution by considering other formulations, without overly altering the terms of the debate or reopening old conflicts. This could also give the United States the possi­bility to agree definitions with its NATO part­ners. Even the ICRC deviates from its established terminology in its 2021 position paper on AWS, and now speaks of “human control and judgement”. The ICRC’s posi­tion could function as a door-opener ex­pand­ing the terminological side of the debate – always presupposing agreement can be reached on the aspect of human/ machine interaction.

Finally it must not be forgotten that not all members of the European Union are also members of NATO. For instance Aus­tria and Ireland participate actively in the GGE talks in Geneva and in NATO’s Part­ner­ship for Peace (PfP). Through this latter for­mat they could be included in the substantive discussion on AWS. Russia also partici­pates in the PfP, but cannot be expected to contribute meaningfully in the foreseeable future. In light of current events that would not be welcomed either.

An open and informal exchange among NATO states would create a good opportu­nity to keep the AWS discussion going. But a discussion between NATO states and will­ing non-NATO states could also con­tribute to agreeing a shared line. The disadvantage would be that NATO would not be holding a dialogue on AWS with all states, but pri­or­itising cooperation with particular third states.

Germany should also press for a broader dis­cussion within the EU, where divergences on AWS still exist. If the member states were able to agree on a joint position the EU could lead the way on international standard-setting and expect others to follow (the “Brussels effect”). This has already occurred in the field of cybersecurity. The European Parliament has discussed the issue of AWS several times and calls for legally binding regulation, while the Euro­pean Defence Fund does not support “actions for the development of lethal autono­mous weapons without the possi­bility for mean­ing­ful human control over selection and engagement decisions when carrying out strikes against humans”.

Discussions directed towards finding a shared European position would be espe­cially useful for the Permanent Structured Cooperation (PESCO). PESCO was launched in 2017 to allow groups of willing member states to coordinate activities and forward planning in areas related to the Common Foreign and Security Policy (CFSP). The twenty-five PESCO states are already work­ing on autonomous functions for weapons systems in joint defence projects. For exam­ple, in an Estonian-led project Germany and nine other EU member states are develop­ing an unmanned ground system. Similar projects are also working on naval and aero­space technologies, often with German involvement. For example an AI-based un­manned anti-submarine-system is being developed under Portuguese leadership.

The extent of autonomous functions in these weapons systems is not entirely clear from the public data. But closer investigation of these systems and the question of human/machine interaction could certainly supply valuable insights and provide mean­ingful input for other forums. Even if PESCO’s focus is on arms development, it is never­theless conceivable and above all pro­ductive also to conduct substantive discus­sions that could subsequently positively influence other processes for example in NATO and the UN.

United Nations – the normative power of resolutions

Even if the GGE has been the central forum for discussions on regulating AWS, it is quite possible that the process could shift to other institutions within the UN. Believing that the GGE has failed, the NGO Campaign to Stop Killer Robots declared in March 2022 that “an alternative process of legal devel­op­ment is now inevitable”. The Latin Ameri­can, African and Asian states of the G-13 group in the GGE would also like to see other UN institutions becoming in­volved. Germany needs to be prepared for such developments and should consider whether and on what terms it would par­ticipate in such a process. Realistically speaking, two UN forums offer potential alternatives:

The first of these is the UN General Assem­bly’s Committee on Disarmament and International Security, also known as the First Committee. Its resolutions are normally adopted unanimously, as in the GGE, so states that disagreed would be able to block any decision. However, non-unani­mous preparatory processes in the could advance the discussion in the General As­sembly, or even lead to the General Assem­bly itself passing a resolution on AWS. The General Assembly is the other UN forum in which AWS could be discussed. Its decisions are non-binding and made by majority vot­ing, so resolutions can be adopted against the resistance of particular states.

The experience of the GGE talks shows that most states favour regulation of AWS. Even China explicitly underlined its sup­port in a position paper last autumn. How­ever, it has also become apparent that the sticking point is not lack of interest but disagreements concerning the rules for human/machine interaction. One benefit of discussing AWS in the General Assembly could lie in the greater political impact of General Assembly resolutions – although there would still be a real risk of failure to make meaningful progress.

If a General Assembly resolution is to generate political momentum it will need to attract broad support. That could prove problematic in light of the current conflicts. For example China might refuse to vote for a resolution that was not supported by Rus­sia. Obviously any resolution on AWS will need the support of all the major powers if regulation is to be credible and enduring.

Recommendations for Germany

It is very likely that the GGE will be declared a failure after this year’s second meeting in July. If Germany is to live up to its com­mit­ment to work to outlaw AWS it needs to develop a strategy for a time after the GGE.

That means preparing a national position on AWS, which should distinguish be­tween fully and partially autonomous weap­ons systems and provide specific guid­ance on the requirements for human/machine in­teraction. It would certainly make sense to include non-lethal systems too. To ad­dress the refusal of the United States and others to accept the term “human control”, Ger­many could sign up to the ICRC ap­proach of speaking of “human control and judge­ment”. The planned German National Secu­rity Strategy could also contain a sec­tion on AWS and new weapons technologies, per­haps modelled on the Swiss Arms Control and Disarmament Strategy 2022–25.

Russia’s invasion of Ukraine appears to mark a paradigm shift in the realm of arms control. Universal forums have been suc­ces­sively sidelined, while regional institutions such as NATO and the EU gain in impor­tance. But AWS issues cannot be settled in regional forums, certainly not inclusively. Yet discussing them within NATO and the EU could at least coordinate partners’ stances more closely and in the medium (and longer) term enable them to enter broader future talks with a single voice.

Germany should also prepare for a dis­cussion on AWS in the First Committee, which the G13 states in particular might initiate. A coordinated approach by the NATO partners and the EU member states could contribute to finding broad majorities in the individual UN institutions.

Under current circumstances none of the potential forums guarantees success. But it would be pointless to allow that to paralyse the AWS discussion. The deployment of new technologies in the war in Ukraine un­der­lines the growing importance of auton­o­my in weapons systems. The announced de­fence spending plans should give occa­sion to reflect on the complexity of auton­o­my in weapons systems, and to develop a national position on AWS. Germany could show the world that it can be counted on to keep its political promises.