AI shows promise as a tool for diplomacy
AI is on everyone's lips. Can it be a useful tool for diplomacy? Where could the limits and risks be? We looked at these questions in a comprehensive study commissioned by the Swiss Foreign Ministry - with a focus on international negotiations.
Our methodic approach divided the landscape of digital tools into two main categories. Firstly, classical quantitative methods, algorithms and software that operate without the capability of self-learning. Secondly, the more advanced digital tools that are equipped with self-learning capability, which we label AI-methods. This categorisation was pivotal in understanding the contrasting capabilities and applications of these technologies in diplomatic contexts. Further, we categorised the type of negotiations according to three different dimensions to enable a systematic analysis: the complexity of the negotiation (from simple to complex), the technicality (from technical to political content), and the structural framework (bilateral and multilateral formats).
Our research included a series of real and hypothetical case studies to empirically evaluate the potential of digital methods. On the one hand, we tested the AI-methods which have self-learning capability. In one case, for example, we utilised the ChatGPT-4 model to formulate a cooperation charter involving the Arab Gulf states, Iran, and Iraq. A comparison revealed that integrating a manually written text with one generated by ChatGPT can yield a synthesis of high quality. Notably, while the text written by human hands exhibited more creative elements, the version produced by ChatGPT stood out for its detailed and comprehensive nature. A related case, where ChatGPT was used to draft a memorandum on nuclear co-operation, showed a similar pattern and also met diplomatic standards. In another case study, ChatGPT was tasked with drafting resolutions for the Human Rights Council. This case illuminated the model's efficiency in expediting document preparation, albeit highlighting the necessity of human oversight to mitigate inaccuracies. The limitations of AI became particularly clear when the ChatGPT model was tasked with developing a new dispute resolution procedure, which is one of the many issues in the ongoing negotiations between Switzerland and the EU. This case illustrated the difficulties of AI in dealing with complex technical negotiations, as the algorithm did no creative intellectual work, but merely confirmed what had been entered in advance as possible ideas.
On the other hand, we evaluated classical quantitative methods without self-learning capabilities. First, we applied a tool to calculate the Pareto front in World Trade Organization (WTO) negotiations. With the help of this tool, the negotiators can determine whether their provisional solutions are theoretically the best possible ones. The main challenge is the need for a quantitative assessment of negotiation outcomes, which is essential for the algorithm's functionality. Second, we tested a tool designed to aid UN budget negotiations. It enables delegates in the UN's 5th Committee to immediately understand the impact of budget change proposals on their positions, enhancing budget planning and control during negotiations. Both tools, though non-AI, have shown considerable utility in aiding complex international negotiations by providing clear, quantitative insights.
From these diverse case studies, several key insights emerged. (i) AI-tools with self-learning capacities are very promising for less complex negotiations, but face challenges in scenarios that require creative and complicated solutions. (ii) In multilateral negotiations, the use of AI-methods might be limited due to a lack of traceability. (iii) Classical quantitative methods, in contrast, are ideal for technical negotiations where quantifiable aspects are predominant. (iv) Overall, both classical and AI-methods have the potential to significantly support foreign policy negotiations.
Given the rapid developments in the field of digital technologies, particularly AI, it is imperative that foreign ministries do not miss out on these advances. The effective utilisation of new technologies hinges on the availability of requisite expertise and know-how, which requires staff training. Furthermore, the strengthening of collaborations with research institutions and scientific organisations to harness technological potential is crucial. Finally, the continuation of exchange and dialogue with other foreign ministries is recommended (as long as the parties do not compete in specific negotiations).
AI is proving to be a valuable tool in diplomacy, provided it is supported by experienced professionals with critical thinking capability and creativity. This synthesis not only enables the undoubtedly considerable potential of AI to be optimally utilised, but also serves to minimize risks and errors.
Dr. Michael Ambühl is the former Swiss State Secretary for Foreign Affairs.
