Inclusion of diverse voices in international negotiations is long due, but progress is very slow. Daniela Lup writes that the need for inclusion became even more urgent when an artificial intelligence agent created by Meta achieved human-like performance in an online game involving strategic negotiation.
Two pieces of news caught my attention in the past month. The first one relates to the low representation of women at the COP27 climate negotiations—hardly news, though, for researchers and representatives of civic society who have long advocated for gender balance in international affairs. The second one is about Cicero, an artificial intelligence agent created by Meta that has become the first AI to achieve human-like performance in an online game involving strategic negotiation.
Taken together, these two readings, unconnected as they might seem, help us make a strong argument that we need balanced gender representation in international negotiations.
Gender representation in international climate conventions
Gender representation in international bodies and conventions has improved over time, but progress remains slow. A detailed report prepared by the United Nations Framework Convention on Climate Change (UNFCCC) ahead of COP27 describes in detail the state of gender representation in previous climate events. For the first time, the report presents figures related to voice diversity, including the length of interventions by women and men. It shows that women remained severely under-represented in some of the most important groups established under the Convention and its Kyoto Protocol. For instance, women represented only 10 per cent of the Clean Development Mechanism (CDM) Executive Board, the body connected with the highly debated carbon offset scheme. The report also shows a dramatic decrease in female representation in the Consultative Group of Experts (CGE), from 52 to 33 per cent. CGE is the body assisting developing countries to fulfil their reporting requirements.
An important figure that captures the under-representation of women in decision-making aspects of the UNFCCC processes relates to the composition of party delegates, who come from all the participating states, with multiple delegates per state. The report shows that only a third of party delegates are women, a stable number across years, but with a slight downward trend. The same trend is present among heads and deputy heads of delegation. Moreover, the percentage varies across age groups in ways that suggest that gender and age need to be viewed from an intersection perspective: while about a third of delegation leaders are women in the age 35-55 category, the percentage falls to an average of about 15% for women aged over 55.
Recognising that influence in negotiations does not depend only on being present in the decision space, but also on being heard, the report includes information about speaking time for men and women. It is the first time that speaking time is measured and reported. The numbers show that while 37 per cent of party delegates were women, they accounted for only 29 per cent of total speaking time. Surprisingly, the number is even lower when considering the speaking time of chairs and facilitators.
What happens when women’s voices are not heard
Some of the existing research suggests that women policymakers tend to voice more gender-equality claims and that the presence of more women in decision-making bodies is associated with more transnational aid for women. Instead, when in minority, women seem compelled to conform and support issues and positions favoured by male decision-makers. These and other similar studies suggest what might be the hidden mechanism through which under-representation and reduced voice affect the norms of the negotiation game and its outcomes. But none of them is able to follow in detail how a negotiation party learns the rules of the game, and what they learn in games in which some stakeholders are under-represented.
The news on Meta’s AI agent sheds light on negotiations
The announcement by Meta that an AI agent has achieved human-like performance in a negotiation game is certainly impressive. The game, called Diplomacy, is set in a pre-World War I Europe and encourages players – who represent the seven major powers at the time – to negotiate as a group or in smaller coalitions to gain control over a total of 18 supply centres. Diplomacy appeared to have been a favourite of President John F. Kennedy and former Secretary of State Henry Kissinger and is currently played as a board game and online.
Two breakthroughs that elevate Cicero to human-like negotiation performance are worth highlighting:
First, Cicero appeared to have mastered perspective-taking, a cognitive ability that involves considering the other party’s point of view. Anyone involved even in basic negotiations would be familiar with the attention they need to pay to the other party, and the complex work of eliciting information about them to understand their interests, motivation, and needs. Perspective-taking becomes much more complicated in multi-party negotiations, in which one must consider not only the unique interests of each party, but also the motivation behind coalitions that tend to form when multiple actors negotiate.
It is good news that Cicero can take perspective with perhaps more ease than a human – think about our time constraints and difficulty to focus and analyse complex interactions. Like the game of Diplomacy, international negotiations are complex, with multiple parties and interests, and one can imagine a future in which AI agents would support humans in more carefully considering and understanding the positions of others.
Second, Cicero has learned to use natural language in a persuasive way to convince counterparties to collaborate with the agent. The learning process is similar to how humans learn from other humans with whom they come in contact: by listening to how and when they speak, by noticing how words and phrases resonate with others, and learning from feedback, implicit or explicit. According to researchers from META, the agent has become “so effective at using natural language to negotiate with people in Diplomacy that they often favoured working with Cicero over other human participants”.
In negotiations, like in the Diplomacy game, how a participant talks to other parties is often more important than the specific proposals that they advance. The corollary to this is that any negotiator who wants to be taken seriously and accepted as a legitimate party at the negotiation table should learn the language spoken at the table. In turn, the rules of speaking and interacting are based on previous iterations of the negotiation process. Cicero has clearly earned a reputation as a player that understands how to speak the language of the Diplomacy game.
Learning to speak and listen to men only might appear as a mere footnote, and indeed the detail is not included in the Meta announcement. Nor does it appear in the main research paper. It is included in the supplementary files to the published paper, in the form of a short note and a recommendation for future research on board games: “This usage of masculine pronouns by the AI agent might lead to misgendering another player or contribute to harmful stereotypes about gender and board games. We recommend that future work in the use of natural language in board games be mindful of the potential for AI agents to perpetuate such representational harms.”
But this mere detail carries implications beyond board games. Cicero’s ability to learn to negotiate ‘like a human’, including its learning of the subtle norms of communication that makes it a legitimate party at the Diplomacy negotiation table is a telling illustration of what might happen in international decision-making settings in which women and the specific issues that they represent are in the minority. With women’s voice barely heard in some of the bodies and conventions most critical for climate change decisions, gender specific issues are less likely to appear or unlikely to occupy much of the discussion time.
One wonders what Cicero would learn and do if thrown in the real game of climate negotiations?! The answer can shed much needed light on the actual rules of various international negotiation games and help us correct some of the imperfections embedded in our international institutions.