Context:

Artificial intelligence (AI) is being increasingly explored as a tool for foreign policy decision-making, with institutions like the Center for Strategic and International Studies (CSIS) experimenting with its potential to resolve international conflicts. AI models are being trained and tested on scenarios involving deterrence and crisis escalation, revealing significant variability in their responses based on the biases of their creators. While some AI models lean towards pacifism, others show a tendency for escalation, highlighting the importance of aligning these tools with institutional approaches. AI's potential to expedite negotiations and monitor ceasefire compliance is promising, yet challenges such as unintended outputs and the inability to replicate human connections remain. Experts believe that while AI can assist in diplomatic tasks, it cannot fully replace the nuanced understanding and personal relationships that human negotiators bring to the table.

Dive Deeper:

• The Futures Lab at CSIS, with funding from the Pentagon, is experimenting with AI technologies like ChatGPT and DeepSeek to transform diplomatic practices, including crafting peace agreements and monitoring ceasefire compliance.

• A study conducted by the lab tested AI models with scenarios of deterrence and crisis escalation, finding that different models, such as OpenAI's GPT-4o and Antropic's Claude, tend to favor peaceful solutions, while others like Meta's Llama and Google's Gemini lean towards more aggressive actions.

• The variability in AI responses is attributed to the biases of the software developers, leading to the need for models to be assessed and aligned with specific institutional approaches to avoid unintended consequences.

• An interactive program called 'Strategic Headwinds' has been developed by CSIS to aid in negotiations, using AI trained on historical peace treaties and news to identify common grounds for ceasefire in the Ukraine conflict.

• Challenges with AI in diplomacy include its lack of human touch, difficulty in assessing long-term consequences, and the need for training to understand complex geopolitical language, as demonstrated by its misinterpretations in Arctic deterrence scenarios.

• Experts caution against over-relying on AI for diplomacy, as it may not effectively address issues in closed societies and lacks the ability to replicate the personal connections vital in negotiations.

• Despite these challenges, AI is seen as a tool that can augment human efforts in diplomacy by automating certain tasks and providing insights, but it cannot fully substitute human judgment and relationships.