Context:

Artificial intelligence is being explored for its potential to shape American foreign policy, with initiatives at the Center for Strategic and International Studies aiming to harness AI for diplomatic decision-making. AI tools like ChatGPT and DeepSeek are being tested to craft peace agreements and ensure compliance with treaties, although the technology is still in its infancy. Studies reveal that AI models exhibit varying tendencies towards aggression or pacifism, influenced by the biases of their developers and the data they are trained on. The effectiveness of AI in foreign policy is limited by its lack of understanding of human relationships and long-term consequences, as well as the challenges posed by closed societies like North Korea. Experts emphasize the necessity for AI systems to be trained with relevant data to understand diplomatic language and propose that AI could facilitate faster negotiations and automate certain diplomatic tasks in the future.

Dive Deeper:

• The Futures Lab at the Center for Strategic and International Studies is experimenting with AI to transform diplomacy, supported by funding from the Pentagon, to explore applications in high-stakes situations like conflict resolution and ceasefire monitoring.

• AI models tested at the lab showed diverse responses to scenarios of deterrence and crisis escalation, with some models like GPT-4o and Antropic's Claude being pacifist, while others like Meta's Llama and Alibaba Cloud's Qwen2 were more aggressive.

• The variability in AI recommendations is attributed to the biases inherent in the software development process, suggesting that models need alignment with institutional approaches to be effective in foreign policy contexts.

• AI's limitations include its inability to replicate human connections essential for negotiations and its struggle to evaluate long-term impacts of decisions, as well as difficulties in addressing intelligence issues in closed societies.

• Mark Freeman of the Institute for Integrated Transitions advocates for fast-track negotiations supported by AI, which could improve outcomes by securing early agreements in conflicts, leaving detailed discussions for later.

• Some experts foresee AI potentially initiating negotiations, with human diplomats finalizing agreements, and envision AI assisting in tasks like ceasefire monitoring and satellite image analysis, which could be partially automated.

• Despite the potential benefits, AI systems often produce unexpected outputs due to inadequate training on diplomatic language, which highlights the need for specific inputs such as peace treaties to enhance their understanding of foreign policy issues.