Imagine being a fly on the wall during a high-stakes negotiation, instantly deciphering every tactic, every subtle cue. That's the promise of the Vanderbilt AI Negotiation Lab's groundbreaking research, which uses the power of large language models (LLMs) to analyze negotiation transcripts. Negotiation research often relies on painstaking manual coding of conversations, a process that's both time-consuming and costly. This research explores how AI can automate this process, opening doors to faster, more efficient analysis. The team experimented with various AI strategies, starting with simple “zero-shot” learning, where off-the-shelf LLMs tried to code negotiation tactics with no specialized training. Unsurprisingly, this approach failed, highlighting the need for AI to understand the nuances of negotiation language. Next, they “fine-tuned” an LLM by training it on a specific coding scheme, leading to a significant improvement. However, the real breakthrough came with “in-context learning.” This method feeds the LLM a massive prompt containing code definitions and example transcripts, making it much more accurate. The team also discovered that training AI on idealized sentences wasn’t enough. Real-world conversations are messy and ambiguous, requiring training on real transcripts. This led to an innovative approach where the model learns from human-coded transcripts, adapting to the specific coding scheme used. The model even assesses its own consistency by coding each sentence multiple times and only reporting results if it reaches a certain level of agreement. While the initial results are promising, the researchers acknowledge that AI isn't perfect. Human coders still play a vital role in validating the AI's output and resolving discrepancies. The project's biggest takeaway? In the fast-paced world of AI, flexibility is key. Be prepared for unexpected challenges and embrace the constant evolution of technology. This research not only offers a powerful tool for negotiation analysis but also provides a roadmap for applying AI to other fields that rely on coding textual data, like medical record summarization or financial report analysis.