Attached Paper In-person November Annual Meeting 2025

The Thoughts of AI

Abstract for Online Program Book (maximum 150 words)

In the summer of 2024, AI development appeared to stagnate with delays in major model releases and concerns about training data limitations. However, the emergence of Deepseek's reasoning model, deepseek-r1, revolutionized AI research by introducing a new "Reasoning Space" component. Unlike traditional transformer LLMs, which operate as black boxes producing token-by-token responses, reasoning models provide transparency into their decision-making process through accessible "reasoning traces."

This advancement enables researchers to examine how AI systems arrive at their conclusions. Building on this technology, this project aims to create AI instances replicating different Gospel versions of Jesus (Mark, Matthew, Luke, John, and Thomas) using reasoning models. The goal is to analyze how these AI representations think differently and whether their reasoning can evolve through interaction with each other. This research could potentially extend to dialogues with other AI-simulated historical or philosophical figures, offering new insights into AI reasoning and simulation capabilities.