System Dynamics Game-Theoretic Model of the AI Development Race
Ariel Gil · Team System Dynamics BCN
Submitted to The AI Forecasting Hackathon. Sprint projects are early-stage work by participants, not Apart Research publications.
A Game theoretic / System Dynamics model of the race dynamics of the US, China, and EU, as a follow up to the Armstrong et al. (2016) paper “Racing to the Precipice”. We find preliminary results where knowledge of competitor capabilities delays the decision to race, contradicting the preceding paper.
Reviews
This is a very exciting topic, I’m glad to see an extension of Armstrong et al. 's model to make it more dynamic.
Some more interpretation of the results would be beneficial - I found it difficult to get useful takeaways from this by looking at the plots, and the current analysis a bit scant.
Assuming the EU as a serious actor in AI development is a bit odd, and would warrant some more justification. I’d also be excited to see a dynamic simulation of just the US and China (two actors).
Future work could explore how enabling aggressive actions could change the outcome. For example, currently, the model doesn’t take into account the possibility of sabotage of others’ AI projects.
I think this is really exciting work that addresses a concrete risk scenario of the AI arms race. The simulation setup seems sound overall. You could possibly look into further exploring how robust your findings/conclusions are against the hyper parameters used, as well as going further into what has changed since the Armstrong (2016) paper.
Cite this project
@misc{gil2025system,
title = {{System Dynamics Game-Theoretic Model of the AI Development Race}},
author = {Ariel Gil},
year = {2025},
month = nov,
note = {Submitted to The AI Forecasting Hackathon, an Apart Research Sprint},
howpublished = {\url{https://apartresearch.com/sprints/projects/system-dynamics-gametheoretic-model-of-the-ai-development-race-dqe3}},
url = {https://apartresearch.com/sprints/projects/system-dynamics-gametheoretic-model-of-the-ai-development-race-dqe3}
}More from The AI Forecasting Hackathon
- View project: ExogenousAI
ExogenousAI
Fibonacci
Current AI capability forecasting methodologies, including EpochAI's Direct Approach and Biological Anchors framework, primarily rely on internal metrics such as training compute and scaling laws while assuming stable …
- View project: AI Incidents Forecasting
AI Incidents Forecasting
KLACE
This research develops a framework for forecasting AI incidents to help predict future risks. We have developed two models that forecasts incidents which include calibrated 90% prediction intervals with backtests. These …
- View project: AI Shared Socioeconomic Pathways
AI Shared Socioeconomic Pathways
AI & Climate change
We propose adapting the IPCC's Shared Socioeconomic Pathways, the standard framework for climate scenario planning, to AI forecasting, exploring how alternative socioeconomic trajectories might shape AI development and …