Catastrophic AGI Risk Prediction

Catastrophic AGI Risk Prediction

Mitchell Reynolds & Jon Menaster

We introduce a custom case scenario based on Open Philanthropy’s AI Worldviews Contest. We estimate a 48% probability of an existential catastrophe from the loss of control of an AGI system. Our approach suggests a starting point for forecasting AGI risk through a programmatic tool. Our future work will use the ITN framework to assign weights to a variety of independent variables within technical and governance research in order to improve the ability of forecasters to accurately predict the extent to which AGI existential catastrophe will occur within a given timeframe.