Simulation-based inference with deep learning suggests speed climbers combine innovation and copying to improve performance

In the Olympic sport of speed climbing, athletes compete to reach the top of a 15-meter wall as quickly as possible. Since the standardization of the speed climbing route in 2007, improvement has been driven by a process of cumulative cultural evolution—new route sequences innovated by some are copied and improved upon by others. In this study, we use simulation-based inference to fit an agent-based model of speed climbing to 12 years of competition times (2007-2019). In the fitted model, innovation and copying are used roughly equally by climbers, with copying having only a slight advantage. Slower agents are more likely to innovate, likely in pursuit of strategies that give them a competitive advantage. Population size negatively predicts innovation, presumably because innovation is not as useful when there are plenty of existing solutions to choose from. An additional analysis of real route sequences from 2012-2019 World Championships suggests that other factors, like height and weight, influence the use of some holds, but may not affect performance or the probability of adopting innovations. Finally, the model suggests that climbers may not be using innovation and copying optimally—continued improvement may require more emphasis on innovation and less reliance on copying.