A new study from MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL) demonstrates a method for training AI models to be more robust and generalizable by learning from ‘counterfactual’ examples. The research focuses on teaching models to understand causal relationships, not just correlations, by exposing them to scenarios where small, irrelevant changes are made to input data. This approach, tested on image classification tasks, helps prevent models from relying on spurious features, such as background textures, and instead focus on the core object. The technique shows promise for creating more reliable AI systems that perform better in real-world, unpredictable environments. For more details, read the full article at https://technologyreview.com/2024/05/15/1099875/ai-counterfactual-training-robustness.