{
  "id": 19215991,
  "title": "Insight about how 'nature makes no leaps' promises better robot predictions",
  "url": "https://urgent.news/2026/11/09/insight-about-how-nature-makes-no-leaps-promises-better-robot",
  "topic": "tech",
  "section": "Tech",
  "published": "2026-11-09T17:00:00.000Z",
  "source": {
    "name": "The Register Science",
    "slug": "the-register-science",
    "url": "https://www.theregister.com/ai-and-ml/2026/11/09/insight-about-how-nature-makes-no-leaps-promises-better-robot-predictions/5308301"
  },
  "original_language": "en",
  "account": "Researchers at Brown University, NYU, and AMI Labs have created a technique to help AI systems better predict the behavior of objects in the real world. Machine learning world models typically rely on observed data to predict the consequences of actions, but they cannot infer possibilities without extensive training. The researchers' method, detailed in a pre-print paper titled \"Counterfactual World Models: Natur Non Facit Saltus,\" allows models to make predictions about alternative scenarios based on observed outcomes.\n\nThe key to this approach is the concept of counterfactuals, which refers to the idea of considering what would have happened if a different action had been taken. The technique uses the discrepancy between a model's prediction and reality to identify mistakes and improve future predictions. By applying this method to any world model that predicts the average next state under any action, the researchers were able to enhance the ability of world models, such as Joint Embedding Predictive Architecture (JEPA), to anticipate counterfactual scenarios.\n\nIn their experiments using off-the-shelf encoders and a small predictor, the researchers found that the counterfactual world model could predict the effects of actions more accurately than traditional models. For example, when simulating a boat crossing a river with an unseen current, both the traditional and counterfactual models predicted the boat's final position. However, the counterfactual model provided a more accurate estimate by accounting for the actual effect of the current.\n\nThe counterfactual world model has potential applications in various scenarios, such as allowing a robot to retry an action after encountering an unexpected obstacle or helping a vehicle adapt to persistent faults like a miscalibrated motor. By providing insights into what would have happened under different circumstances, these models can assist in debugging failures, distinguishing between bad decisions and bad luck, and guiding future actions for improved performance.",
  "summary": "Leibniz lives!",
  "key_points": [],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 1,
    "also_reported_by": []
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}