Mind's Eye: Grounded Language Model Reasoning through Simulation
This addresses the issue of physical reasoning errors in language models for AI communication and robotics applications, representing a novel approach rather than an incremental improvement.
The paper tackles the problem of language models lacking grounded physical reasoning by introducing Mind's Eye, a paradigm that uses a physics engine to simulate outcomes and integrates them into LM inputs, resulting in large accuracy improvements (e.g., 27.9% zero-shot and 46.0% few-shot gains) and enabling smaller models to match much larger ones.
Successful and effective communication between humans and AI relies on a shared experience of the world. By training solely on written text, current language models (LMs) miss the grounded experience of humans in the real-world -- their failure to relate language to the physical world causes knowledge to be misrepresented and obvious mistakes in their reasoning. We present Mind's Eye, a paradigm to ground language model reasoning in the physical world. Given a physical reasoning question, we use a computational physics engine (DeepMind's MuJoCo) to simulate the possible outcomes, and then use the simulation results as part of the input, which enables language models to perform reasoning. Experiments on 39 tasks in a physics alignment benchmark demonstrate that Mind's Eye can improve reasoning ability by a large margin (27.9% zero-shot, and 46.0% few-shot absolute accuracy improvement on average). Smaller language models armed with Mind's Eye can obtain similar performance to models that are 100x larger. Finally, we confirm the robustness of Mind's Eye through ablation studies.