When you play a video game, your brain seamlessly blends reasoning—like figuring out the next move—with action, such as pressing a button. A new study shows that artificial intelligence models trained to focus on action mimic this dual process in surprising ways, and that one type of AI is particularly good at predicting activity in your brain’s motor-planning areas.
The Research
Researchers at IIT Hyderabad and Microsoft (lead author Subba Reddy Oota) scanned the brains of participants playing Atari-style video games using fMRI. They compared how well two types of AI models—vision-language models (VLMs) that combine images and text, and large-action models (LAMs) that specialize in generating actions—could predict neural activity. Both models were prompted with either action-focused or reasoning-focused descriptions of the game state.
The results, published on arXiv (2605.19352), show that VLMs and LAMs both outperform reinforcement-learning baselines in predicting brain activity, even when controlling for model complexity. However, prompt-driven improvements were not uniform across the brain: frontal-parietal and motor-planning regions saw the largest gains (up to ~27% unique variance for action prompts in LAMs), while early visual cortex improved only half as much.
Strikingly, variance partitioning revealed that VLMs are “prompt-symmetric”: action and reasoning prompts each explained about 13% of unique variance. LAMs, in contrast, were “prompt-asymmetric”: action prompts accounted for 27% unique variance, while reasoning prompts explained almost none (−5%). This asymmetry was strongest in frontal-motor cortex, suggesting that action-specialized fine-tuning reorganizes the model’s representations to align with human action-related neural computations.
Why It Matters
This study reveals that your brain doesn’t treat reasoning and action as separate modules—they are tightly integrated, but action-specialized AI can capture the motor-planning aspects more precisely. For anyone interested in cognitive training, this implies that actively engaging in tasks that require both planning and executing—like video games or sports—may strengthen the neural circuits that underlie action-reasoning integration.
What You Can Do
Try playing strategy-based video games that demand quick decisions and precise actions. This combination may enhance your ability to link thought to movement, potentially improving reaction time and planning skills. Even 15 minutes a day could stimulate the frontal-parietal network highlighted in this research.
Source: arXiv q-bio.NC
Curious about your own brain? Take our free adaptive IQ test or try 306 brain training levels.
