Led by PhD candidate Clemens Bartnik and computational neuroscientist Iris Groen, the research team examined how people evaluate potential movements in different settings. Using MRI scans, participants viewed photos of diverse environments and indicated whether they would walk, swim, drive, or climb in each. While they responded, the researchers monitored activity in the visual cortex.
The results showed distinct patterns of brain activation not linked to visible objects alone. Groen explained, "These brain areas not only represent what can be seen, but also what you can do with it." Remarkably, these responses occurred even without explicit prompts-implying the brain automatically assesses action potential.
This is the first time affordances have been directly observed as a neurological process rather than just a psychological theory. The discovery marks a significant step in understanding how humans blend perception with motor possibilities.
To assess AI's grasp of these affordances, the team tested multiple models, including image recognition systems and GPT-4. While AI could mimic some action judgments when trained specifically for the task, its internal representations diverged significantly from human brain activity. "Even the best AI models don't give exactly the same answers as humans, even though it's such a simple task for us," Groen said.
She attributes this gap to the fundamental difference in embodiment. "We connect our perception to our experience in a physical world. AI models can't do that because they only exist in a computer."
Groen believes this work could influence the next generation of AI, particularly in applications like robotics and autonomous navigation. "Machines need to understand not only what something is, but also what it can do," she emphasized.
The study also raises sustainability concerns. Today's AI training methods demand enormous computational power. Insights from the human brain, which processes information with high efficiency, may offer blueprints for building smarter, more economical, and human-aligned systems.
Research Report:Representation of locomotive action affordances in human behavior, brains, and deep neural networks
Related Links
Universiteit van Amsterdam
All About Human Beings and How We Got To Be Here
| Subscribe Free To Our Daily Newsletters |
| Subscribe Free To Our Daily Newsletters |
