AI Model Predicts Human Planning Failures in Clinical Test
AICON, a reactive gradient-descent framework initially created for robotic manipulation, was utilized by researchers to tackle the Tower of London test, which assesses cognitive function in individuals with Parkinson's disease, mild cognitive impairment, and those who have suffered a stroke. Remarkably, AICON achieved a nuanced ordering of difficulty across 24 tasks without relying on lookahead planning or insights into human cognition, surpassing structural task parameters and generalizing effectively to unseen problems. Notably, AICON excelled compared to a planning baseline in groups with diminished planning abilities, whereas the planning baseline was more effective for healthy controls—an outcome anticipated by the original AI model.
Key facts
- AICON is a reactive gradient-descent framework for robotic manipulation.
- Applied to Tower of London test for planning assessment.
- Tower of London test used in Parkinson's, mild cognitive impairment, and stroke.
- AICON reproduced human difficulty ordering across 24 problems.
- Outperformed structural task parameters.
- Generalized to held-out problems in leave-two-out evaluation.
- AICON outperformed planning baseline for reduced planning capacity groups.
- Planning baseline better captured healthy controls.
Entities
—