CONCEPT
Physical Intelligence
The competence to act in the three-dimensional, dynamic, unforgiving physical world—the capacity that a two-year-old takes for granted and that remains the most unsolved frontier of artificial intelligence.
Physical intelligence is the term
Daniela Rus uses for the competence that the AI field’s recent triumphs have not touched: the capacity to act competently in the messy, unpredictable, three-dimensional world that living organisms navigate without conscious effort. A two-year-old can pick up an unfamiliar object, adjust her grip when it turns out to be heavier than expected, and cross a room strewn with toys without stepping on any. No robot can reliably do all of this. The tasks that are hardest for artificial systems are often the ones trivial for a toddler or a bird, because the physical world demands something categorically different from the statistical fluency of
large language models: real-time adaptation to a domain that is dynamic, uncertain, and unforgiving, where decisions must be made now and mistakes have physical consequences. Physical intelligence is not a more demanding version of text intelligence; it requires fundamentally different architectures, differently embodied systems, and a humility before biological solutions that the symbolic and neural-network traditions have often