Moravec’s paradox
It’s an idea that comes from 1988, written by Hans Moravec.
Moravec addresses a counterintuitive finding in AI and robotics: the tasks that require the most effort and computation in AI research are often the ones humans (and even small children) perform effortlessly, and vice versa.
It is easy to make computers exhibit adult-level performance on intelligence tests or playing checkers, and difficult or impossible to give them the skills of a one-year-old when it comes to perception and mobility.
Moravec’s Hypothesis
Moravec’s hypothesis for this paradox is based on natural selection and evolutionary timeline.
Ancient skills (difficult for AI): skills related to sensory input, motor control and spatial reasoning (walking or recognizing objects) have been refined through biological evolution over hundreds of millions of years. These processes are so deeply embedded and efficient that we are no longer consciously aware of the massive computational effort required.
Recent skills are (easy for AI): abstract reasoning, logic, mathematics, and symbolic manipulation are relatively new additions to the human cognitive toolkit, evolving primarily in the last few thousand to million years. Consequently, their biological implementation is less efficient and requires conscious, deliberate thought.
Modern Context
The paradox remains fundamentally true, even though deep learning has significantly softened the edge of the paradox since 2012:
- Computer Vision can now recognize faces, objects and scenes with a near-human accuracy, in 80s it was ofocourse impossibly difficult.
- Embodied AI: advancements in reinforcement learning are allowing robots to learn complex motor skills through trial and error, mimicking the evolutionary learning process.
It is no longer impossible for AI to solve perception but it still requires vast amounts of data and computational power that greatly exceed the demands of a complex game like chess.