In the 1980s, AI researchers encountered a baffling contradiction. While computers were quickly mastering complex logic like chess and calculus, they struggled to perform tasks any one-year-old child could do: recognizing a face or lifting a cup without crushing it.
This phenomenon, known as Moravec's Paradox, suggests that high-level reasoning requires very little computation, but low-level sensorimotor skills require enormous computational resources. It fundamentally shifted our understanding of biological and artificial intelligence.
The "Hard" Problems are Actually Easy
From the perspective of early computer science, we assumed intelligence was defined by the ability to solve abstract problems. We built machines that could calculate trajectories and beat Grandmasters. These tasks, though mentally taxing for humans, follow strict rules that are easy to program.
Paradoxically, the tasks we consider intellectually "superior" are computationally trivial for AI:
- Symbolic Logic: Performing complex algebra or geometry proofs is straightforward for rule-based systems.
- Data Analysis: Analyzing millions of stock market transactions happens in milliseconds.
- Strategy Games: Chess and Go have finite states, making them solvable through brute force computation.
The Evolutionary Explanation
Hans Moravec offered an evolutionary answer: We have spent millions of years evolving systems for perception and movement. These processes are heavily optimized and run subconsciously. Logic, however, is a new "thin layer" of human evolution.
- Motor Skills: Walking on uneven terrain requires thousands of real-time adjustments per second.
- Perception: Distinguishing a cat from a dog in poor lighting is an immense pattern-matching challenge.
- Dexterity: Folding a towel involves complex physics prediction that baffles most robots.
🚀 Implications for the Future
This paradox explains why we have AI that can write poetry (LLMs) but we still don't have reliable robot butlers. The "digital" world is easy to simulate; the "physical" world is messy, unpredictable, and computationally expensive.
To reach true Artificial General Intelligence (AGI), AI must master the physical world. This is the new frontier of "Embodied AI"—moving intelligence out of the server and into the sensor.
"It is comparatively 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."