Moravec's Paradox

In 1988, roboticist Hans Moravec made a puzzling observation: AI could beat chess grandmasters, prove mathematical theorems, and optimize complex logistics. But it couldn't walk across a room, recognize a face, or catch a ball.

Tasks that require years of human education - calculus, chess, logical reasoning - were relatively easy to program. Tasks that every toddler masters - walking, grasping objects, understanding social cues - remained almost impossibly hard.

This isn't a quirk of AI design. It's a fundamental insight about what intelligence actually is and how evolution shaped our brains.

The things we think of as "hard" (chess, math) are recent evolutionary hacks.
The things we think of as "easy" (walking, seeing) are ancient and deeply optimized.

Explore the timeline of AI achievements. Notice a pattern: abstract reasoning tasks were solved decades before sensorimotor tasks. Chess fell in 1997. Walking is still being worked on.

Drag the slider through time. Watch how abstract tasks fall while physical tasks remain unsolved.

If human difficulty correlated with AI difficulty, all points would cluster along the diagonal line. Instead, they split into two distinct groups: sensorimotor tasks (easy for humans, hard for AI) and abstract tasks (hard for humans, easy for AI).

The paradox becomes clear: tasks in the upper-left (Moravec Zone) are trivially easy for humans but remain grand challenges for AI. Tasks in the lower-right (AI Sweet Spot) require years of human training but were among the first problems AI solved.

The explanation lies in evolution. Sensorimotor skills have been optimized for hundreds of millions of years. Vision evolved 540 million years ago. Locomotion even earlier. Every ancestor that was bad at these tasks was eaten.

Abstract reasoning? Chess has existed for 1,500 years. Mathematics as a formal discipline for about 4,000 years. Programming for 75 years. Evolution has barely begun optimizing these skills.

This means sensorimotor tasks aren't actually "easy" - they're heavily optimized. Your brain dedicates massive computational resources to tasks you don't consciously notice. Abstract reasoning is a thin veneer on top of ancient, sophisticated machinery.

The brain's resource allocation reveals the truth. Most of your neural real estate is dedicated to sensory processing and motor control - not abstract thought. The visual cortex alone uses a third of the cortex. The cerebellum, which coordinates movement, contains half of all neurons in the brain.

When you catch a ball, your brain performs real-time calculations of trajectory, adjusting for wind, spin, and your own body position - all in milliseconds, unconsciously. When you solve a math problem, you consciously step through logic using a tiny fraction of your brain.

Compare the computational resources required for different tasks. A chess move requires evaluating millions of positions. Walking requires coordinating hundreds of muscles, processing proprioceptive feedback, and making real-time balance adjustments - trillions of operations per step.

The difference is stark: tasks humans find effortless require vastly more computation than tasks humans find challenging. We just don't notice because evolution did the optimization work for us.

Moravec's paradox has proven remarkably predictive. We got language models before robot butlers. AI writes poetry while still struggling to fold towels. Understanding the paradox helps predict what AI breakthroughs will come next - and which will take decades longer than expected.

Use Moravec's paradox to predict how hard any task will be for AI. The key question: does it require manipulating the physical world, or can it be done with pure computation?

Moravec's paradox isn't just about AI - it reveals something profound about intelligence itself:

The hard problems are easy. The easy problems are hard.
Evolution knew all along.