The Moravec Paradox
The paradox nobody expects
In 1988, roboticist Hans Moravec wrote something that still sounds backwards today: "It is comparatively easy to make computers exhibit adult-level performance on intelligence tests, and difficult or impossible to give them the skills of a one-year-old when it comes to perception and mobility."
Read that again. A computer beat the world chess champion in 1997. It still can't reliably fold a towel the way a toddler can stack blocks. The things we associate with 'being smart', math, logic, language, strategy, turned out to be the easy part for a machine.
The things a one-year-old does without thinking, recognizing a cup in clutter, grabbing it without crushing it, balancing while reaching for it, turned out to be the hard part. This is the Moravec paradox, and it flips most people's intuition about which jobs are 'safe' from AI.
Guess which one is harder for AI
For each pair below, pick the task you think is HARDER for an AI system to do well. Trust your gut: then see if it matches reality.
Which is harder for AI?
Which is harder for AI?
Which is harder for AI?
Which is harder for AI?
Which is harder for AI?
Why evolution explains it
Here's the mechanism behind the paradox. Abstract reasoning: chess, math, logic, language structure, is recent. Humans have been doing it for maybe 100,000 years. It's a thin, late layer on top of the brain, and because it's recent, it's also relatively simple to describe with explicit rules and data.
Perception and movement are ancient. Walking, grasping, balancing, reading a face, these were refined over roughly 3.5 billion years of evolution, under brutal selection pressure from predators, terrain, and survival. The result is unbelievably optimized, and almost entirely unconscious. You don't know HOW you catch a ball, you just do it.
AI learns from explicit patterns in data: rules, examples, checkable outcomes. Reasoning tasks are easy to turn into that format, a chess move is right or wrong, a legal citation exists or doesn't. Sensorimotor skill isn't stored as rules anywhere, not even in your own head, which makes it brutally hard to extract and hand to a machine.
Quick check
Why is abstract reasoning easier for AI than physical perception and movement?
This flips the job map
Most predictions about automation got this backwards for decades, assuming factory and manual labor would go first, while 'knowledge work' stayed safely human. The Moravec paradox says the opposite is structurally more likely.
Tasks built on reasoning over clean, structured information, drafting contracts, writing code, analyzing spreadsheets, summarizing reports, sit exactly where AI is strongest. Tasks requiring fine motor skill in unpredictable physical environments, plumbing, elder care, electrical repair, cooking in a real kitchen, sit exactly where AI is weakest, and will for a while.
This connects directly to what you learned in the last lesson: it's not about white-collar versus blue-collar. It's about whether a task is clean-pattern-with-no-accountability (AI's zone) or messy-physical-and-judgment-heavy (still yours). Moravec's paradox is why that second zone includes far more 'unskilled-looking' work than anyone expected.
Does this mean you should quit your digital career and become a plumber? No. It means that if your value in an office is just 'writing clean code' or 'summarizing spreadsheets,' you are in the danger zone. Your job won't be saved by your degree, it will be saved by your ability to shift from a pure 'reasoner' to an architect who uses AI to handle the reasoning, while you inject the judgment, the strategy, and the accountability. The knowledge workers who understand Moravec's paradox are the ones who will design the tools for everyone else.
What you learned
- ✓The Moravec paradox: tasks that feel intellectually hard (chess, exams, composing music) are often easy for AI; tasks that feel effortless (walking, grasping, reading a noisy room) are often hard for AI.
- ✓This is backwards from most people's intuition, and from most outdated predictions about which jobs automation would hit first.
- ✓The reason is evolutionary: reasoning is a recent, rule-like skill; perception and movement are billions of years deep and largely unconscious, which makes them hard to formalize into data AI can learn from.
- ✓Skilled physical work in unpredictable environments is more AI-resistant than it looks, and clean, structured "knowledge work" is more exposed than it looks.