Moravec's Paradox

Moravec's paradox is the observation that the hard problems for humans are often easy for machines, and the easy problems for humans are often hard for machines. Reasoning, algebra, and logic — which we find difficult and teach late — are tractable for AI and computers. Perception, movement, and common-sense intuition — which we do effortlessly from infancy — have been the bottleneck for robotics and AI. The paradox is attributed to Hans Moravec and others (e.g. Minsky) in the 1980s.

The usual explanation is evolution. We've had millions of years to optimise perception and motion; they're baked into ancient, efficient neural circuitry. Abstract reasoning is a recent add-on, less optimised and more accessible to formalisation. So we find reasoning "hard" (it's effortful and slow) and perception "easy" (it's automatic). For machines, the opposite holds: formal reasoning is just symbol manipulation; perception and sensorimotor control require processing huge amounts of noisy, ambiguous data and coordinating many degrees of freedom. The paradox predicts where automation will land first: not "easy" human tasks (driving, folding laundry), but "hard" human tasks (calculation, rule-based analysis, pattern matching in structured data).

In strategy and hiring, the implication is to automate and outsource what machines do well (logic, search, optimisation) and to value and retain what they still do poorly (physical dexterity, social nuance, open-ended judgment in novel situations). It also warns against assuming that human difficulty maps to machine difficulty — the map is inverted.

The paradox is a moving frontier. Image recognition and speech-to-text have shifted from machine-hard to machine-easy in the last decade. What remains stubbornly machine-hard tends to be tasks that mix perception, context, and one-off judgment — e.g. handling a novel customer complaint, navigating an unstructured environment, or making a decision with incomplete and ambiguous data. Updating your view of where the frontier is matters for automation roadmaps and hiring.

Look for domains where humans perform a task effortlessly and machines struggle, or the reverse. When someone says "that's trivial for humans" or "that's just common sense," consider whether it's actually the kind of thing that has been hardest to automate. When a machine excels at something humans find hard, that's the other side of the paradox.

Common misapplication: Treating the paradox as static. As AI and robotics improve, the boundary shifts. What was "hard for machines" (e.g. image recognition, translation) has moved. The principle — invert human difficulty when thinking about machine difficulty — remains; the specific frontier does not.

Second misapplication: Using it to dismiss automation. The paradox doesn't say "don't automate." It says automate where machines have comparative advantage (logic, scale, consistency) and don't assume that human-easy tasks will automate first. Use it for prioritisation, not for complacency.

Moravec's paradox connects to automation, comparative advantage, and where humans and machines should each focus. The models below either reinforce it (automation, division of labour), create tension (abstraction, technical debt), or extend to strategy (bottlenecks, comparative advantage).

The quote states the paradox directly. The implication for builders and strategists: don't use human difficulty as a proxy for automation difficulty. The skills we take for granted — seeing, moving, reading the room — are the ones that have been slowest to replicate. Plan and invest accordingly.

Summary: Moravec's paradox is that what is hard for humans (reasoning, logic) is often easy for machines, and what is easy for humans (perception, motion, common sense) is often hard for machines. The usual explanation is evolution: we're optimised for the latter. Use it to prioritise automation (automate the formal and rule-based first) and to value human skills that remain machine-hard (judgment, dexterity, social nuance). Don't assume human difficulty maps to machine difficulty — the map is inverted. The frontier shifts as AI and robotics improve; the principle remains.

Technical debt is the cost of shortcuts. In AI systems, we often patch "human-easy" gaps with rules and heuristics — accruing debt because we can't yet automate the full task. The tension: the paradox explains why some debt persists — the missing capability is the machine-hard part. Pay down debt when the frontier moves and that part becomes tractable.