Moravec's paradox
AI struggles with tasks easy for humans but excels at complex computations.
Moravec's paradox describes a strange situation in artificial intelligence. Tasks that are very easy for humans, like recognizing a face or walking across a room, are incredibly difficult for AI. Conversely, tasks that are hard for humans, such as complex mathematical calculations or playing chess at a grandmaster level, are relatively easy for AI.
Think about a baby learning to walk. It takes years of practice and a sophisticated understanding of balance and movement. Yet, a computer can perform billions of calculations per second. This paradox highlights the difference between human intelligence, which is deeply rooted in sensory experience and physical interaction, and AI's current strengths, which lie in pattern recognition and logical processing.
Moravec's paradox posits that high-level reasoning, which is difficult for humans, is easy for AI, while low-level sensory-motor skills, which are easy for humans, are hard for AI. This is because human intelligence evolved over millions of years to deal with the physical world, embedding these skills deeply. AI, on the other hand, has primarily developed through symbolic manipulation and statistical pattern matching, making abstract reasoning more tractable than embodied interaction.
The difficulty for AI in replicating human-like sensory-motor skills stems from the vast, unstructured, and noisy nature of real-world data, requiring robust perception and control systems. Current large language models, while impressive in language tasks, still face significant challenges in grounding their knowledge in physical reality and performing complex physical manipulations.
Moravec's paradox highlights the ongoing challenge in AI development: tasks that seem simple for humans, like understanding context or performing physical actions, remain difficult for AI. Recent advancements in large language models, like those from OpenAI, are making progress on complex reasoning, but the paradox reminds us that true general intelligence requires overcoming these 'easy' human tasks.
- How Fyxer built an AI executive assistant people trustOpenAI · OpenAI News ·
- OpenAI ChatGPT
- OpenAI's work on models that can draft personalized replies and build trusted AI assistants touches on Moravec's paradox by tackling complex human communication, while the underlying challenges of real-world interaction and nuanced understanding persist.
- AI alignment
- AI alignment focuses on ensuring AI systems act in accordance with
- AI safety
- AI safety is concerned with preventing catastrophic risks from advanc
- transfer learning
- Transfer learning is a technique where a model trained on one
The paradox was first articulated by Hans Moravec in his 1988 book 'Mind Children'. He observed that robotics researchers found it much harder to teach robots basic skills like walking and seeing than abstract reasoning tasks. Source ↗
Ask an AI chatbot to describe the feeling of stubbing your toe or the taste of a strawberry. Then, ask it to calculate the trajectory of a projectile given specific initial conditions. The AI will likely excel at the calculation but struggle to convey the subjective human experience.
The Playbook: reusable how-tos, prompts, configs and tips pulled out of the news.
- PromptRole-play a student struggling with math for AI tutor simulationSimulate a student's math struggles and uncertainty for AI tutor evaluation. · 5 min
- TemplateReusable workflow templates for healthcare teamsTeams can create and share templates for common workflows like documentation or prior authorizations. · 15 min
- TemplateTemplate for building an AI tool tailored to your workA structured approach to defining and building a custom AI tool for specific business needs. · 15 min
- TemplateEvals CSV prompt list for agent skillsA CSV file template for listing prompts used in evaluating agent skills. · 20 min
- HeuristicDouble coding agent task completion with docs linksProviding coding agents with a documentation link significantly increases their task completion rate. · 2 min
Questions people ask
- Why is walking hard for robots but easy for babies?
- Walking involves complex, real-time adjustments to balance, terrain, and body position, requiring sophisticated sensory feedback and motor control that AI is still developing. Babies learn through years of physical interaction and adaptation.
- Can AI ever overcome Moravec's paradox?
- Researchers are actively working on embodied AI and integrating AI with robotics to improve sensory-motor skills. Progress is being made, but replicating the full spectrum of human physical intelligence remains a significant challenge.
First seen on Subvolts Sep 14, 2026. Updated Sep 21, 2026. Written by an AI model from the items above and reviewed against them weekly; report an error.