Weak AI

Weak AI is artificial intelligence built to perform a specific task or a bounded set of tasks, such as recognizing faces, transcribing speech, or recommending products, with no general understanding or awareness of what it does. In everyday usage, weak AI is a synonym for artificial narrow intelligence (ANI), and by most definitions every AI system in use today belongs to it. "Weak" refers to scope: inside its task, a weak AI system can outperform every human expert.

Origin of the Term#

Philosopher John Searle introduced the weak and strong labels in his 1980 paper "Minds, Brains, and Programs", the paper that presented the Chinese Room argument. In Searle's usage, weak AI is the claim that computers are powerful tools for simulating and studying the mind, and strong AI is the claim that an appropriately programmed computer literally has a mind and understands. His thought experiment imagines a person in a closed room who answers questions written in Chinese by following an English rulebook for manipulating Chinese characters. The answers convince native speakers outside, yet the person understands no Chinese, and Searle argued that a computer running a program is in the same position.

Over the following decades the labels shifted from philosophical positions to categories of capability. Today "weak AI" usually means narrow, task-specific AI, and "strong AI" is used as a synonym for artificial general intelligence (AGI).

Characteristics of Weak AI#

• Task-bound competence: A weak AI system does what it was built or trained for. A model trained to read chest X-rays cannot read mammograms, let alone play chess, without new training data and retraining.
• Sensitivity to data drift: Modern weak AI learns statistical patterns from data, so its accuracy holds only for inputs that resemble its training data. When real-world inputs shift, it can fail while still reporting high confidence.
• No understanding or goals of its own: The system maps inputs to outputs. It has no awareness of what a pedestrian or a tumor is and no aims beyond the objective it was optimized for.
• Superhuman within scope: Narrowness is compatible with exceptional performance. DeepMind's AlphaFold 2 predicts protein structures with accuracy close to laboratory methods, work that earned Demis Hassabis and John Jumper a share of the 2024 Nobel Prize in Chemistry.

Weak AI vs. Strong AI#

• Weak AI (ANI): Exists today and covers all deployed AI. "ANI" stresses the narrow scope of capability, while "weak AI" also carries Searle's original question of whether the machine understands anything.
• Strong AI (AGI): Hypothetical. In common usage, a system that matches humans across virtually all cognitive tasks and learns new ones without task-specific engineering; in Searle's sense, a machine with a genuine mind.
• Artificial superintelligence (ASI): Hypothetical. A system that surpasses the best humans in virtually every domain, the scenario behind the technological singularity.

Examples of Weak AI#

• Computer vision: Driver-assistance systems detect pedestrians and lane markings, and factory cameras inspect parts for defects. Ultralytics YOLO26 is narrow in exactly this way: one model family handles detection, instance and semantic segmentation, depth estimation, classification, pose estimation, and oriented boxes, and each trained model handles only the task and classes it was trained on.
• Speech recognition: Dictation software, captioning tools, and voice assistants convert spoken audio into text.
• Recommendation systems: Streaming and shopping services rank items for each user from viewing and purchase history.
• Games: IBM's Deep Blue beat world chess champion Garry Kasparov in 1997 and could do nothing else.

Is Generative AI Still Weak AI?#

Large language models (LLMs) write code, translate, summarize, and answer questions about images, which strains the narrow label. A 2023 Microsoft Research paper on GPT-4 described "sparks of artificial general intelligence", and those who keep LLMs in the narrow category point to uneven abilities, a pattern called jagged intelligence, and to weights that stop learning when training ends. The EU AI Act sidesteps the weak and strong labels and regulates such systems as general-purpose AI models, with obligations for their providers applying since August 2, 2025.