commonsense reasoning
In plain wordsThe everyday background knowledge and reasoning people take for granted — dropped glasses break, hungry people eat, nobody is in two places at once — which machines long struggled to match.
The ability to make the everyday inferences about the physical and social world that people make effortlessly, and the AI research area that tries to give machines this ability. It covers naive physics (objects fall, liquids spill), intuitive psychology (people act on beliefs and goals), typical events and object properties, and reasoning about time, space and causation, usually through defaults with exceptions. Symbolic approaches hand-encoded such knowledge in logic (CycCyc, prop. n.: An AI system built by specialists who spent decades typing in millions of everyday truths, such as 'people cannot be in two places at once', hoping to give computers common sense., the situation calculussituation calculus, n.: A logical language for describing how the world changes as actions happen, so that a program can work out what will be true after any sequence of actions.) or collected it as plain assertions (ConceptNet, ATOMIC), confronting the frame problemframe problem, n.: The difficulty of getting an AI to work out what does and does not change when something happens, without spelling out every one of the countless things that stay the same., non-monotonicitynon-monotonic reasoning, n.: Reasoning in which new information can overturn earlier conclusions — you assume a bird can fly until you learn it is a penguin — as people do constantly but classical logic does not allow. and sheer scale. Large language modelslarge language model, n.: A computer program trained on enormous amounts of text to predict the words that come next. At vast scale, that simple skill lets it chat, answer questions, translate, summarize and write code. absorbed much commonsense knowledge from text and now score highly on benchmarks such as the Winograd schema challengeWinograd schema challenge, prop. n.: A test of machine common sense built from pairs of sentences in which changing one word changes what a pronoun refers to — something people resolve instantly using everyday knowledge., HellaSwag and PIQA, but how robust that reasoning is — in unfamiliar situations, physical interaction and multi-step problems — remains debated.
The robot failed a basic commonsense test: it tried to store the soup in the cutlery drawer.
Origin
John McCarthy's 'Programs with Common Sense' (presented 1958, published 1959) launched the program of formalizing everyday knowledge. Patrick Hayes's 'Naive Physics Manifesto' (1978), Douglas Lenat's Cyc project (from 1984) and Ernest Davis's Representations of Commonsense Knowledge (1990) shaped the field; benchmarks and text-trained models later shifted its focus.
Usage note
Whether language models possess common sense or only its statistical traces is contested: models that ace benchmarks can fail simple rephrasings, and some researchers argue that embodied experience is needed.
Related
- CycCyc, prop. n.: An AI system built by specialists who spent decades typing in millions of everyday truths, such as 'people cannot be in two places at once', hoping to give computers common sense.
- frame problemframe problem, n.: The difficulty of getting an AI to work out what does and does not change when something happens, without spelling out every one of the countless things that stay the same.
- Winograd schema challengeWinograd schema challenge, prop. n.: A test of machine common sense built from pairs of sentences in which changing one word changes what a pronoun refers to — something people resolve instantly using everyday knowledge.
- non-monotonic reasoningnon-monotonic reasoning, n.: Reasoning in which new information can overturn earlier conclusions — you assume a bird can fly until you learn it is a penguin — as people do constantly but classical logic does not allow.
- knowledge representationknowledge representation, n.: The study of how to record knowledge about objects, categories, rules and events in a precise format, so that a program can reason with it and reach conclusions nobody typed in.
- world modelworld model, n.: An AI's internal simulation of how its world works, which lets it predict what will happen next — especially in response to its own actions — and so plan, imagine or practice without acting for real.
- Moravec's paradoxMoravec's paradox, n.: The observation that what is hard for people, like chess or algebra, is often easy for computers, while what toddlers do effortlessly, such as seeing, walking and picking things up, is very hard for machines.
Sources
- McCarthy, J. (1959). Programs with Common Sense. In Mechanisation of Thought Processes. HMSO, London.
- Davis, E. & Marcus, G. (2015). Commonsense Reasoning and Commonsense Knowledge in Artificial Intelligence. Communications of the ACM 58(9), 92–103.
Mentioned in
abductive reasoningabductive reasoning, n.: Reasoning from an observation to its most likely explanation — the grass is wet, so it probably rained — the kind of educated guess that doctors and detectives make., AI-completeAI-complete, adj.: Describes a problem believed to be so hard that solving it would require general, human-level intelligence, such as flawless translation or truly understanding a story., CycCyc, prop. n.: An AI system built by specialists who spent decades typing in millions of everyday truths, such as 'people cannot be in two places at once', hoping to give computers common sense., frame problemframe problem, n.: The difficulty of getting an AI to work out what does and does not change when something happens, without spelling out every one of the countless things that stay the same., John McCarthyJohn McCarthy, prop. n.: The American computer scientist who coined the term 'artificial intelligence', invented the Lisp programming language and championed formal logic as the route to machine common sense., knowledge representationknowledge representation, n.: The study of how to record knowledge about objects, categories, rules and events in a precise format, so that a program can reason with it and reach conclusions nobody typed in., reasoningreasoning, n.: Working out conclusions step by step from facts, rules or evidence, as in solving a math problem or planning a trip. Whether today's AI truly reasons, or imitates reasoning, is debated., ConceptNetConceptNet, prop. n.: A free, crowd-built network of everyday facts — 'a knife is used for cutting', 'rain makes things wet' — in many languages, which AI systems use to make sense of ordinary situations., conceptual dependencyconceptual dependency, n.: A 1970s theory for representing the meaning of sentences with a small set of basic actions, so that 'John gave Mary a book' and 'Mary got a book from John' come out the same., coreference resolutioncoreference resolution, n.: Working out which words in a text are talking about the same someone or something — for instance, that 'she', 'the senator' and 'Ms. Okafor' all mean one woman., Douglas LenatDouglas Lenat, prop. n.: An American computer scientist who spent four decades building Cyc, a giant hand-written database of everyday common sense meant to let computers reason like people., HellaSwagHellaSwag, prop. n.: A test in which an AI must pick the most sensible ending for a description of an everyday scene; the wrong endings were machine-written to fool computers, so people found it easy and early AI did not., natural language inferencenatural language inference, n.: Deciding whether one sentence follows from another, contradicts it, or neither: if 'a man is playing a guitar on stage', is it true that 'a man is performing music'?, non-monotonic reasoningnon-monotonic reasoning, n.: Reasoning in which new information can overturn earlier conclusions — you assume a bird can fly until you learn it is a penguin — as people do constantly but classical logic does not allow., scriptscript, n.: In classic AI, a stored outline of a familiar sequence of events, such as eating at a restaurant, that helps a program understand a story by filling in what normally happens., SuperGLUESuperGLUE, prop. n.: A harder follow-up to the GLUE language-understanding test, released in 2019 after AI systems had already beaten the human baseline on the original., Terry WinogradTerry Winograd, prop. n.: An American computer scientist who built SHRDLU, a celebrated early program that conversed in English about a world of blocks, and later became a critic of symbolic AI., Winograd schema challengeWinograd schema challenge, prop. n.: A test of machine common sense built from pairs of sentences in which changing one word changes what a pronoun refers to — something people resolve instantly using everyday knowledge.