Computational common sense
The “common sense problem” within AI is longstanding; notable early failures include computing systems ‘suggesting boiling a kidney to cure an infection, and attributing reddish spots on a Chevrolet to a case of the measles’ (Cantwell Smith, 2019: 37).
Advanced AI systems are still often seen to lack common sense, understood as ‘the ability to reason intuitively about everyday situations and events, which requires background knowledge about how the physical and social world works’ (Choi, 2022: 139).
Logic-based AI’s quest to “codify common sense” through translating tacit human knowledge into machine-readable knowledge is perhaps best encapsulated by the Cyc project, launched by computer scientist Doug Lenat and colleagues in 1984.
CyC was designed to navigate new situations by drawing analogical links to what it already “knows” – which required that it be programmed with a substantial base of knowledge and an organised body of reasoning methods (Lenat et al, 1985). Through mobilising higher-order logic, Cyc began to abstract, generalise, and learn from its experience.
Over time, however, systems like Cyc failed to achieve robust flexibility and intuitive sense-making – deficits linked not only to difficulties in scaling up from isolated ‘micro-worlds’ (Davis, 1990) but also to logic-based AI’s account of ‘intelligence as a passive receiving of context-free facts into a structure of already stored data’ (Dreyfus, 1992: 34).
Later AI initiatives, such as MIT’s Open Mind Common Sense project (1999-2016), drew on machine learning and internet-sourced data to simulate human common sense knowledge.
Key actors now recommend a hybrid “solution” to AI’s persistent common sense problem that integrates the speculative pattern recognition of generative AI and language-based (as opposed to logic- or rules-based) formalisms (Choi, 2022).
Yet critics argue that such “neuro-symbolic” approaches continue to neglect how human common sense is emerges out of situated socio-affective contexts and relations (Suchman, 2024). Also at stake is the capacity of mainstream AI research to confront common sense’s pervasive political, ideological and ethical elements (Pedwell, 2024).