Abduction
Abduction, as theorised by the mathematician and philosopher Charles Sanders Peirce, is reasoning that infers the best explanation from incomplete data.
Linked throughout Peirce’s writing to ‘hypothetical thinking, imagination, intuition and guessing’ (Fortes, 2022: 1), abduction consolidates for him in the 1890s as a kind of inference involving the ‘generation and evaluation of explanatory hypotheses’ amid uncertainty (Thagard, 2007: 226).
While twentieth century AI built logic-based systems capable of deductive and inductive reasoning, twenty-first century deep learning is said to specialise in abductive reasoning, going beyond what has been taught to discover otherwise unknowable patterns and correlations.
Differently from deductive reasoning (starting with a general rule and applying it to concrete cases) or inductive reasoning (inferring a general rule from facts about individual cases), abductive reasoning in machine learning (asking what would need to be the case for a theory to be true) operates as a speculative experiment in which ‘what one will ask of the data is a product of patterns and clusters derived from the data’ (Amoore, 2020: 47).
When referring to deep learning operations as “abductive”, however, it is important to recognise that, for Peirce, abduction is a multi-modal mode of discovery in which what is most important is not necessarily the data itself but rather subjects’ affective reactions: abduction is initiated by ‘the feeling of puzzlement and ends at the satisfaction of knowing’ (Fortes, 2022: 5).