This chapter focuses on the epistemological foundations of Keynes’s economic thought, as developed in his early work, A Treatise on Probability, where he worked out an antireductionist approach crucial for understanding his later economic theories. Motivated by a critique of utilitarianism’s reliance on calculable probabilities for moral decisions, Keynes developed a logical theory where probability is a relation between propositions applicable to non-demonstrative arguments. Though Keynes followed the anti-psychologism and the ‘linguistic turn’ of contemporary philosophers like Russell, he diverged from them by emphasizing the crucial role of complexity and the limits of formal probability calculus. Key concepts introduced include probabilistic causality, and the ‘weight of argument,’ reflecting the incompleteness of relevant information underlying a probability judgment. Keynes noted that more evidence does not always increase the weight of argument because it may reveal greater relevant ignorance. This framework, acknowledging the crucial role of organic units, the incommensurability of alternative theories, and the limits of probability calculations, underpins Keynes’s critique of classical probability theory for its Ignoratio Elenchi, that is its systematic use of assumptions inappropriate for answering the research questions raised by the complexity of the real world. In his subsequent economic works, Keynes extended similar criticisms to classical economic theory.

This chapter proposes a framework for analysing scientific pluralism in economics using taxonomies an evolutionary approach, emphasising analogies and differences with the approach of biological evolution. It rejects Kuhn’s concept of a monolithic ‘disciplinary matrix’ dominating ‘normal science,’ arguing instead for the persistent coexistence and interaction of multiple scientific visions, even during periods of hegemonic consensus. The chapter introduces ‘epistemic matrices’ as ideal-types representing distinct scientific visions, characterized by specific assumptions about individual agents and the economic system. Three matrices are outlined: 1) (neo)classical approach, where the Homo economicus acts with unbounded rationality, generating linear dynamics with a unique stable equilibrium, applicable only to ‘closed’ worlds; 2) algorithmic behaviourism, where the Homo ‘algorithmicus’ acts with procedural rationality and learns via error correction, generating multiple attractors, irreversible time, applicable in adaptive ‘open worlds’; 3) complexity approach, where the Homo faber acts with designing rationality and proactive expectations, generating complex dynamics, applicable to evolving ‘open worlds’. It argues that while realism increases from matrix 1 to 3, formal tractability decreases. The chapter ends by sketching the ‘evolutionary tree of economic theories’, tracing the evolution of aggregate analysis from the proto macroeconomics of mercantilism and monetarism, to the New Consensus, and beyond.