Why agency and cognition are fundamentally not computational

Summary

“To live is to know.” (Maturana, 1988) “Between the stimulus and the response, there is a space. And in that space lies our freedom and power to choose our responses.” (Frankl, 1946, 2020) “Voluntary actions thus demonstrate a ‘freedom from immediacy.’ ” (Haggard, 2008; channeling Shadlen and Gold, 2004) 1 Introduction All organisms are limited beings that live in a world overflowing with potential meaning (Varela et al., 1991; Weber and Varela, 2002; Thompson, 2007), a world profoundly exceeding their grasp (Stanford, 2010). Environmental cues likely to be important in a given situation tend to be scarce, ambiguous, and fragmentary. Clear and obvious signals are rare (Felin and Felin, 2019). Few problems we encounter in such a “large world” are well-defined (Savage, 1954). On top of this, organisms constantly encounter situations they have never come across before. To make sense of such an ill-defined and open-ended world—in order to survive, thrive, and evolve—the organism must first realize what is relevant in its environment. It needs to solve the problem of relevance. In contrast, algorithms—broadly defined as automated computational procedures, i.e., finite sets of symbols encoding operations that can be executed on a universal Turing machine—exist in a “small world” (Savage, 1954). They do so by definition, since they are embedded and implemented within a predefined formalized ontology (intuitively: their “digital environment” or “computational architecture”), where all problems are well-defined. They can only mimic (emulate, or simulate) partial aspects of a large world: algorithms cannot identify or solve problems that are not precoded (explicitly or implicitly) by the rules that characterize their small world (Cantwell Smith, 2019). In such a world, everything and nothing is relevant at the same time. This is why the way organisms come to know their world fundamentally differs from algorithmic problem solving or optimization (Roli et al., 2022; see also We...