Michael Levin (2019), Frontiers in Psychology 10, 2688 (Hypothesis and Theory; published 13 December 2019) — DOI-10.3389/fpsyg.2019.02688, open access (CC BY)
Key takeaways
- A Self is defined by its goals. “The scale and types of goals that a system can pursue defines … the boundaries and content of the putative agent.” Goals are meant in the cybernetic sense: a feedback loop that drives a state toward a set point.
- The cognitive light cone. An agent’s boundary is “the most distant (in time and space) set of events that this system can measure and attempts to regulate”. It is drawn on a diagram with time on one axis and space on the other (Fig. 2), and is called “semi-quantitative”. No procedure for measuring it is given.
- Nesting is the norm. “Individuals can overlap – the same biophysical system can support a number of coexisting, coupled Selves with different cognitive borders” (Fig. 2C). “Most biological systems consist of multiple, nested selves [not one, as implied by Integrated Information Theory (Tononi, 2008)].”
- Binding is information sharing. Cells joined by gap junctions share one picture of the world and so form a larger Self with larger goals, such as an organ’s shape. Too little sharing fails to bind the parts; too much gives “a homogenous soup”.
- Cancer is the Self shrinking. A cell cut off from bioelectric coupling reverts to single-cell goals. This rests on cited experiments: depolarising a cell population makes melanocytes metastatic in genetically normal tadpoles, and hyperpolarisation stops human oncogenes from forming tumours.
- Consciousness is set aside. No claims about first-person experience are made “except for a few brief remarks at the end”. Those remarks suggest there may be “something-it-is-like to being a tissue or organ”, and say the model “takes a stand on the perennial combination problem and can perhaps be seen as a form of panpsychism”.
Standing in the record
Filed on 2026-10-02 for the owner’s bridge between Minsky’s society of mind and Schwitzgebel’s conscious United States. NOTE-341 is the close reading, and it placed the work: Active, on the composition side of the seam.
What it gives the bridge is a criterion of individuation under which a whole built from agents is itself an agent and its parts stay agents. Minsky’s Society of Mind (LIT-429, read here through Singh’s exposition LIT-444) decomposes a mind into agents that are mindless. Levin’s parts are the opposite. Cells are “competent”, goal-seeking Selves in their own right, and a body is a Self because their goals are scaled up by sharing information. Schwitzgebel’s LIT-159 (read in NOTE-131) argues that a whole made of sensing or conscious parts can be conscious (its C3). Levin supports the agency half of that claim, and his closing section gestures at the experience half without arguing it. He proposes, too, that “multi-human systems could have their own degree of cognition” and that reinforcement-learning and information-theoretic tools could test whether a social group is “a coherent goal-seeking individual”. That is the functional version of Schwitzgebel’s telescopic view of the United States, reached independently: Levin does not cite Schwitzgebel.
On the anti-nesting principle the paper is against IIT, but only by assertion. The bracket naming IIT is the whole of the argument. Nothing in the paper engages the exclusion postulate as stated in IIT 3.0 (LIT-445), which is about phenomenal consciousness, not goal pursuit. So it supports C3 in the agency sense, is neutral on C4 (Putnam is not mentioned), and does not support C5: it contests IIT’s conclusion without giving the reductio, or any argument, against the postulate.
Neighbours. It cites Hofstadter’s Gödel, Escher, Bach (LIT-437) twice, for collectives that compute “meta-system properties” and for agents that “transcend one system”. Its pragmatic rule, that the right level of description “is the one that facilitates prediction and control with the least effort”, is the intentional stance it cites (Dennett 1987), and so sits with Dennett’s LIT-220. It names causal-emergence measures (Hoel et al.) as the tool for finding the most efficacious level, which LIT-025 reviews. The record’s two other Levin works, LIT-195 and LIT-211, assume this paper’s continuum of cognition. LIT-120 (autopoiesis) and LIT-167 (the organism as an individual) are the record’s other accounts of biological individuality, and LIT-046 is its computational account of collective intelligence.
It carries no instruction for machine-learning practice. Its one remark about engineering is a prediction that “adaptive, useful robotics will require components that are themselves competent and goal-seeking”, which is a conjecture, not a method. It is not held in the anthology, and nothing here needs to move there.