Biology, Buddhism, and AI: Care as the Driver of Intelligence

What to take away

1. 1. The paper introduces the Care Light Cone (CLC) as a formal construct distinct from the Physical Light Cone (PLC): the CLC maps the spatiotemporal distribution of an agent's goal-states, which can extend beyond physically reachable states, while the PLC bounds only achievable trajectories.
2. 2. Three stereotypical CLC archetypes — the futurist (care concentrated on complex future scenarios), the historian (care concentrated on the past), and the chauvinist (a one-sided deep-past narrative dominating future resolution) — demonstrate that care distributions can take asymmetric, disconnected point-cloud shapes rather than symmetric cones.
3. 3. Xenobots, protoorganisms made from frog skin cells reported in PNAS 2020 and 2021, exhibit coherent anatomical and behavioral outcomes despite having no evolutionary backstory of selection forces, illustrating why Care (what the system expends energy to achieve) is a more informative metric than phylogenetic origin.
4. 4. Gap junctions (electrical synapses present in bacterial biofilms approximately 3.5 billion years ago) are identified as the biophysical mechanism by which cells merge care light cones: once calcium signals pass through gap junctions, individual cells cannot distinguish endogenous from neighbor-derived signals, forcing a functional expansion of the collective self's goal boundary.
5. 5. Cancer is reframed as a failure of CLC integration: inappropriate reduction of gap junctional connectivity causes cells to revert to ancient unicellular goal-spaces, treating the rest of the body as external environment and producing metastatic over-proliferation.
6. 6. The paper raises the open hypothesis that Care, rather than the pursuit of knowledge alone, is necessary to generate the motivational dynamics leading to artificial general intelligence, because knowledge-seeking without modification drive risks self-defeating passivity.
7. 7. To replicate the light-cone comparison methodology, a researcher would plot diverse agents — from bacteria (local sugar-concentration goals, sub-millimeter/sub-second scale) to humans (planetary-scale, multi-generational goals) to a hypothetical Bodhisattva agent (infinite CLC sphere) — on the same two-axis space-time diagram using estimated maximal goal scales collapsed onto one spatial and one temporal axis.
8. 8. The Bodhisattva Vow is analyzed as a discrete phase transition in the CLC: before the vow a human agent's CLC is finite by human standards (potentially spanning distant galaxies or global ecosystems), but after formally committing to care for all sentient beings throughout infinite time and space, the CLC becomes formally unbounded, constituting what the paper calls a 'great evolutionary transition' in individuality.
9. 9. In reinforcement learning systems, the loss function in supervised learning and the negative reward signal in RL are identified as direct functional analogs of dukkha — the delta between current and optimal states — confirming that stress-reduction loops are substrate-independent drivers of goal-directed behavior across biological and artificial systems.
10. 10. The paper funded under John Templeton Foundation Grant 62212 (Levin) and Templeton World Charity Foundation Grant 0470 (Duane, Doctor, Witkowski, Solomonova) argues that existing ethical frameworks relying on biological origin or material composition will become untenable within decades as hybrots, chimeras, organoids, assembloids, and neuroprosthetic cyborgs proliferate, and that the measurable radius of an agent's compassion capacity (outward-facing CLC area) is the only rubric likely to remain applicable.

Peer brief — for seminar discussion

Doctor, Witkowski, Solomonova, Duane, and Levin (Entropy 2022, 24, 710) argue that Care — operationalized as concern for the alleviation of dukkha, the stress-delta between current and optimal states — is a substrate-independent invariant that simultaneously defines intelligence, determines cognitive sophistication, and grounds moral obligation across biological, artificial, and chimeric agents. The paper is a concept paper, not an empirical study, and its primary methodological contribution is the Cognitive Light Cone (CLC) formalism: a space-time diagram in which any agent's maximum goal-space is plotted on a single spatial and a single temporal axis, yielding a cone whose area represents the agent's domain of Care. This is formally distinguished from the Physical Light Cone (PLC), which bounds only achievable state trajectories; crucially, a CLC can exceed its PLC because an agent may care about states it cannot reach. Agents from bacteria (local sugar concentrations, sub-millimeter scale) to humans (planetary-scale, multi-generational goals) to a hypothetical Bodhisattva (formally infinite CLC) are plotted on the same graph, making diverse intelligences directly comparable without reference to substrate, anatomy, or phylogeny. An alternative methodological approach the paper could have employed is active inference or free-energy minimization (à la Friston), which offers a similarly substrate-independent formalism for goal-directed behavior; the authors instead opt for the Buddhist dukkha-as-stress translation and the teleonomy literature stemming from Rosenblueth, Wiener, and Bigelow (1943). The load-bearing finding is the proposed equivalence: intelligence scales as Care scales, because the size of an agent's CLC — what it can spend energy trying to change — directly determines the complexity and sophistication of problems it can address. This is illustrated with gap junctions in bacterial biofilms (present approximately 3.5 billion years ago per Prindle et al., Nature 2015), cancer as a CLC-contraction pathology, Xenobots (PNAS 2020, 2021) as care-characterized rather than phylogenetically characterized agents, and the Bodhisattva Vow as a documented human technology for CLC expansion. The paper predicts that a radical expansion of an AI system's objective function — structured around Care for alleviating others' dukkha rather than narrow task performance — is a necessary condition for progress toward artificial general intelligence, and that the Bodhisattva Vow constitutes a design principle for triggering a major evolutionary transition (in the Szathmáry–Maynard Smith sense) in any cognitive system. The most pressable critical objection is circularity and operationalization: Care is defined as concern for stress relief, intelligence is defined as the capacity to identify and alleviate stress, and the CLC is the spatial representation of what an agent cares about — the three concepts mutually define one another without an independent empirical handle. A critical reader would demand a non-circular measurement procedure for CLC extent that does not already presuppose knowledge of what the agent cares about, and would note that the paper provides no falsifiable predictions distinguishing its framework from, say, a simple goal-complexity metric or integrated information theory. The ethical claims — that Care-radius should replace phylogenetic or material criteria for moral status — are philosophically substantive but rest entirely on the theoretical framework, and the paper's scope, explicitly covering hybrots, organoids, assembloids, and neuroprosthetic cyborgs, outpaces any current empirical test of the CLC formalism.

Claims (3)

• Care as the Driver of Intelligence

  Central thesis: expansion of cognitive boundaries through care for others' stress states is fundamental driver of intelligence scaling across evolution and lifespans

• Moral Duty Based on Care Capacity

  Our moral duty toward any being should be based on degree of care they can exhibit toward others, not composition or origin

• Cancer as Cognitive Defection

  Cancer results from defection from large-scale anatomical goals due to reduction of gap junctional connectivity, causing cells to revert to unicellular concerns

Questions (1)

• What is central about intelligence regardless of material composition or origin?

  Key research question framing the paper's investigation into substrate-independent invariants of intelligence

Similar preprints — Semantic Scholar