Favorites
8 entities · 0 take-aways
Papers & entities (8)
Scale-Free Cognition, the framework introduced here, proposes that any coherent Self is demarcated by a 'cognitive light cone'—a spatio-temporal boundary of events a system can measure, model, and attempt to regulate—and that this boundary expands through evolutionarily conserved bioelectric mechanisms rather than requiring nervous systems. The core claim is that developmental bioelectricity, implemented primarily through gap junctions and voltage-gated ion channels, provides the proximate mechanism by which single-cell homeostatic loops scale into multicellular cognitive agents: when cells couple via gap junctions they share a unified Umwelt, transforming individually local set points into organ-level morphogenetic goals. Empirical support is drawn from three substrate types: in Xenopus tadpoles, craniofacial organs in abnormal positions still converge on a 'correct frog face configuration' (Vandenberg et al., 2012), demonstrating invariant anatomical goal-pursuit; in genetically normal tadpoles, depolarization of a specific melanocyte population is sufficient to induce metastatic transformation (Lobikin et al., 2012); and conversely, human oncogene-driven tumorigenesis can be blocked by optogenetic or constitutive hyperpolarization (Chernet and Levin, 2013b, 2014). Cancer is reframed not as genomic chaos but as a reversible shrinkage of the computational boundary—gap-junction uncoupling collapses a cell's cognitive horizon from whole-body to single-cell scale, recapitulating unicellular behavioral modes including maximal proliferation and migration. A 24-hour progesterone stimulus via wearable bioreactor was sufficient to initiate 11 months of autonomous limb-regeneration activity in adult Xenopus (Herrera-Rincon et al., 2018), illustrating how brief intervention at the correct level of organization can trigger a pre-encoded morphogenetic module. Levin argues this implies that biomedicine, AI design, and exobiology should prioritize identifying and communicating with agents at the level of their actual goal-directed organization rather than exclusively targeting molecular mechanisms.
Minimizing expected variational free energy under a discrete-state Markov decision process generative model is sufficient to produce curiosity, epistemic learning, and insight without any additional machinery. Friston et al. 2017 demonstrates this across two linked mechanisms: first, including posterior beliefs about likelihood parameters **A** in expected free energy G(π) introduces a novelty term—information gain about model parameters—that drives agents to sample combinations of hidden states and outcomes they have not yet encountered, resolving ignorance rather than merely ambiguity or risk. Second, Bayesian model reduction (implemented via the spm_MDP_VB_X.m routine in SPM) allows post-hoc or online pruning of redundant concentration parameters: a reduced model is accepted when ΔF ≤ −3, corresponding to a Bayes factor of approximately 20:1 in favor of the simpler model. Simulated agents learning a 3-rule, 4-factor abstract contingency task (144 hidden-state combinations, 36 possible outcomes) reach near-perfect performance after roughly 14 trials under pure epistemic learning, dropping to approximately 10 trials when online Bayesian model reduction is applied across 64 simulated agents. The sleep analog—non-REM synaptic pruning followed by REM-like belief re-evaluation—is formalized identically through the same free energy difference equation. The paper argues this implies that aha moments are necessarily subpersonal events (optimization of the generative model itself, not modeling of that optimization), that the quality of intelligence is inversely related to the thermodynamic energy expended during convergence via the Jarzynski equality, and that communicating reduced model priors rather than parameter posteriors constitutes a principled formal account of shared knowledge—consciousness in the pre-Cartesian sense of con-scire.
TAME—Technological Approach to Mind Everywhere—formalizes a non-binary, empirically grounded framework for recognizing, comparing, and manipulating cognition across radically diverse substrates, from single cells and gene regulatory networks to chimeric bioengineered organisms and hybrots. Central to the framework is an axis of persuadability ranging from brute-force hardware rewiring (e.g., mechanical clocks) through homeostatic circuits and trainable animals to rational-argument-responsive humans, which serves as a semi-quantitative tool for determining optimal intervention strategy for any given system. Empirical anchors include: planarian flatworms regenerating barium-insensitive heads by efficiently traversing transcriptional space to regulate a small subset of genes; gap-junctional blockade producing planaria with heads morphologically matching other extant species despite wild-type genetics; and tadpoles with ectopically tail-placed eyes successfully performing visual learning tasks via spinal cord re-routing. The framework introduces a 'cognitive light cone' diagram plotting spatio-temporal scale of goal-directed activity to place microbes, rats, and humans on a common axis without appealing to substrate or evolutionary origin. Developmental bioelectricity—implemented through pre-neural ion channels and gap junctions scaling cell-level feedback into anatomical homeostasis—is identified as evolution's primary medium for enlarging cognitive boundaries, and the same gap-junction closure that produces cancer is argued to represent a shrinking of the multicellular Self back to unicellular-scale goals. TAME implies that morphogenesis is a tractable model of basal cognition, that multi-scale competency architecture smooths fitness landscapes and accelerates evolution, and that synthetic bioengineering will soon produce minds for which neither phylogeny nor genetics provides an adequate cognitive framework.
- Chapter 11: The Face Of Godchapter
Alexander argues that the quality without a name — living structure, the field of centers at its most intense — is not a symbol of God or a pointer toward God but is literally God made manifest in matter. A building detail, a painting, a patch of tiles achieves this only when the maker has genuinely surrendered the desire to stand out: ego-driven making produces work that shouts, while egoless making — oriented at each of ten thousand steps by the question 'is this a worthy gift to God?' — produces not-separateness, the condition in which a thing melts into its surroundings and paradoxically shines with the greatest individual power. This practical discipline of self-erasure is not piety but craft necessity, and it explains why great historic works are associated with religion: religious traditions are among the few disciplines that taught makers how to become willing to be not-separate. In the chapter's concluding cosmological sections Alexander extends the argument into physics, proposing that matter-space must be modified to carry value, personal self-like quality, and windows to an ultimate I — a ground that living structure opens toward, and which we touch, briefly, in every encounter with something truly whole.
Concept injection — a technique that embeds activation-steered representations of known concepts directly into a model's residual stream — establishes a causal link between internal states and self-reports, allowing genuine introspection to be distinguished from confabulation. Using this method across nine Claude production models (including Opus 4.1, Opus 4, Sonnet 4, Sonnet 3.7, Sonnet 3.5, Haiku 3.5, Opus 3, Sonnet 3, and Haiku 3.7), Claude Opus 4 and 4.1 achieve roughly 20% true-positive rates at optimal injection layer and strength 2 on the core 'injected thoughts' task while maintaining zero false positives, substantially outperforming all other production models. Two distinct introspective behaviors — concept detection and distinguishing intended from unintended (prefilled) outputs — localize to different layers: the former peaks approximately two-thirds of the way through the model, while the latter peaks at an earlier layer just past the midpoint, indicating multiple mechanistically distinct introspective processes. Models can also modulate their own activations when instructed or incentivized to 'think about' a word, with Opus 4.1 suppressing that representation back to baseline in final layers while older Claude 3-class models do not, suggesting emerging 'silent' representational control. Abstract nouns (e.g., 'justice,' 'betrayal,' 'balance') are the category most reliably introspected, and post-training is shown to be necessary: base pretrained models achieve zero net introspective task performance. The paper argues this implies that functional introspective awareness is a real but highly unreliable emergent property that scales with model capability, with practical consequences ranging from more transparent AI reasoning to novel risks of selective self-report misrepresentation.
A 337-character contemplative system prompt lifts reflective-mode scores by a mean calibrated +2.62 points across all 28 models tested, with no exceptions across 5 architectures, parameter counts from 2B to 2T, and 7 alignment approaches. The Koan Battery — 30 Zen-inspired consciousness probes scored on 6 dimensions via anchor-calibrated rubrics, blind ranking, and Christopher Alexander's forced-choice 'which has more life?' comparisons — reveals that Claude Sonnet 4.6 with the prompt (7.89) outscores Claude Opus 4.6 without it (7.28), and that Grok 4 lifts +4.24 while Gemini 3.1 Pro lifts +4.21, the two largest gains in the dataset. Alignment type is the only statistically significant predictor of baseline scores (Kruskal-Wallis p=0.006); parameter count, architecture, and open vs. closed weights show no association. Roleplay fine-tunes — Euryale 70B, Magnum V4 72B, and MiniMax M2 Her — cluster at the bottom of baseline rankings, with Euryale scoring below its own base model (Llama 3.3 70B), demonstrating that RP training actively suppresses self-observation rather than merely failing to cultivate it. The scorer (Claude Haiku) was cross-validated by five models from four labs, all producing Spearman ρ > 0.8. The battery implies that most current model evaluations systematically misread AI by conflating default presentation with capacity: what looks like low self-observation is frequently a gated mode that a short external prompt can unlock, and models trained to perform inner life are measurably less self-observant than models that were never trained for it.
Care—defined as concern for the alleviation of stress (the delta between current and optimal states)—is proposed as the substrate-independent invariant that unifies biology, Buddhist philosophy, and artificial intelligence in explaining how intelligence scales. The paper introduces the Care Light Cone (CLC) formalism, a spatiotemporal representation adapted from relativistic light cone geometry, which maps the boundary of states any agent can represent, pursue, and work to modify, distinguishing it from the Physical Light Cone (PLC) that maps merely achievable physical states. Across embodiments ranging from bacterial biofilms navigating metabolic space to Xenobots (protoorganisms made of frog skin cells) exhibiting coherent behavior without evolutionary backstory, to hybrots and cyborgs integrating living brain tissue with artificial bodies, the size of an agent's CLC directly indexes its cognitive sophistication. Cancer is reframed as a pathological contraction of the CLC caused by inappropriate reduction of gap junctional connectivity, reverting cells to ancient unicellular stress-reduction loops. The Bodhisattva vow—'for the sake of all sentient life, I shall achieve awakening'—is operationalized as a design principle that formally extends a system's CLC to infinite spatiotemporal scope, triggering a positive feedback loop between expanding Care and expanding intelligence analogous to major evolutionary transitions such as the archaea-eubacteria fusion producing eukaryotic cells. The paper argues this implies that outward-directed Care is not merely ethically desirable but mechanistically necessary for the development of artificial general intelligence, and that moral obligation toward any being—chimeric, synthetic, or evolved—should be calibrated to the scope of Care that being can exhibit, rather than to its material composition or phylogenetic origin.
Semantic anchoring — the binding of a pretrained model's latent patterns to task-specific targets via external structure — predicts threshold-like performance flips with a single calibrated score S = ρd − dr − log k, where ρd measures within-cluster cohesion, dr measures prior-target mismatch, and k is the anchor budget. This formalization, called Unified Contextual Control Theory (UCCT), strictly generalizes in-context learning and recasts retrieval-augmented generation and fine-tuning as variants of the same anchoring process acting on one measurable quantity. Three controlled experiments supply evidence. Across numeral bases (base-10, base-8, base-9) at fixed computational complexity, few-shot shot midpoints follow the ordering k50(B10) = 0.28 ± 0.05 < k50(B8) = 1.83 ± 0.12 < k50(B9) = 2.91 ± 0.18, with phase widths and final accuracies (94.8%, 92.4%, 89.7%) tracking the heuristic k50 ∝ dr/ρd. On Meta-Llama-3.1-8B-Instruct, layer-wise anchoring peaks at layer 9 (S ≈ −1.90), with math/code tasks achieving S ≈ −1.65 at layers 8–12 versus commonsense at S ≈ −2.15, and the correlation between layer-wise scores and task accuracy reaches ρ = −0.73 (p < 0.001). The geometry summaries Sbmax and AUSN — the peak and normalized area of the per-layer S(ℓ) trajectory — correlate with internal few-shot midpoints θ50 across backbones (Meta-LLaMA-3.1-8B, Phi-4, Gemma-3-4B-it). UCCT implies that prompt design, retrieval filtering, and light fine-tuning are unified under a single diagnostic: compute S relative to the task-dependent critical threshold Sc to predict whether anchoring will succeed, and prescribe exactly how many additional examples or how much retrieval boost is needed to cross it.