paper
archived
1985
ⓒ 146
paper:synthetic-article-review

Synthetic Article & Review

ByB.C. Barber
DOI 10.1080/01431168508948262OpenAlex W1975419631

TL;DR

Evolutionary individuality, organismic individuality, and cognition are coextensive — this is the load-bearing conjecture Watson and Levin develop across 22 pages in Collective Intelligence (2023, DOI: 10.1177/26339137231168355). The argument rests on dissolving three apparent categorical distinctions: all individuals are collectives (from molecules to multicellular organisms via major evolutionary transitions), all intelligences are collectives (brains are the archetypal case), and cognition is substrate-independent. The framework they introduce is evolutionary connectionism, which maps the algorithmic principles of connectionist models — specifically Hopfield networks, multi-layer perceptrons, deep belief networks, and Hebbian unsupervised learning — onto gene-regulatory networks, bioelectric networks, and ecological community networks. Empirically, the framework is grounded in planarian bioelectric pattern memory (transient gap-junction modulation producing heritable two-headed regeneration despite unaltered genomes), ectopic Xenopus eyes wired via spinal cord synapses that still support light-mediated learning, and newt kidney tubules that achieve correct diameter whether using ~8 normal-sized cells or a single enlarged cell bending on itself. Morphogenesis is reframed as basal cognition — anatomical problem-solving in morphospace — and the Hopfield network is used to show how unsupervised, distributed Hebbian learning at the component level produces system-level associative memory and generalisation without centralized control or global reward feedback. The paper argues this implies that the capacity to represent non-linearly separable functions (requiring multi-layer interaction depth) is both the necessary condition for genuine collective intelligence and the very thing that constitutes a new level of evolutionary individuality, with direct consequences for regenerative medicine, cancer as dysregulated morphogenesis, and the engineering of synthetic living machines.

What to take away

  1. 1. Evolutionary individuality, organismic individuality, and cognition are proposed to be coextensive: the causal structures required to produce fitness that belongs properly to a collective are identical to those required to produce information integration and coordinated action characteristic of a cognitive self.
  2. 2. Transient bioelectric modulation of gap-junction networks in planaria — without any genomic alteration — shifts the body-wide pattern memory to a persistent two-headed state that is inherited through subsequent rounds of amputation and exhibits bistability, demonstrating that physiological signals can re-write anatomical target morphology independently of genetic hardware.
  3. 3. Ectopic eyes grafted to Xenopus tadpole tails still support light-mediated learning because eye primordia cells successfully wire the optic nerve to the spinal cord rather than the brain, illustrating multi-scale competency architecture in morphospace navigation.
  4. 4. Newt kidney tubule diameter is achieved whether ~8 normal-sized cells communicate to set it, fewer large cells do the same, or a single very large cell bends around itself using cytoskeletal mechanisms — meeting William James's 1890 definition of intelligence as achieving the same goal by multiple means.
  5. 5. The framework the paper introduces — evolutionary connectionism — maps Hopfield network dynamics, multi-layer perceptron depth requirements, deep belief network architectures, and Hebbian unsupervised learning onto gene-regulatory, bioelectric, metabolic, and ecological networks to specify necessary and sufficient interaction structures for collective intelligence.
  6. 6. Representing non-linearly separable functions (e.g., XOR-type fitness epistasis, or reciprocal sign epistasis per Weinreich et al. 2005) requires interaction depth beyond a single-layer perceptron, and this depth requirement is argued to be the precise structural condition distinguishing genuine collective intelligence from mere coordination.
  7. 7. Individual-level reinforcement learning acting on connection strengths between components produces dynamics that are formally equivalent to unsupervised Hebbian learning at the system scale, meaning system-level associative memory and generalisation can emerge without system-level reward or selection (demonstrated in ecological community network models, Power et al. 2015).
  8. 8. Cancer is interpreted within this framework as disconnection from the bioelectric tissue network releasing cells from higher-level collective goals, and the cancer phenotype can be suppressed by restoring bioelectric coupling among cells even in the presence of strong oncogenic mutations (Chernet and Levin 2013), validating the practical translational arm of the theory.
  9. 9. A replicable methodology for testing substrate-independent cognition claims is to apply the Hopfield network as a null model — its fixed-point attractors represent baseline local energy descent — and then ask whether a biological network (gene-regulatory, bioelectric, or ecological) exhibits learning-induced improvement in constraint satisfaction beyond that baseline, using the Watson et al. 2011a self-modelling dynamical system protocol.
  10. 10. An open hypothesis the paper raises is whether major evolutionary transitions in individuality correspond specifically to the emergence of deep (hierarchical, multi-layer) interaction structures rather than single-level symmetric networks, predicting that the 'chunking' or rescaling of search space demonstrated in deep optimisation models (Caldwell et al. 2018, Mills 2010) will be found to parallel the algorithmic signature of each transition.

Peer brief — for seminar discussion

Watson and Levin's 2023 paper in Collective Intelligence (DOI: 10.1177/26339137231168355) develops a unified theoretical framework by arguing that the apparent categorical boundary between individual intelligence and collective intelligence dissolves once three propositions are accepted: all individuals are themselves collectives at multiple hierarchical levels (from pre-cellular replicators through eukaryotic cells to multicellular organisms), all intelligences are collectives (neural networks being the archetypal case), and cognition is substrate-independent, implementable by any network of signals and nonlinear responses. From these premises, the authors derive what they call evolutionary connectionism — a formal mapping of connectionist machine learning architectures (Hopfield recurrent networks, single-layer perceptrons, multi-layer perceptrons, deep belief networks from Hinton et al. 2006, and deep autoencoders) onto biological networks including gene-regulatory, bioelectric, metabolic, and ecological community networks. The framework is anchored to three empirical cases: planarian pattern memory re-written by transient gap-junction modulation producing heritable two-headed regeneration despite unaltered genomes (Durant et al. 2016, 2017); ectopic eyes in Xenopus tadpole tails that synapse onto spinal cord rather than brain yet still support light-mediated learning (Blackiston and Levin 2013); and newt kidney tubule diameter regulation achieved identically by ~8 normal cells, fewer enlarged cells, or a single cell bending on itself (Fankhauser 1945a, 1945b). The load-bearing finding is that representing non-linearly separable functions — the XOR-type credit-assignment problem that a single-layer perceptron cannot solve — requires interaction depth, and this depth requirement is both the necessary structural condition for genuine collective intelligence and, by hypothesis, the condition that constitutes a new level of evolutionary individuality. An alternative theoretical vehicle the paper could have used is active inference / free-energy minimization (Friston et al. 2015 is cited but not adopted as the primary formalism), which would have offered a different account of morphogenetic goal-directedness without invoking connectionist learning architectures. What implies the most about future research is the unsupervised learning result: individual-level Hebbian reinforcement on connection weights produces system-scale associative memory and generalisation without any system-level selection or reward, meaning evolutionary individuality and collective cognition could bootstrap each other before higher-level selection takes effect. The paper makes an explicit prediction: major evolutionary transitions in individuality correspond algorithmically to the emergence of deep, hierarchical interaction structures rather than single-level symmetric networks, and this predicts that the search-space rescaling demonstrated in deep optimisation models will appear as a signature of each transition. The contestable point a critical reader should press is the empirical thinness of the substrate-independence claim: the three core biological examples (planaria, Xenopus, newt) demonstrate morphogenetic plasticity consistent with basal cognition, but none directly demonstrates that the specific connectionist learning algorithm — Hebbian weight update on recurrent connections producing generalising associative memory — is the actual causal mechanism operating in those tissues rather than a useful formal analogy. The paper acknowledges this gap explicitly in its roadmap table, flagging several biological systems where the relevant topology is unknown, but the gap between 'formally equivalent' and 'mechanistically identical' is precisely what a skeptical empiricist would demand be closed before treating evolutionary connectionism as a predictive rather than a heuristic framework.

Findings (2)

Claims (3)

Hypotheses (1)

Original abstract (expand)

Abstract Digital synthetic-aperture radar (SAR) imaging techniques have previously only been reported in the literature in a fragmentory manner. This article presents a comprehensive review of the theory of digital SAR imaging from Earth-orbiting satellites. The digital SAR imaging process is explained, including a discussion of various aspects which are specific to satelliteborne SAR. A number of relevant digital-processing techniques are reviewed and it is shown how these techniques may be applied to the processing of digital SAR data. The range migration problem is discussed and various techniques for overcoming it are presented. The paper should be useful not only to the designer of SAR processors, but also to the user of digital SAR data and images.

Related work— refs + corpus + external arXiv

Cited / in-corpus / arXiv badges show which signals surfaced each row. Multi-source rows weighted higher.

+18 more

Similar preprints — Semantic Scholar

Cross-corpus bridges (12)

same_concept_as · Nomic cosine

External markdown files that talk about the same concept as this entity.

  • aboutblank_kb
    Are evolutionary individuality and organismic individuality dependent on the same cognitive architecture?questions/are-evolutionary-individuality-and-organismic-individuality-dependent-on.md0.853
  • aboutblank_kb
    What is the relationship between individual learning, collective intelligence, and the emergence of organismic individuality?questions/what-is-the-relationship-between-individual-learning-collective.md0.852
  • aboutblank_kb
    Are evolutionary individuality, organismic individuality, and cognition coextensive phenomena arising from the same deep principles?questions/are-evolutionary-individuality-organismic-individuality-and-cognition-coextensive.md0.848
  • aboutblank_kb
    Connectionist Framework For Collective Intelligenceframeworks/connectionist-framework-for-collective-intelligence.md0.836
  • aboutblank_kb
    Can connectionist models of individual cognition and learning be validly extended to explain collective intelligence in non-neural systems?questions/can-connectionist-models-of-individual-cognition-and-learning.md0.833
  • aboutblank_kb
    Connectionist Approach To Evolutionary Transitions In Individualityframeworks/connectionist-approach-to-evolutionary-transitions-in-individuality.md0.814
  • aboutblank_kb
    Vol.:(0123456789)papers/cleaned/s10071-023-01780-3.md0.813
  • aboutblank_kb
    Can unsupervised learning mechanisms alone generate collective intelligence without system-level feedback or pre-existing evolutionary units?questions/can-unsupervised-learning-mechanisms-alone-generate-collective-intelligence.md0.810
  • aboutblank_kb
    Hopfield Networksframeworks/hopfield-networks.md0.802
  • aboutblank_kb
    Design For An Individual (Hypothesis H1)frameworks/design-for-an-individual-hypothesis-h1.md0.802
  • aboutblank_kb
    Hierarchically Mechanistic Mindframeworks/hierarchically-mechanistic-mind.md0.790
  • aboutblank_kb
    Connectionist Models Of Cognitionconcepts/ai/connectionist-models-of-cognition.md0.785