Learning

This note is about the concept of learning. It is related to the domain of Teaching and Knowledge.

Definition

Tentative (SR): learning is entropy-resisting pattern formation. Formation here includes acquisition, retention, transmission.

Learning, capacity to change behaviour. Results from individual experiences. Achieves adjustments that are difficult to encode genetically.

Memory, capacity to retain learned information to influence future behaviour.

Wyatt, Tristram D. Animal Behaviour: A Very Short Introduction. Oxford: Oxford University Press, 2017.

In education, learning involves the construction of meaning. Cf. Techniques.

In education:

Ambrose, Susan A., Michael W. Bridges, Michele DiPietro, Marsha C. Lovett, and Marie K. Norman. How Learning Works: Seven Research-Based Principles for Smart Teaching. San Francisco: John Wiley & Sons, 2010.

Learning is a process that leads to change. It occurs as a result of experience and increases the potential for improved performance and future learning.

Aspects:

Learning is a process, not a product. Observers can only infer it from products or performances.
Learning accumulates over time and produces a lasting change in knowledge, beliefs, behaviours or attitudes.
Learning is not something done to learners. Learners do it themselves through conscious or unconscious responses to experience.

^{Summary by Diana Laurillard}

Laurillard, Diana. Teaching as a Design Science: Building Pedagogical Patterns for Learning and Technology. New York: Routledge, 2012.

How can this diagram be redrawn to include distributed and nonhuman learning?

Thoughts

Ignorance (or the continuation of the current state) is often more comfortable, or energetically prudent, than learning. Learning implies the work of changing.

Learning and teaching are forms of Violence. They are energetically and structurally expensive. They are risky.

The value of learning depends on its scope. Learning new to all life ranks highest, followed by learning new to humanity, then to human cultures, generations and groups. Learning that only matches the tropes of an existing group culture ranks even lower. The lowest is the learning by individuals (e.g., human individuals), such as self-learning. Such within-culture learning can empower human cultures, but it also causes harm because of its limited scope.

By contrast, valuable learning is slow and addresses relationships, mutualistic or otherwise. It relates to justice through the active inclusion of related agents.

Valuable learning accrues. Otherwise, learning becomes an expensive indulgence. In many situations, such accrual requires active and innovative support.

Any interpretation of reality is intrinsically violent. Cf. ontological violence. Learners can avoid physical violence, but some ontological violence is intrinsic to all knowledge.

Oksala, Johanna. Foucault, Politics, and Violence. Evanston: Northwestern University Press, 2012.

Science

By a serious cognitive neuroscientist:

Dehaene, Stanislas. How We Learn: Why Brains Learn Better Than Any Machine ... for Now. New York: Viking, 2020.

Learning in Living Systems

Bartlett, Stuart, and Michael L. Wong. ‘Lyfe: Learning to Learn Better.’ Interface Focus 15, no. 6 (2025): 20250019. https://doi.org/10.1098/rsfs.2025.0019.

Gunawardena, Jeremy. ‘Learning Outside the Brain: Integrating Cognitive Science and Systems Biology’. Proceedings of the IEEE 110, no. 5 (2022): 590–612. https://doi.org/10/gqd36c.

A General Definition of Learning

Bartlett and Wong propose learning as one of four pillars of ‘lyfe’ (alongside dissipation, autocatalysis and homeostasis). They define learning as:

the ability of a system to record information about its external and internal environment, process that information and carry out actions that feed back positively on its probability of surviving or proliferating.

This definition separates learning from memory. Memory only records. Learning also processes the record and acts on it. The Moon retains bootprints, but the Moon does not learn. The distinction matters for Cognition and Intelligence because it sets a functional rather than a substrate-bound threshold. Any system that closes the loop from sensing to action with persistent benefit qualifies, regardless of whether it has neurons.

This frame extends, rather than replaces, Darwinian evolution. Bartlett and Wong treat natural selection as one mechanism of biological learning among many. Cf. Evolution.

Multiple Mechanisms and Time Scales

Life on Earth learns through layered mechanisms that operate at different time scales:

Darwinian evolution. Intergenerational selection on heritable variation. The slowest channel.
Epigenetics. Intragenerational regulation of gene expression that can pass to offspring. Couples experience to inheritance.
Horizontal gene transfer. Exchange of genetic material across lineages. Builds the pangenome and reframes life as a web rather than a tree.
CRISPR-mediated adaptive immunity. Prokaryotes that record viral sequences and reuse them to recognise future attacks.
Protein computation. Subsecond information processing through allosteric switches, phosphorylation cascades and signalling networks. Bacterial chemotaxis builds a chemical map of the environment.
Supramolecular processing and evo-devo. Cytoskeletal memory, bioelectric signalling and developmental modification, which permits exaptation and the evolution of evolvability.
Cultural memory, memes and the dataome. Symbol-idea transmission through language, story, artefact and infrastructure. Indigenous Australian oral traditions encode environmental knowledge across more than 10,000 years.
Bayesian inference, science and causal discovery. Iterative model updating against world states.
Machine learning and artificial intelligence. A new dataomic lineage that coevolves with humans. Cf. Artificial Intelligence.

Each mechanism trades speed against scope. Faster mechanisms respond to short-term fluctuation. Slower mechanisms encode persistent structure. A planetary biosphere learns at all of these scales at once.

Learning as Optimisation Plus Novelty Search

Bartlett and Wong argue that learning combines two pressures:

Optimisation. Refining what already works.
Novelty search. Generating difference for its own sake.

Novelty search escapes local optima and supports open-endedness. Without it, populations and minds stagnate. With it, they can find paths that pure fitness gradients hide. This balance maps directly onto Creativity and onto design practice. It also reframes ‘failure’ as a productive contribution to a longer learning trajectory.

Three Feedbacks That Shape Learning Systems

Bartlett and Wong propose three feedbacks that may govern any learning system:

Learning and dissipation. Better prediction lets a system harness more free energy, which in turn funds further learning. Information has a thermodynamic price (Landauer). Cognition is energetically expensive.
Learning and environmental complexity (autocatalysis of learning). Learners change the environments they learn from. Each successful adaptation undermines its own conditions and forces deeper learning. Cf. Complexity and Red Queen dynamics.
Self-modelling negative feedback. As a cognitive system grows, it must devote a rising fraction of its capacity to modelling itself. Beyond a threshold, this slows further outward growth. The trajectory of learning power may therefore be sigmoidal rather than exponential, with implications for AI and the Fermi paradox.

Learnability Niche Construction

Life may not only adapt to its environment but also modify the environment to make it more learnable. Domestication, agriculture, the built environment and scientific instruments all illustrate this. Cf. Gaia. The reverse also matters: a niche can be deliberately rendered less learnable for some agents (concealment, deception, propaganda). Cf. Niche.

Implications for the Initial Definition

The ‘entropy-resisting pattern formation’ formulation above is broadly consistent with Bartlett and Wong, but it understates two points:

Learning is not just pattern retention. It is pattern use that feeds back on persistence.
Learning entails dissipation. Resisting entropy locally always exports entropy elsewhere. The cost is part of the definition, not external to it.

More-than-Human Learning

Cf.

^{More-than-human mutual learning.}

As an extension of Vygotsky's Zone of Proximal development and societal scaffolding. This diagram does not need to presume that an agent is an organism or an individual but for an example presumes a human and an animal. It could be a group of any kind, an organisation, an ecological community (need to consider the implications).

This learning can occur as exchanges of meanings, habits, etc. through chains or networks (birds see a cat and make a noise, dog hears the noise and barks because it knows a cat is around, humans hear the dog and the birds and know this too).

The others do not need to be more knowledgeable in general, they just might know different things.

Group members can have different levels of knowledge, ability, experience, etc.

Scaffolding and support are necessary for learning in human societies and this also applies to more-than-human collectives.

Vygotsky, Lev. Mind in Society: The Development of Higher Psychological Processes. Edited by Michael Cole, Vera John-Steiner, Sylvia Scribner, and Ellen Souberman. Cambridge: Harvard University Press, 1978.

^{Mapping of knowledge production and mutual learning.}

These zones of learning can be regional. They need active maintenance and reproduction, or re-learning. They are inherently collaborative, but the flow of meaning can break even when it remains in potential.

The volume and resolution of this knowledge varies. Scientific, engineering, geometrically mapped knowledge of contemporary human societies further overlays this in a patchwork fashion.

It should be possible to produce volumetric maps of knowledge density to illustrate gap and opportunities. For example, human knowledge will diminish further away from the shores, with depth under the water and with height in the atmosphere.

Similarly, it should be possible to assess and map the continuity and multiplicity of the knowledge quilt, effectively measuring the degree and patterns of fragmentation.

Based on this, it should be possible to express and map a measure of cohesion.

Further, it should be possible to estimate and map rates of knowledge production and dissipation.

Such measures can be important to underpin the valuing of the planetary places and as a way to inform approaches to more-than-human political representation.

Questions:

the total volume of available knowledge
means of knowledge production
patterns of knowledge-to-action transfer
effects and benefits of knowledge

Sterelny, Kim. ‘The Evolution and Evolvability of Culture’. Mind & Language 21, no. 2 (2006): 137–65. https://doi.org/10/bv4c68.

Engagement

What forms of engagement are possible with nonhuman beings?

With humans:

Behavioural: participation
Emotional: positive and negative
Cognitive: effort to comprehend ideas or skills

With nonhuman beings

behavioural
cultural
genetic
co-living
metabolic exchange
labour exchange
...

Open Questions for Future Research

The 'lyfe' frame opens several questions that connect learning to design, pedagogy, intelligence and political representation.

On the nature of learning.

Which of the nine biological learning mechanisms have analogues in human institutions and design processes? Where are the gaps?
Can the optimisation-plus-novelty-search frame replace the loose use of ‘innovation’ in design discourse? Cf. Creativity.
How do we measure the learnability of a place, a building or a curriculum? Could a quantitative ‘learnability index’ guide design briefs?

On more-than-human learning.

Can the Vygotskian Zone of Proximal Development extend to interspecies and ecosystem-scale collectives without anthropomorphic distortion? What scaffolding can a designer provide to a more-than-human learning collective?
How does learnability niche construction by humans (cities, monocultures, lighting, noise) erode the learnability of the same niche for other agents? Cf. Biosemiotics.
Can volumetric maps of knowledge density (proposed above) include nonhuman knowledge holders? What metrics make such inclusion meaningful?

On energetics and complexity.

If learning autocatalyses environmental complexity, design that aims at long-term planetary inhabitance must learn to slow, not only accelerate, certain learning loops. Which loops, and by what means?
The thermodynamic cost of cognition implies an energy ethics for AI, education and research infrastructure. How should this cost enter design and policy decisions? Cf. Value.

On consciousness-driven negative feedback.

If high-capacity cognitive systems must spend a growing share on self-modelling, what does this imply for institutional reflexivity, for design pedagogy and for the governance of AI?
The ‘sapezoic’ idea (life recognising its own causal power on a planetary scale) reframes the role of design as a planetary self-modelling practice. What competencies does that require?

On pedagogy and design education.

1Laurillard’s conversational framework should be redrawn to include nonhuman, distributed and machine learners. What does the redrawn diagram look like?
Studio teaching already balances optimisation (critique against a brief) with novelty search (open exploration). Can the Bartlett and Wong frame make this balance explicit and assessable?
The ‘misinformation death spiral’ that Bartlett and Wong describe for AI-generated content has direct implications for design research, citation practice and the integrity of evidence. What pedagogical defences scale?

On political representation and justice.

If valuable learning relates to justice through inclusion of related agents, then loss of Indigenous, ecological and craft knowledge is a loss of planetary learning capacity. How should design valorise and protect these channels?
Can the Dataome concept ground a politics of more-than-human knowledge commons? Cf. Commons.