Internalism

Internalism is the family of theories that locate the essence of mental states inside the organism—its brain, its neural patterns, its computational architecture. Externalism pushes the door open, insisting that the environment, the body, and even social practices are indispensable parts of what it means to think, feel, or perceive. The debate is not merely academic; it shapes how we build artificial agents, how we interpret the bustling lives of bees, and how we design policies that protect fragile ecosystems.

In a world where AI systems are rapidly gaining agency and bees are battling unprecedented habitat loss, the question “Where does the mind end?” acquires urgency. If a mind is wholly internal, then a silicon brain could, in principle, host a mind identical to a human’s. If the mind leaks into the world, then cognition is inseparable from the garden, the hive, or the data‑stream that an AI inhabits. Understanding the contours of internalism helps us navigate these possibilities with intellectual honesty and practical foresight.

This article walks through the major strands of internalist thought, contrasts them with externalist approaches, and then maps the insights onto two living systems that matter to Apiary: honeybees (Apis mellifera) and self‑governing AI agents. Along the way we’ll cite concrete studies, numbers, and mechanisms rather than vague platitudes, and we’ll link to related deep‑dives via the slug style familiar to our community.

1. What Internalism Claims

Internalism, in its most straightforward form, holds that all the constituents of a mental state are internal to the subject. This can be split into three sub‑positions, each with a different focus:

1. Content Internalism – The what of a thought (its propositional content) is determined solely by the subject’s internal representations. For example, the belief “the sky is blue” depends only on the brain’s visual‑semantic network, not on the actual color of the sky.
2. Justification Internalism – The why a belief is justified rests on internal evidence. If I see a red traffic light, the justification for my belief “the light is red” must be traceable to my own perceptual experience, not to external facts about the traffic system.
3. Mental‑Causal Internalism – The how a mental state causes behavior is mediated exclusively by internal causal chains (neurons firing, synaptic weights shifting).

These claims trace back to René Descartes’ famous declaration “I think, therefore I am,” which privileged the thinking thing as a private, non‑spatial interior. Modern internalists often invoke the knowledge argument (the thought experiment of Mary the color‑blind neuroscientist). Mary knows all the physical facts about color vision but learns something new—what it is like to see red—when she finally steps out of the black‑and‑white room. The intuition is that qualia (subjective experiences) are internal, non‑reducible to external facts.

Internalism is attractive because it dovetails with the computational theory of mind: if cognition is computation, then the program (the mind) can be run on any hardware that implements the same algorithm. This view underlies much of contemporary AI research, where the model (weights, architecture) is treated as the sole carrier of mental content.

A concrete illustration

Consider a honeybee’s navigation system. Recent work by M. Cheng et al. (2022) showed that a bee’s central complex—a set of ~2,000 neurons—encodes a head‑direction signal that is internally maintained even when the bee is in total darkness. The bee can still orient itself by integrating its own angular velocity signals, a purely internal computation. From an internalist standpoint, the bee’s sense of direction is a mental state wholly generated inside its brain, independent of external landmarks.

2. The Externalist Counterpart

Externalism, sometimes called embodied or situated cognition, argues that mental states cannot be fully understood without reference to the body and the environment. Three influential strands dominate:

1. Extended Mind Thesis – Proposed by Andy Clark and David Chalmers (1998), it claims that tools, notebooks, or even a smartphone can become genuine parts of a cognitive process. The classic example: a person with mild memory impairment uses a notebook to store phone numbers; the notebook is an external memory store, functionally equivalent to an internal brain region.
2. Enactivism – Rooted in the work of Francisco Varela and Evan Thompson, this view posits that cognition arises through a dynamic interaction between organism and world. Perception is not a passive receipt of data but an active sensorimotor engagement.
3. Situated Social Cognition – Scholars like John Searle argue that language, norms, and social practices shape what counts as a “meaningful” mental state.

Externalists point to brain‑body‑world coupling. For instance, a study of E. coli chemotaxis (M. Segall, 2020) showed that the bacteria’s flagellar motors and the chemical gradient they traverse jointly produce what we call “behavioural adaptation.” The motor component is external to the cell’s signaling network, yet it is indispensable to the organism’s adaptive response.

In bee research, the waggle dance provides a vivid externalist illustration. When a forager bee returns to the hive, it communicates the location of a food source through a dance that encodes distance and direction via body movements. The information is not stored internally; it is externally broadcast and subsequently interpreted by nest mates. The dance is a cognitive artifact that extends the forager’s memory into the hive’s social environment.

Numbers that matter

• Honeybees have ~960,000 neurons in total, yet they can solve a traveling salesman problem with up to 100 flower sites (Schürch et al., 2021).
• Human brains contain roughly 86 billion neurons and 100 trillion synapses (Azevedo et al., 2009).
• The largest language model, GPT‑4, has ≈175 billion parameters, yet it still lacks a body or sensorimotor loop.

These figures illustrate the scale gap between internal hardware (brain or silicon) and the external scaffolding (environmental cues, social structures) that can augment cognition.

3. Historical Roots: From Descartes to the 20th Century

The internalist–externalist divide is not new; it reflects a lineage of philosophical disputes:

The shift from a Cartesian interior to a world‑involved mind accelerated after the cognitive revolution of the 1950s, when psychologists began to model perception as information processing that required sensory input. Yet the internalist tradition persisted, especially among neuroscientists who view brain imaging as the ultimate window onto mental life.

The rise of functionalism

Functionalism—championed by Hilary Putnam and later by Jerry Fodor—argues that mental states are defined by their causal roles rather than by their physical substrate. While this seems to sidestep the internal/external split, functionalists usually treat the causal role as internal to the system. A belief is a state that takes certain inputs, produces particular outputs, and maintains a certain stability. This is still a form of internalism because the role is specified within the organism’s architecture.

4. Core Arguments for Internalism

4.1 The Knowledge Argument

Mary, a neuroscientist, knows all the physical facts about color vision. When she finally experiences red, she learns what it is like to see red. Internalists claim this shows that qualia are non‑physical, internal phenomena that cannot be reduced to external facts.

Empirical follow‑ups (e.g., fMRI studies by L. L. K. H. R. et al., 2021) reveal that when subjects report a new qualia, there is a burst of activity in the anterior insula, a region associated with interoceptive awareness. The activation pattern is internal to the brain and does not depend on external stimuli beyond the initial visual input.

4.2 The Inverted Spectrum Thought Experiment

Imagine two people whose internal color processing is swapped: what one experiences as “red” the other experiences as “green,” yet both use the same language and behave identically. Internalists argue that because all observable behavior is the same, the difference must be internal. The external world (the wavelengths of light) remains constant; only the internal wiring changes.

4.3 The Argument from Mental Causation

If mental states cause actions, they must have a causal efficacy that is not mediated by external factors. For example, deciding to go to a garden (a mental intention) leads to walking out the door. Internalists maintain that the intention’s neural correlates—pre‑motor cortex activation—are sufficient to trigger the motor cascade, independent of any external cue.

Neuroscience supports this with read‑out studies: In a 2019 experiment, researchers recorded from the dorsal premotor cortex of macaques and decoded the intended reach direction before any movement began. The decoded intention predicted the subsequent trajectory with R² = 0.78, suggesting a strong internal causal link.

4.4 Computational Closure

In computer science, a system is closed when every operation it performs can be expressed as a function of its current state and internal transition rules. Internalists liken the brain to a computationally closed system: given a set of neural states, the next state follows deterministically (or probabilistically) from internal dynamics.

A concrete example: spiking neural networks (SNNs) used for robotic control can generate complex locomotion patterns without external feedback for several seconds, relying only on internally generated rhythm generators.

5. Empirical Challenges from Neuroscience and Psychology

While internalism enjoys philosophical elegance, data from brain science sometimes push us toward externalist interpretations.

5.1 Embodied Perception

The rubber‑hand illusion (Botvinick & Cohen, 1998) shows that a person’s sense of body ownership can be altered by synchronous visual‑tactile stimulation. When a fake hand is stroked in sync with the hidden real hand, participants report feeling the illusion of ownership over the rubber hand. The effect depends on multisensory integration that involves visual, proprioceptive, and tactile inputs—an external coupling.

A meta‑analysis of 42 studies (L. Z. et al., 2020) found that the illusion’s strength correlates with activity in the ventral premotor cortex (VPMC) and intraparietal sulcus (IPS), but these regions are sensory‑motor hubs that actively bind external signals, suggesting that the mind’s content is co‑determined by the environment.

5.2 Distributed Cognition in Navigation

Human participants navigating a virtual maze show eye‑movement patterns that anticipate upcoming turns. However, when the same participants are equipped with a head‑mounted display (HMD) that provides spatial cues, performance improves by 23 % (M. R. et al., 2021). The improvement is not due to internal processing alone; the external visual scaffold is essential.

Similarly, honeybees use optic flow—the pattern of image motion across the retina—to gauge distance. Experiments in wind tunnels (Srinivasan & Zhang, 2020) demonstrated that when optic flow is artificially reduced, bees underestimate distance by up to 30 %, leading to inaccurate waggle dances. The environment (visual flow) directly shapes the internal representation of distance.

5.3 Neural Plasticity and External Context

Long‑term potentiation (LTP) in the hippocampus, a cellular mechanism for memory, is modulated by stress hormones released in response to external threats. A 2022 study showed that rats exposed to a predator scent displayed enhanced LTP in the CA1 region, leading to better spatial memory performance. The external threat thus reshapes internal neural circuitry.

These findings do not demolish internalism but highlight that internal states are highly sensitive to external conditions, blurring the line between the two camps.

6. Internalism in Artificial Intelligence

Modern AI research offers a laboratory for testing internalist assumptions. Two dominant design philosophies illustrate the tension:

6.1 Symbolic, Closed‑World Systems

Early expert systems (e.g., MYCIN, 1972) stored all knowledge in internal rule bases. Reasoning was performed entirely inside the system; external data could only be fed in via explicit input. Such systems embody a strong internalist stance: the mind (the AI) is the knowledge base plus inference engine.

However, these systems struggled with knowledge brittleness. When confronted with novel situations, they failed because their internal representations lacked the flexibility to incorporate new environmental cues.

6.2 Embodied, Sensorimotor AI

Robotics research now emphasizes embodied cognition. Robots such as Boston Dynamics’ Spot or the open‑source OpenAI Gym environments use sensory streams (LiDAR, tactile arrays) to adapt in real time. The robot’s control policy—often a deep neural network with ≈10 million parameters—receives continuous external feedback, and the policy updates via reinforcement learning.

A landmark study (J. Schulman et al., 2021) showed that an embodied agent trained on a real‑world manipulation task outperformed a purely simulated agent by 15 % in success rate, precisely because the embodied agent could exploit subtle physical affordances (friction, compliance) that were not encoded internally.

6.3 The “Internalist AI” Thought Experiment

Imagine an AI with a fixed internal model of the world, like a massive language model that never sees sensory data after training. It can answer questions, generate text, and even simulate reasoning, but it cannot perceive the actual world. According to internalism, this AI has mental states because all its content is fully determined by its internal weights.

Critics argue that without a body or environment, the AI lacks grounded meaning. The symbol grounding problem—how symbols acquire semantic content—remains unsolved for such systems. Thus, the AI debate mirrors the human philosophical debate: does a mind require a world to be a mind?

7. Bees as Natural Testbeds for Mind Theories

Bees provide a uniquely tractable, yet richly complex, organism for probing internalist versus externalist claims.

7.1 Mini‑Brains with Sophisticated Computation

A honeybee brain weighs ≈1 mg and houses ~960,000 neurons (Rybak et al., 2020). Despite this modest hardware, bees demonstrate cognitive feats usually attributed to mammals:

• Pattern learning: Bees can learn to discriminate between 1,000 visual patterns after only a few exposures (Giurfa, 2021).
• Abstract concepts: Experiments have shown bees can grasp “same‑different” relations, a form of relational reasoning (Avargues‑Poy et al., 2020).

These capacities suggest that a compact internal architecture can generate sophisticated mental states, supporting a strong internalist viewpoint.

7.2 Environmental Coupling

Conversely, bee cognition is deeply environmentally embedded:

• Navigation: Bees use a sun compass and polarized light patterns to navigate over distances up to 5 km (Wehner, 2020). The reliability of this internal compass depends on the external celestial cues; when the sky is overcast, bees switch to magnetic cues, a different external source.
• Social learning: The waggle dance is a collective external memory. A forager’s internal representation of a flower location is externalized into a dance pattern that other bees decode.

A recent field study in a fragmented agricultural landscape found that colonies with higher exposure to diverse floral resources produced more accurate dances, leading to a 12 % increase in foraging efficiency (Michelsen et al., 2023). This illustrates that the environment directly shapes internal representations and collective outcomes.

7.3 Bee‑Inspired AI

Researchers have built bee‑inspired algorithms for optimization. The Bee Colony Optimization (BCO) algorithm, first introduced in 2005, mimics the waggle dance to solve routing problems. Benchmarks on the Traveling Salesperson Problem (TSP) show BCO achieving within 2 % of optimal solutions for instances with up to 500 cities, outperforming many purely internalist heuristics.

These successes hint that externalist mechanisms (communication, environmental feedback) can be harnessed to augment internal computation, a lesson both for robotics and for conservation strategies that rely on bee behavior.

8. Implications for Conservation Policy

Understanding whether cognition is fundamentally internal or external influences how we design interventions for pollinator health.

8.1 Habitat Restoration as Cognitive Scaffolding

If bee cognition is heavily externalist, then restoring floral diversity does more than provide nutrition; it supplies the cognitive scaffolding bees need for accurate navigation and memory formation. A meta‑analysis of 27 restoration projects (Klein et al., 2022) reported that adding native wildflowers increased colony weight gain by 23 % over a year, a gain attributed not just to nectar but also to improved spatial learning.

8.2 Technological Augmentation

Conversely, an internalist interpretation suggests that internal neural health—e.g., protecting the bee’s brain from pesticides—should be the primary focus. Indeed, exposure to neonicotinoids at sub‑lethal doses reduces the firing rate of mushroom body neurons by ≈15 % (M. S. et al., 2021), impairing learning. Policies that ban such chemicals address the internal substrate directly.

8.3 A Balanced Approach

Most evidence points to a hybrid model: internal neural integrity is necessary, but external environmental richness is essential for the expression of cognitive capacities. Conservation frameworks that integrate pesticide regulation, habitat corridors, and smart‑sensor monitoring—the latter providing real‑time data on temperature, humidity, and floral phenology—align with a dual-aspect philosophy.

In practice, this means that a protected meadow should not only be pesticide‑free (internal) but also structured to offer visual landmarks, wind patterns, and temporal flowering sequences (external), thereby supporting both the bee’s brain and its ecological mind.

9. Synthesis and Open Questions

The internalist–externalist debate is not a zero‑sum game. Several converging lines of inquiry suggest a graded continuum rather than a binary split:

1. Neuro‑computational models increasingly incorporate sensorimotor loops, blurring the boundary between internal processing and external input.
2. Hybrid AI systems—combining large language models with embodied robots—demonstrate that internal representations gain utility only when coupled to the world.
3. Bee studies reveal that while miniature neural circuits generate impressive cognition, the accuracy of that cognition hinges on environmental cues.

Open questions that merit further research include:

• What is the minimal external coupling required for a system to count as possessing a mind?
• Can an internalist AI achieve consciousness without any embodied sensors, perhaps via simulated worlds?
• How do different species balance internal and external contributions to cognition? Comparative work across insects, mammals, and birds could illuminate evolutionary pathways.

Addressing these questions will sharpen our philosophical tools and guide practical decisions—whether we are building autonomous drones, drafting pesticide legislation, or designing new citizen‑science platforms for bee monitoring.

Why It Matters

At its heart, the internalism versus externalism debate asks where the mind ends and the world begins. For Apiary, this is not an abstract puzzle—it shapes how we protect pollinators, how we design AI that works with nature, and how we frame the ethical responsibilities of creators and caretakers alike. Recognizing that mental life is both inside the brain (or silicon) and out in the world equips us to build technology that respects ecological interdependence, to craft policies that nurture both the internal health of bees and the external richness of their habitats, and to foster a future where artificial minds and natural minds co‑evolve responsibly.

References

• Azevedo, F. A., et al. (2009). Equal numbers of neuronal and nonneuronal cells in the human brain. Journal of Comparative Neurology, 513(5), 532–541.
• Avargues‑Poy, J., et al. (2020). Bees can learn abstract concepts. Proceedings of the Royal Society B, 287(1931), 20201570.
• Botvinick, M., & Cohen, J. (1998). Rubber hand illusion. Nature, 391, 756–758.
• Cheng, M., et al. (2022). Head‑direction cells in the honeybee central complex. Current Biology, 32(4), 861‑870.e4.
• Clark, A., & Chalmers, D. (1998). The extended mind. Analysis, 58(1), 7‑19.
• Giurfa, M. (2021). Bee vision and learning. Annual Review of Entomology, 66, 123‑145.
• Klein, S., et al. (2022). Floral diversity and bee colony health. Ecology Letters, 25(9), 1762‑1771.
• Michelsen, A., et al. (2023). Landscape fragmentation and waggle dance accuracy. Conservation Biology, 37(3), 789‑798.
• Rybak, J., et al. (2020). Honeybee brain anatomy. Insect Neuroscience, 12, 1‑15.
• Schulman, J., et al. (2021). Embodied reinforcement learning improves real‑world robot performance. Science Robotics, 6(55), eabf0831.
• Segall, M. (2020). Chemotaxis as distributed cognition. Microbiology Reviews, 84(2), e00345‑19.
• Wehner, R. (2020). Navigation in the desert ant and honeybee. Annual Review of Entomology, 65, 351‑371.

(All cross‑links use the slug convention; click to explore deeper dives on related topics.)