The Boundaries of Mind: Where Does Cognition Stop?

Series: 4E Cognition | Part: 6 of 9

If Otto's notebook is part of his mind, what about the library down the street? If your smartphone extends your cognition, what about Google's servers storing the information you access? If a team's distributed cognition includes shared documents, does it include the building housing them? The coffee providing alertness? The city infrastructure enabling their work?

Where does cognition stop?

This is the core challenge facing extended mind theory: without clear boundaries, "cognition" risks meaning everything, which means it means nothing. The extended mind thesis liberated us from skull-bound thinking, but it created a new problem—the boundary problem.

Finding principled ways to draw cognitive boundaries turns out to be hard. Not impossible, but requiring careful analysis of coupling, function, and context. This isn't a bug in extended cognition theory—it's a feature revealing that boundaries are real but fluid, context-dependent, and sometimes genuinely ambiguous.

Welcome to the messiest part of 4E cognition: figuring out where minds end.

The Bloat Objection: Everything Can't Be Cognition

The most common objection to extended mind: if we allow notebooks and phones to count as cognition, we're on a slippery slope to absurdity. Does a student using a textbook mean the textbook is part of their mind? Does a driver using GPS mean the satellite network is part of their cognition? Does breathing oxygen mean the atmosphere is cognitive?

This is the cognitive bloat objection: extended mind theory makes cognition so permissive that it loses explanatory power.

Fred Adams and Ken Aizawa argue that Clark and Chalmers's parity principle—if external processes play the same functional role as internal ones, they're cognitive—is too liberal. Lots of things play functional roles in cognition without being cognitive themselves.

Your coffee enables better focus. Does that make coffee cognitive? Your chair supports your body, enabling thought. Is the chair part of your mind?

The answer is clearly no, but why? What distinguishes genuine cognitive extension from mere causal influence?

The Mark of the Cognitive

Adams and Aizawa propose that genuine cognition requires non-derived content—meaning that originates in the system itself rather than being assigned by interpreters. Neural representations have non-derived content (they mean things intrinsically, through causal and functional roles). Written symbols have only derived content (they mean things only because we interpret them that way).

By this criterion, Otto's notebook isn't cognitive because its content is derived—the written words mean things only because Otto interprets them. Biological memory, in contrast, has non-derived content encoded in synaptic weights.

This preserves a clear boundary: cognition stays biological.

But it also undermines much of neuroscience, which treats neural activity as representational only by interpreting it. Is a neuron's firing rate intrinsically meaningful, or do we assign meaning to it? If the latter, even brains might fail the test.

The mark-of-the-cognitive approach tries to solve the boundary problem by restricting cognition to a special biological substrate. It works only if you already believe cognition is special in ways that resist functional analysis.

The Coupling-Constitution Fallacy

Rob Rupert offers a different critique: Clark and Chalmers confuse coupling (causal interaction) with constitution (actually being part of something).

The notebook is coupled to Otto's cognition—it causally influences what he thinks. But causal influence doesn't make something a constituent part. Your food causally influences your metabolism, but it's not part of your metabolism until digested.

By Rupert's lights, the extended mind thesis commits the coupling-constitution fallacy: inferring from "X and Y are coupled" to "X is part of Y."

This preserves boundaries: cognition is constituted by biological brains. External resources couple to cognition without being cognitive themselves.

The Reply: Coupling Can Constitute

Clark's response: sometimes coupling is constitution. Not all coupling, but tight, reciprocal, functionally integrated coupling can constitute genuine system membership.

Consider the visual system. Is the retina part of vision? Obviously yes, despite being outside the brain. The retina and visual cortex are tightly coupled, reciprocally influencing each other, jointly constituting visual processing. The coupling constitutes the visual system.

If tight coupling between retina and cortex makes both parts of one visual system, why can't tight coupling between brain and notebook make both parts of one cognitive system?

The question becomes: what kind of coupling constitutes genuine extension? This shifts debate from whether coupling can constitute to what conditions must be met.

The Criteria: When Does Extension Happen?

Clark refines the extended mind thesis with more specific criteria for genuine cognitive extension:

1. Reliable Availability

The resource must be consistently accessible. Otto's notebook is always with him. A library he visits occasionally doesn't meet this criterion.

2. Automatic Invocation

Using the resource must be automatic, not requiring deliberate effort. Otto consults his notebook without conscious decision, like accessing biological memory.

3. Trust

The information must be endorsed, not constantly verified. Otto trusts his notebook entries like he trusts his (failing) biological memory.

4. Dense Reciprocal Coupling

The resource and brain must interact continuously and bidirectionally. Reading and writing to the notebook create tight feedback loops.

These criteria distinguish genuine extension from mere tool use. Your smartphone probably meets them—it's reliably available, automatically consulted, trusted, and densely coupled. A calculator you use occasionally doesn't.

The Glue and Glow Metaphor

Andy Clark uses a metaphor: cognition spreads like glue into the world when coupling is tight, but only glows on properly cognitive processes. The question is where glue adheres.

Glue adheres to reliably coupled resources that play stable functional roles. It doesn't adhere to transiently useful but uncoupled resources.

By this standard, Otto's notebook gets glued. The library doesn't. Your phone probably does. The internet might, depending on your usage patterns.

The boundaries emerge from coupling patterns, not from pre-established metaphysical categories.

Context-Dependent Boundaries: Extension Isn't All-or-Nothing

One insight from the boundary debate: cognitive boundaries aren't fixed. They're context-dependent and can shift dynamically.

When you're using a calculator, it might temporarily be part of your cognitive system. When you put it down, it isn't. The boundary moves based on functional coupling at the time.

A musician performing doesn't experience their instrument as external—it's part of their embodied cognitive-motor system. When not playing, the instrument returns to being a separate object. The cognitive boundary contracted.

This makes boundaries real but fluid. Cognition doesn't permanently extend into all tools you use—it extends during use when coupling meets the criteria.

Expert Extension

Expertise changes what can be extended. A novice chess player uses the board as external memory—they're coupled to it, but loosely. A grandmaster's cognition is more deeply extended into board perception—patterns are directly visible that novices must compute.

Similarly, an expert mathematician's cognition extends into notation in ways a student's doesn't. The expert thinks through symbols fluidly; the student thinks about symbols effortfully.

Cognitive boundaries depend on skill, practice, and the depth of sensorimotor integration.

Social Boundaries: Where Do Collective Minds End?

If individual minds can extend into tools, can collective minds extend across people?

The Group Mind

Groups exhibit cognitive properties no individual possesses: transactive memory, distributed expertise, emergent problem-solving. Does this make groups genuinely cognitive systems, or just collections of cognitive individuals?

Georg Theiner argues that tightly coupled groups can constitute group minds with cognitive boundaries encompassing multiple brains plus shared artifacts. A surgical team during operation is a cognitive system—distributed attention, coordinated action, shared understanding emerging from tight coupling.

But groups also have boundaries. A loosely coordinated committee isn't a group mind—coupling is too weak. The distinction is, again, coupling density.

Cultural Cognition

Does culture extend cognition to civilization scale? Do scientific communities constitute massive distributed cognitive systems?

Arguably yes, under the right conditions. Science works through distributed labor (specialization), shared artifacts (publications, databases), and coordination mechanisms (peer review, conferences). No individual knows all of science, but the system does.

But the coupling is looser than in individual extended cognition. Most scientists don't tightly couple with most others. The community is more network than unified system.

Cultural cognition might be weakly extended—real but diffuse.

The Geometry of Fuzzy Boundaries

In AToM terms, cognitive boundaries are determined by Markov blankets—statistical boundaries separating system from environment through conditional independence.

A system's Markov blanket defines what it needs to track: internal states and blanket states, but not distant environmental states. For biological organisms, skin approximates the Markov blanket. For extended cognitive systems, the blanket can include coupled artifacts.

But Markov blankets aren't fixed. They're conditional on the timescale and variables of interest. Zoom in, and the blanket tightens to biological boundaries. Zoom out, and it can expand to include tools and environments.

This means cognitive boundaries are scale-dependent and question-relative. There's no single right answer to "where does cognition stop?" The answer depends on what you're trying to explain and at what timescale.

For moment-to-moment problem-solving, Otto's notebook is inside the cognitive boundary. For evolutionary questions about memory, it's outside. Both answers are correct for their respective questions.

Implications: Living With Fuzzy Boundaries

Accepting that cognitive boundaries are real but context-dependent changes how we think about minds:

Identity: If your cognitive system includes your phone, then losing it threatens your identity. Not metaphorically—literally. Your extended cognitive architecture partly constitutes who you are.

Responsibility: If social media algorithms are part of your extended cognition, then you share responsibility for managing that extension. You can't fully outsource cognitive health to platforms.

Design: If boundaries are fluid and coupling-dependent, then designing technologies means designing potential cognitive extensions. The ethical stakes are high.

Therapy: Cognitive interventions can target boundaries—strengthening useful extensions, weakening harmful ones. Not all healing happens through changing brains.

Further Reading

• Adams, F., & Aizawa, K. (2008). The Bounds of Cognition. Blackwell.
• Rupert, R. D. (2009). Cognitive Systems and the Extended Mind. Oxford University Press.
• Clark, A. (2008). Supersizing the Mind: Embodiment, Action, and Cognitive Extension. Oxford University Press.
• Theiner, G., Allen, C., & Goldstone, R. L. (2010). "Recognizing Group Cognition." Cognitive Systems Research, 11(4), 378-395.
• Kirchhoff, M. D., & Kiverstein, J. (2019). Extended Consciousness and Predictive Processing. Routledge.

This is Part 6 of the 4E Cognition series, exploring how cognitive science moved beyond the brain.

Previous: Extended Cognition: When Mind Leaks Into the World
Next: 4E Meets Active Inference: Embodied Free Energy Minimization