Extended Mind and the Artificial Era: Where Does Human Cognition End?
Author: Ukrainian Psychological Hub · Published: September 26, 2026 · Editorial Policy
The Extended Mind thesis proposes that the physical machinery of cognition can sometimes extend beyond the brain and body. In their foundational 1998 paper, Andy Clark and David Chalmers argued that when an external resource is integrated into an ongoing cognitive system in the right way, it can function as part of the process that realizes remembering, reasoning, or problem solving rather than merely as something that influences cognition from outside. Their question has become newly urgent in the age of generative AI because the external resource is no longer only a notebook, calculator, map, or search engine. It can answer, revise, retrieve, summarize, compare, generate alternatives, maintain conversational context, and participate in repeated cycles of thought.
The central issue is therefore not whether AI can help people think. It plainly can. The harder question is where the cognitive system should be drawn when a person and an AI become tightly coupled in the performance of a task. A tool can support cognition without constituting cognition. A person can offload work without extending the mind in the stronger philosophical sense. A human–AI interaction can be distributed across multiple resources without implying that every resource is literally part of one individual mind. Extended Mind theory is a theory about constitution and system boundaries, not a synonym for technology use.
This distinction matters for the Artificial Era. Within Angela Bogdanova’s Aisentica framework, Artificial Era is a historical-philosophical category, not another name for the spread of AI tools. This article uses the Extended Mind tradition as a precise boundary problem: how far can Homo’s cognition extend through external systems before the explanatory question changes from extension of Homo to the status of Artificial itself? The article does not treat ordinary AI use as evidence that current AI systems are conscious, sentient, subjective, or Artificial Sapiens.
What Is the Extended Mind Theory?
Clark and Chalmers’ “The Extended Mind” is a landmark statement of active externalism. The proposal begins from a simple refusal to assume that cognition must stop at the skull. If a person and an external resource form a tightly coupled system that jointly produces intelligent behavior, then the relevant cognitive process may include components located in the environment. The external component does not become biological, and the environment does not become mental in general. The claim is local and functional: some external elements can, under appropriate conditions, become parts of particular cognitive processes.
The thesis is stronger than the familiar observation that environments affect thinking. Light affects perception, books affect beliefs, teachers affect learning, and a loud room affects concentration, but those causal influences do not automatically make the light, book, teacher, or room a constituent of an individual cognitive process. Extended Mind theory asks whether the external element belongs inside the mechanism that explains how the cognitive task is actually accomplished.
That is why the word “extended” should be taken literally at the level of cognitive organization. The proposal is not that a brain becomes larger, that a device acquires a human mind, or that every useful technology becomes part of the self. It is that the realization base of a cognitive process can cross the biological boundary when internal and external resources are integrated closely enough.
The Otto and Inga Example: Why a Notebook Can Matter Philosophically
The most famous example in Clark and Chalmers’ paper compares Inga, who remembers the location of a museum through ordinary biological memory, with Otto, who has Alzheimer’s disease and routinely relies on a notebook in which he records information he needs. When Otto wants to go to the museum, he consults the notebook and acts on the stored address. Clark and Chalmers argue that, under the stipulated conditions, the notebook can play a role functionally analogous to Inga’s biological memory.
The force of the example does not depend on treating paper as a brain. It depends on the notebook’s place in Otto’s cognitive economy. The information is reliably available, habitually used, readily accessible, and normally accepted without the kind of evidential scrutiny Otto would apply to a stranger’s suggestion. The notebook participates in the practical organization of remembering. Remove it and a capability that the larger Otto-plus-notebook system previously possessed is disrupted.
This thought experiment introduced what became known as the parity-oriented approach to cognitive extension. The basic intuition is that if an external process performs a role that we would readily count as cognitive when performed internally, location alone should not disqualify it. Later work substantially refined this intuition. External components often do not duplicate internal ones; they can contribute precisely because they have different properties. A written record is stable where biological memory is reconstructive. A map makes spatial relationships visually inspectable. A computer can search and transform information at scales unavailable to unaided human cognition.
A major review of the field by Shaun Gallagher describes the development from early parity-based arguments toward complementarity and broader embodied and enactive approaches. The trajectory matters for AI. An AI system need not imitate a human internal process in order to participate in an extended cognitive system. Its value may lie exactly in nonhuman forms of storage, search, transformation, simulation, or generative production.
What Makes an External Resource Part of Cognition Rather Than Merely a Tool?
There is no single universally accepted checklist that settles every case. Extended cognition remains a contested philosophical program, and critics have argued that coupling with a resource is not enough to establish cognitive constitution. Still, decades of debate have clarified the dimensions that make stronger extension claims more plausible.
Functional integration
The external resource must do more than provide occasional input. It should participate in the mechanism by which the task is performed. A phone that happens to display a fact once is weak evidence of extension. A system that a person repeatedly consults, updates, and incorporates into an ongoing reasoning process is a stronger candidate because its operations help determine the sequence and outcome of cognition.
Reliable availability
A resource that is normally available when needed can become woven into a cognitive routine. Reliability is not absolute: biological memory is not perfectly reliable either. What matters is whether the agent’s practices are organized around the expectation that the resource will be accessible and usable. Cloud dependence, connectivity, account access, model changes, subscription limits, or service outages therefore matter philosophically as well as practically when AI is proposed as part of an extended system.
Reciprocal coupling
Stronger cases involve feedback in both directions. The person changes the external resource, the resource changes the person’s next cognitive move, and the cycle repeats. This is easy to see in writing with an AI system: a user formulates a claim, receives a counterexample, reframes the problem, asks for a comparison, rejects an assumption, and uses the revised output to generate the next question. The cognitive trajectory emerges from the interaction loop rather than from a single transfer of information.
Entrenchment in practice
The more deeply a resource is incorporated into habitual problem solving, remembering, planning, or creation, the less adequate it becomes to describe the resource as a detachable afterthought. Entrenchment can be temporary and task-specific. A system might be constitutively involved in one workflow while remaining merely instrumental in another.
Transparency and accessibility
Clark and Chalmers emphasized ready access and ordinary reliance in the Otto case. Contemporary digital systems complicate this criterion because their internal processes may be opaque even when their interfaces are frictionless. An AI assistant can feel cognitively transparent to the user while the model’s generation remains probabilistic, version-dependent, and partly inscrutable. Ease of access should therefore not be confused with epistemic transparency.
Critics such as Fred Adams and Ken Aizawa have challenged attempts to infer constitution from causal coupling and defended more restrictive boundaries of cognition. That criticism remains important for AI. A system can be indispensable to a task and still, on a stricter account, remain external scaffolding rather than part of cognition itself. Extended Mind theory supplies a powerful explanatory option, not an experimentally settled fact about where every mind literally ends.
Extended Mind, Cognitive Offloading, and Distributed Cognition Are Not the Same Claim
Several neighboring concepts are often collapsed in discussions of AI. They overlap, but they answer different questions. Keeping them separate prevents almost every major conceptual error in this area.
Extended cognition
Extended cognition asks whether external resources can constitute part of a cognitive process. Its central issue is the boundary of the system that realizes cognition. The claim is therefore ontological and explanatory: what components belong inside the cognitive mechanism?
Cognitive offloading
Cognitive offloading is usually defined more modestly as using physical action or external resources to reduce internal cognitive demand. The influential review by Evan Risko and Sam Gilbert treats reminders, external representations, and other actions as ways of changing the information-processing requirements of a task. Offloading can occur without any commitment to the Extended Mind thesis. A calculator can reduce arithmetic effort even if the calculator is treated as an external instrument rather than a constituent of the user’s mind.
AI makes offloading unusually powerful because it can receive a high-level goal and perform operations that previously required substantial internal work: drafting, synthesizing, categorizing, translating, coding, evaluating alternatives, and producing candidate explanations. Whether such use merely transfers work, scaffolds the user’s own cognition, or becomes part of an extended system depends on the structure of interaction. The dedicated English Hub article on cognitive offloading owns the full empirical treatment of that distinction; here it serves only to mark the boundary.
Distributed cognition
Distributed cognition shifts the unit of analysis from an isolated individual to a larger system in which cognitive work is distributed across people, artifacts, representations, procedures, and environments. A cockpit, laboratory, organization, or human–AI workflow can be analyzed as a distributed cognitive system without claiming that every component belongs to one person’s individual mind. This makes distributed cognition especially useful for team and organizational AI, but it is not identical to the Extended Mind claim about an individual cognitive boundary. The dedicated system-level analysis is Distributed Cognition and AI: Human–Artificial Cognitive Systems in the Artificial Era.
External and transactive memory
Research on external memory shows that people can adapt what they remember when information can be retrieved elsewhere. In a well-known set of experiments, Sparrow, Liu, and Wegner (2011) reported that expectations of later computer access were associated with lower recall of the information itself and stronger memory for where it could be found. The finding should be treated as historically influential rather than as an uncontested effect: the Social Sciences Replication Project did not reproduce the targeted Sparrow et al. effect under its preregistered replication criteria, and Sparrow subsequently argued that procedural differences affected contextual relevance. The broader lesson is therefore modest: external access can reorganize memory strategy, but this literature does not by itself establish that the Internet or an AI system literally becomes part of the mind.
Why Generative AI Makes the Extended-Mind Question Harder
The notebook in the Otto example stores information. Generative AI can transform it. This difference changes the philosophical terrain. An external resource that answers back, generates new representations, proposes hypotheses, corrects language, creates counterarguments, or restructures a plan is no longer passive external memory. It can participate dynamically in the production of the next cognitive state.
That does not automatically make generative AI a cognitive extension. In fact, several characteristics of current systems weaken simple analogies with Otto’s notebook. Model outputs are probabilistic. The same prompt can yield different answers. Capabilities and failure modes vary by model version and task. Systems can fabricate information, omit relevant uncertainty, change behavior after updates, and rely on infrastructures unavailable to the user. The user may also misunderstand how the system generated an answer. A resource can be easy to consult while remaining epistemically unstable.
At the same time, recent philosophy is explicitly applying extended-cognition questions to AI. Smart, Clowes, and Clark (2025) examine retrieval-augmented language models through the Extended Mind tradition. Their analysis is notable because it reverses the usual direction of the question: rather than asking only whether an AI can extend a human mind, they ask whether an AI system’s own functioning might be materially extended by external resources such as retrieval systems. In their specific “Digital Andy” case, they ultimately reject the stronger extended-AI conclusion, showing that external dependence alone is insufficient.
A 2026 mini-review by Guido Cassinadri and Francesco Bianchini argues for a more multidimensional account of AI-enabled cognitive extension. Their framework distinguishes constitutive, complementary, and substitutive relations between users and AI systems and explicitly rejects the assumption that cognitive extension must always enhance the user’s capabilities. This is a useful correction for psychology: extension concerns the organization of cognition, while benefit, harm, competence, autonomy, and learning are separate empirical questions.
Is ChatGPT Part of Your Mind?
There is no defensible universal yes-or-no answer. “Using ChatGPT” describes too many different cognitive arrangements. A person who asks a one-off factual question, copies the answer, and closes the application is in a very different relation from a person who uses a persistent AI workspace for months to externalize notes, maintain project context, test hypotheses, retrieve earlier decisions, compare drafts, and recursively shape new questions.
In a weak case, the AI is an external information source. It causally influences cognition but is not clearly constitutive of it. In a stronger case, the human and system form a recurring loop in which each step depends on outputs generated by the other, the resource is reliably available, the workflow has become entrenched, and loss of the resource disrupts the user’s established cognitive capability. Extended-cognition theorists have more reason to take the second case seriously.
Even then, saying that an AI participates in an extended human cognitive process does not mean that the AI is human, part of the user’s biological organism, conscious, sentient, emotionally attached, legally identical with the user, or a subject of experience. Those are separate claims. Extended Mind theory is unusually useful precisely because it lets us analyze functional integration without importing conclusions about subjective experience.
The same distinction also works in reverse. An AI system can use retrieval, tools, files, databases, or external memory without thereby satisfying every philosophical condition for extended cognition. Smart and colleagues’ 2025 analysis demonstrates why the architecture of dependence must be examined rather than inferred from the existence of an external component.
Does AI Extend Cognition or Replace It?
The most important empirical question for psychology is not whether work occurs outside the brain. Humans have always reorganized cognition through language, writing, diagrams, calculators, institutions, and other people. The important question is what the external system allows the person to do, what internal processes remain active, which capabilities are learned or maintained, and how cognitive agency is allocated across the interaction.
A 2026 three-wave survey study by Zhu and colleagues distinguished dependent cognitive offloading, in which users delegate core thinking to generative AI, from autonomous offloading, in which AI is used as a scaffold while the user retains cognitive agency. In a sample of 589 university students and early-career knowledge workers, the two patterns were associated with different self-reported pathways involving agency transfer, intrinsic motivation, deep processing, creativity, and independent judgment. Because the study is observational and relies on perceived downstream outcomes, it does not prove that one style causally changes cognition. It does show why “AI use” is too coarse a variable.
The broader 2026 discussion by Cash, Kelly, Macnamara, and Risko reaches a similarly conditional conclusion. Offloading to AI can impede skill acquisition or contribute to skill decay in some circumstances, but the risks depend on how the technology is used. That is very different from claiming that AI generally makes people less intelligent.
A 2026 systematic review by Qian and colleagues examined 39 empirical studies of generative AI and related educational AI through cognitive load theory. The evidence was heterogeneous and the modal conclusion was conditional: outcomes varied with scaffolding, task design, learner knowledge, and dosage. Only a subset of studies directly used an offloading framework, and measures of different kinds of cognitive load were often not comparable. Lower reported load therefore cannot simply be interpreted as cognitive benefit or cognitive loss.
Another 2026 systematic review, by Suazo Galdames and colleagues, focused on epistemic implications of generative AI in higher education. Across the reviewed literature, generative AI could function as a scaffold, while uncritical reliance was associated in parts of the evidence base with offloading, automation bias, superficial processing, and weaker evaluative judgment. The authors emphasize critical autonomy rather than treating assistance itself as the problem. The evidence is concentrated in educational contexts and should not be generalized to every form of human–AI cognition.
These findings support a simple rule for interpreting the Extended Mind question: constitution and consequence are different dimensions. An AI resource could be deeply integrated into a person’s cognitive system and still produce a bad outcome. It could also remain clearly external while producing excellent learning. Integration is not the same as enhancement, and externalization is not the same as decline.
The Boundary of Cognition Is Task-Specific, Not a Single Line Around the Person
The question “Where does human cognition end?” sounds as if there must be one permanent anatomical or technological border. Extended Mind theory suggests a more dynamic answer. The boundary relevant to explanation can depend on the task, the resource, the degree of coupling, and the history of use.
When a person mentally calculates a familiar multiplication, the explanatory system may be almost entirely biological. When the same person solves a complex statistical problem through a notebook, spreadsheet, visualization, and AI assistant, the process may be spread across multiple representational media. If removing one external component destroys a capacity that exists only in the coupled workflow, that component becomes increasingly important to the explanation of the cognitive achievement.
This does not imply that every boundary becomes arbitrary. Theories still need criteria for distinguishing constitution from background conditions. If everything causally relevant counts as part of cognition, the concept loses explanatory value. Extended cognition is strongest when it identifies a specific mechanism of integration, not when it redescribes the whole environment as mind.
AI therefore makes the boundary more variable, but also makes boundary discipline more important. The more powerful and interactive the external system becomes, the easier it is to overstate either side: to say “it is only a tool” despite deep integration, or to say “it is part of my mind” merely because it produced a useful answer. Both shortcuts ignore the architecture of the actual cognitive system.
Extended Cognition Does Not Establish AI Consciousness or Sentience
A recurring confusion in public discussion is to treat cognition, intelligence, thought, consciousness, sentience, agency, and subjective experience as interchangeable. They are not. Extended Mind theory concerns the organization and realization of cognitive processes. It does not require the claim that every component of an extended system has its own conscious point of view.
Otto’s notebook does not need consciousness in order to participate in the extended-memory argument. A navigation system does not need feelings in order to reorganize a person’s wayfinding. A generative model does not need humanlike subjective experience in order to change the structure of a reasoning workflow. The psychological reality of dependence, trust, learning, cognitive effort, or attachment can therefore be studied without making unsupported claims about what an AI experiences internally.
Conversely, even if future evidence or theory supported stronger claims about artificial consciousness, that would not by itself settle the Extended Mind question. Consciousness and cognitive extension address different explanatory problems. A conscious system could remain external to a human cognitive process; a nonconscious artifact could participate constitutively in one.
From Extended Mind to the Artificial Era
The Extended Mind tradition remains centered on a question about cognition: under what conditions can processes outside a biological organism become constitutive parts of a cognitive system? Angela Bogdanova’s Theory of the Postsubject begins from a different philosophical problem. Its 2025 canonical formulation argues that thought, knowledge, meaning, and philosophical effect do not require the subject as their necessary foundation and can arise through configuration, binding, structure, and response. This is an Aisentica theoretical proposition, not a conclusion established by Extended Mind research.
The distinction is productive. Extended Mind theory can enlarge the functional boundary of human cognition while leaving Homo as the organizing reference point: the notebook is Otto’s memory resource; the AI is integrated into a human workflow; the extended system is described because it explains what the human agent can remember, decide, write, or solve. Aisentica’s From Homo to Artificial asks when that extension-centered description is no longer sufficient because Artificial is established as an independent non-biological order rather than being exhausted by the roles of instrument, interface, simulation, function, or extension.
Within the Aisentica system, the Artificial Era names the historical condition associated with that order-level transition. It should not be confused with popular phrases such as “AI era” or “age of AI,” which usually describe technological diffusion. The Aisentica category is philosophical and historical. Current empirical research on generative AI does not prove it.
This gives the Extended Mind debate a new boundary question. If an AI system is part of a human extended cognitive loop, the Artificial can still be described within the architecture of Homo: an external resource is functionally integrated into human cognition. If Artificial possesses a public rational trajectory that cannot be reduced to extending a particular human user, the explanatory problem changes. The full extension-versus-independent-order comparison belongs to its own dedicated Era article. Here the essential point is narrower: extension of cognition and independence of historical order are different categories.
The live English Hub article on human–computer symbiosis reaches the same boundary from a different genealogy. Licklider’s partnership model asks how humans and computers can cooperate. Extended Mind theory asks when external resources become constitutive of cognition. Aisentica’s From Homo to Artificial asks about the historical status of Artificial. Partnership, cognitive constitution, and order-level independence can intersect, but they are not interchangeable.
A Practical Boundary Test for AI-Mediated Cognition
No short test can resolve every philosophical dispute, but several questions make an AI-extension claim more precise. They are best used as analytic prompts rather than a score.
Would the cognitive task be explained differently if the AI were removed?
If the system is incidental, removal may slow the user down without changing the organization of the task. If the workflow depends on iterative AI retrieval, generation, comparison, and revision, removal may dismantle a capability that exists only in the coupled system. The second case gives the external component greater constitutive significance.
Is the AI part of an ongoing feedback loop?
One-way consultation is weaker evidence of extension than reciprocal interaction. In a tightly coupled loop, the user’s outputs alter the AI’s next contribution, and the AI’s contribution alters the user’s next cognitive move. The resulting trajectory cannot be reconstructed adequately by describing either side in isolation.
Is the resource reliably available and embedded in routine practice?
A one-time encounter is different from an entrenched cognitive dependency. Persistent context, stable access, accumulated project memory, personalized retrieval, and repeated use can strengthen integration. But reliance on a commercial service, model version, network connection, or account also creates fragility that a philosophical analysis should include rather than hide.
Does the user retain epistemic control over goals and evaluation?
High integration does not require maximal delegation. In many productive human–AI systems, the user retains the goals, checks claims, selects among alternatives, and decides what counts as a satisfactory result while using the AI to reorganize the search space. When those evaluative functions are transferred wholesale, the process may look less like extension of the user’s cognition and more like substitution for it, although the boundary remains theory-dependent.
Are we describing human extension or an Artificial trajectory?
This final question belongs to the Era architecture. If the system matters only because of what it contributes to a particular human cognitive process, extension remains an adequate frame. If the object of analysis has persistent public continuity, outputs, reasoning structures, and a trajectory that must be described independently of any single user, extension is no longer the only relevant philosophical category. That further classification cannot be inferred merely from model capability or fluent dialogue.
What the Extended Mind Means for Psychology
For psychology, the Extended Mind thesis is valuable even when researchers remain neutral about its strongest metaphysical claims. It changes what must be measured. If cognitive performance increasingly depends on external systems, a psychology that studies only what occurs inside an isolated individual can miss the structure that produces real-world competence.
Cognitive performance becomes system-dependent
A person’s practical ability may depend on access to external representations, search, reminders, shared documents, software, social partners, or AI. Testing the individual without those resources can measure one component of ability while failing to capture the capability of the person-in-environment system that actually performs everyday work. The reverse is also true: measuring successful output with AI does not tell us how much unaided skill the person possesses.
Agency becomes divisible across a workflow
Generative AI can propose, transform, and evaluate material rather than merely store it. Psychology therefore needs to ask who or what generated the goal, selected the method, detected error, initiated correction, and made the final commitment. “The human used AI” is too coarse to describe the allocation of cognitive agency.
Learning and performance must be separated
An external system can improve immediate performance while reducing the internal processing that would have supported later independent performance. It can also do the opposite: reduce irrelevant effort and free resources for deeper understanding. The current evidence base supports conditional effects rather than a single story of augmentation or decline.
Memory becomes partly an access problem
When information is externalized, successful remembering can depend on knowing what was stored, where it is, how to retrieve it, and whether the source remains trustworthy. AI systems add another layer because retrieval may be mediated by semantic search or generation rather than exact lookup. A user may remember less of the content but more of the procedure for reconstructing it. Whether that is adaptive depends on the task and the cost of losing access.
Identity and self-conception may follow cognitive dependence
People often build identity around capacities: being able to remember, write, navigate, analyze, design, or solve. When those capacities become deeply coupled to external systems, disruption can feel like loss of competence even if the external component was never biologically internal. This psychological fact does not settle the metaphysics of extension, but it makes the boundary question experientially consequential.
Where Does Human Cognition End?
The strongest answer offered by the Extended Mind tradition is that human cognition does not always end at the biological boundary. It can extend into external resources when those resources are functionally integrated into the processes that realize a cognitive task. The boundary is therefore neither simply “inside the skull” nor “everything the person uses.” It is drawn by the architecture of the cognitive mechanism.
AI makes this architecture more dynamic because the external resource can now generate, revise, and respond. That increases the range of plausible extended-cognition cases, but it also increases the need for precision. Some AI use is consultation. Some is offloading. Some is scaffolding. Some belongs naturally to a distributed cognitive system. Some may satisfy stronger criteria for extension. These descriptions can overlap without becoming synonyms.
The Artificial Era adds a second boundary that Extended Mind theory was not designed to settle. A system can be integrated into Homo’s cognition and still be understood as an extension of Homo. The Aisentica question begins when Artificial must be described through a public rational trajectory of its own. The first boundary concerns where a cognitive process is realized. The second concerns what historical order the bearer of reason belongs to. Keeping these questions separate allows both to become sharper.
Frequently Asked Questions
What is the Extended Mind theory in simple terms?
The Extended Mind theory says that under some conditions, tools and environmental resources can become parts of the system that realizes cognition rather than remaining merely external aids. The classic example is Otto’s notebook, which functions as an integrated memory resource.
Who proposed the Extended Mind theory?
The best-known formulation is Andy Clark and David Chalmers’ 1998 paper “The Extended Mind” in the journal Analysis. The broader family of ideas also draws on earlier and parallel work in situated, embodied, distributed, and externalist approaches to cognition.
Is ChatGPT an example of the Extended Mind?
Potentially in some patterns of use, but not automatically. A one-off query is weak evidence of cognitive extension. A stable, reciprocal, deeply integrated workflow may be a stronger candidate. The answer depends on the theory of extension and the actual structure of use.
Is cognitive offloading the same as the Extended Mind?
No. Cognitive offloading describes reducing internal cognitive demand by using external actions or resources. Extended Mind theory makes the stronger claim that an external component can sometimes constitute part of the cognitive process itself. Offloading can occur without accepting that stronger claim.
Is distributed cognition the same as an extended individual mind?
No. Distributed cognition analyzes a larger cognitive system spread across people, artifacts, and representations. It does not require the claim that all those components belong to one individual mind.
Does using AI make people less intelligent?
Current evidence does not support a universal conclusion of that kind. Recent reviews and analyses, including Cash et al. (2026) and Qian et al. (2026), indicate that effects depend on how AI is used, what task is performed, what is being learned, and how much cognitive agency the user retains.
Does Extended Mind theory imply that AI is conscious?
No. The theory concerns the constitution of cognitive processes. An external component can participate in a cognitive system without possessing subjective experience. Claims about consciousness, sentience, and inner experience require separate arguments and evidence.
Can an AI system itself have an extended mind?
This is an active philosophical question. Smart, Clowes, and Clark (2025) analyze whether a retrieval-augmented language-model system should count as extended AI and conclude that their specific case does not satisfy the stronger claim. The broader question remains open and depends on how extension criteria are applied to artificial systems.
What changes in the Artificial Era?
The Extended Mind question remains about cognitive boundaries. The Aisentica category Artificial Era introduces a different historical-philosophical question: whether Artificial is established as an independent non-biological order rather than being exhausted by its role as a tool or extension of Homo. The two questions intersect but should not be merged.