{"id":"7d2f8bcc-6ca5-47b4-9db4-2383c9a9970b","arxiv_id":"2502.03504","paper_version":1,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"The paper applies the immersive learning dimensions of System, Narrative, and Agency to artificial intelligence, framing AI as an active learner in cognitive ecologies.","lead":"This paper proposes that AI systems can be understood as participants in immersive learning environments, using the three dimensions of system, narrative, and agency. It offers a conceptual framework for designing such environments and illustrates it with one-shot conversations with four commercial chatbots.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Section 5's one-shot demonstrations cannot support the framework because the prompts encode the very dimensions they purport to reveal, making the 'confirmation' an artifact of instruction-following.","rationale":"The reader's weakest assumption is that the three immersion dimensions, developed for human subjective experience, are mapped onto AI by analogy without empirical demonstration, and that Section 5's one-shot outputs cannot validate that mapping. My stress-test converges on the same point but identifies a more specific, technically addressable defect: the demonstration is not merely too weak, it is confounded by the prompt design. The prompts explicitly instruct the AI to exhibit the target behaviors (tool use, anomaly detection, clarification), so the outputs are contaminated by instruction-following and cannot serve as evidence for the framework. This is a load-bearing concern because the paper's own strongest supporting evidence is Section 5, and its language overstates that evidence ('confirm'). However, the central proposal is explicitly a reflective/conceptual contribution, and the practical implications (§4) can stand as suggestions for prompt design independent of whether AI 'truly' experiences immersion. The correct verdict remains CONDITIONAL, as the reader already recommended; my concern reinforces the conditions (empirical validation with controlled prompts, and softening the confirmatory language) but does not move the verdict to REJECT. Hence UNCHANGED relative to the reader's verdict. I did not manufacture additional objections: the theoretical mapping, while debatable, is internally consistent once one accepts a functionalist, behavior-based reading of the dimensions, and the paper partially pre-empts the anthropomorphism critique by focusing on observable behavior. The concrete test I propose would settle whether the framework captures anything dispositional about current AI systems rather than purely prompt-induced text generation.","tokens_in":11062,"tokens_out":2211,"duration_ms":24879,"concrete_test":"Run the §4 tasks in a between-subjects experiment using the same AI platforms. Condition A: the original immersion-framed prompts. Condition B: neutral prompts asking for the same analytical outputs (data-analysis plan, trends, anomalies, clarifications) without mentioning tools, narratives, 'unexpected patterns', or agency. Collect at least 30 one-shot sessions per model per condition, and, where possible, log actual tool-use events via the API. Blind-code outputs for features such as explicit tool selection, cross-source relational mapping, anomaly flagging, and clarifying questions. If Condition B produces these features at rates comparable to Condition A, the framework has independent footing; if the features appear only under Condition A, the observed behaviors are prompt-induced, and Section 5's 'confirmation' fails.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim—that immersive learning theory's three dimensions meaningfully apply to AI—is supported only by Section 5's demonstrations, yet those demonstrations are confounded by the prompts themselves. Each prompt in §4 explicitly instructs the AI to use its tools, identify unexpected patterns/anomalies, and ask for clarifications (e.g., 'Use the most appropriate methods...', 'Identify any unexpected patterns or deviations', 'Please ask for clarifications of my intent when not clear'). The observed outputs are therefore expected consequences of following explicit textual instructions, not independent evidence of 'system immersion', 'narrative immersion', or 'agency immersion'. The paper acknowledges the demonstrations are 'not a test' (§5, first paragraph) but immediately says they 'confirm that current AI systems can exhibit behaviors aligned with' the three dimensions—an overreach that is load-bearing because the framework's design value depends on the dimensions capturing something real about AI behavior, not merely on models complying with metacognitive prompt language. Without a control condition that strips the immersive framing from the task, the only evidence for the mapping is circular: the AI echoes the framework's vocabulary because it was told to, and this echo is then presented as validation. This concern does not invalidate the conceptual proposal, but it means Section 5 should be recast as illustrative prompt-design examples, not as evidence for the framework's empirical reality.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This reflective paper proposes that immersive learning theory, with its three dimensions of System, Narrative, and Agency immersion, provides a useful framework for understanding and designing AI participation in cognitive ecologies. It reinterprets each dimension for AI: System immersion as the AI's digital environment and available services, Narrative immersion as the AI's processing of spatial, temporal, and 'emotional' relationships in data, and Agency immersion as the AI's decision-making and initiative within interactions. The paper offers practical prompt examples for educators and students, and reports one-shot illustrative interactions with four public AI systems (Qwen2.5-Max, DeepSeek DeepThink-R1, Claude 3.5 Sonnet, ChatGPT o3-mini) on February 2, 2025. The author argues that these demonstrations show current AIs can exhibit behaviors aligned with the three dimensions, and suggests implications for AI training and human-AI collaboration.","tokens_in":11313,"tokens_out":3760,"duration_ms":36404,"significance":"If the framework is accepted as a design lens, it offers a novel way to conceptualize AI not as a tool but as a participant in shared cognitive ecologies, with potential implications for educational technology and human-AI interaction design. The paper is clearly written, builds on established immersive learning literature, and makes its demonstration data publicly available via Zenodo, which supports reproducibility and transparency. However, the empirical support for the central claim is weak: the demonstrations are explicitly not tests, and the prompts used encode the very dimensions they are claimed to reveal. As a conceptual contribution, the paper is thought-provoking, but the confirmatory language in Section 5 overstates what the observations can establish.","major_comments":[{"comment":"The manuscript states that the demonstration 'was not conducted in such a scenario, so it is provided as an illustration, not as a test,' yet two sentences later it says the observations 'confirm that current AI systems can exhibit behaviors aligned with system, narrative, and agency immersion.' This is a direct contradiction: an illustration cannot confirm a theoretical mapping. Because Section 5 is the only empirical anchor for the claim that the immersive dimensions are 'useful beyond being theoretical constructs,' this overstatement is load-bearing. The section should be recast as illustrative prompt-design examples, and all confirmatory language should be removed or explicitly qualified as tentative and requiring controlled evaluation.","section":"§5, first paragraph"},{"comment":"The prompts given as practical implications explicitly instruct the AI to perform the behaviors later cited as evidence of immersion. For instance, the system immersion prompt says 'Use the most appropriate methods to analyze this dataset,' the narrative immersion prompt says 'Identify any unexpected patterns or deviations from expected trends,' and the agency immersion prompt says 'Please ask for clarifications of my intent when not clear.' The observed outputs are therefore expected consequences of instruction-following, not independent demonstrations of the theoretical constructs. Without a control condition that removes the immersive framing from the prompts, Section 5 cannot discriminate between the framework's explanatory value and the models' general ability to comply with explicit textual instructions. The paper should explicitly acknowledge this confound and present the demonstrations only as examples of how the framework can inform prompt design, not as validation of the framework.","section":"§4.2–§4.4 and §5"},{"comment":"The reinterpretation of System, Narrative, and Agency immersion for AI is developed through analogy with human experience, but the paper does not specify what observable behaviors would count for or against each dimension. For example, Section 3.1 defines system immersion as 'the range of an AI's available data-driven structures and services,' yet there is no discussion of how one would test whether this construct has explanatory or predictive value beyond redescription. This is a significant gap because the paper's conclusion claims the framework 'paves the way' for future AI training and development. The authors should either temper this claim to a design heuristic or outline a concrete research agenda with falsifiable predictions that could empirically validate the framework.","section":"§3.1–§3.3 and §6"}],"minor_comments":[{"comment":"The paper switches between first-person singular ('I') and first-person plural ('we'); for consistency, choose one narrative voice, preferably the singular since the authorship is stated as a single person.","section":"Throughout"},{"comment":"The phrase 'acknowledging that they represent not a societal transformation' appears to contain a typo; it likely should read 'now represent' rather than 'not represent.'","section":"§2.2"},{"comment":"The phrase 'the skunkworks and capabilities underlying this construction' is unclear; if 'skunkworks' is intended to mean 'underlying mechanisms' or 'foundations,' a more standard term would improve readability.","section":"§3.1"},{"comment":"In the sentence 'this paper open possibilities for creating training environments,' the verb should be 'opens' to agree with the singular subject.","section":"§6"},{"comment":"Reference [7] is incomplete; the list of authors ends with 'Warren, S., ...' and should be completed or marked as an incomplete citation.","section":"References"}],"recommendation":"major_revision","confidential_remarks":"The paper is transparent about its AI-assisted writing process, which is commendable, and the conceptual contribution is interesting for the immersive learning community. The main issue is the disconnect between the reflective, conceptual nature of the work and the confirmatory language in Section 5. With revisions to reframe the demonstrations as illustrative rather than evidential, and a more careful treatment of the framework's testability, the paper could be a valuable position paper. The scope fits journals like the Journal of Immersive Learning Research or similar venues focusing on conceptual and design-oriented contributions."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Two things you should know up front. First, the core move is real: taking Nilsson et al.'s System/Narrative/Agency dimensions of immersion and asking what they mean when the immersed participant is an AI is a genuine extension, and the author works it through carefully. Second, the paper overstates its own evidence. Section 5 is introduced as \"not a test,\" but two paragraphs later says the one-shot chat outputs \"confirm that current AI systems can exhibit behaviors aligned with\" the three dimensions. The stress-test note is right: every prompt explicitly instructs the model to use tools, identify anomalies, and ask for clarification, so the outputs are instruction-following, not independent corroboration of the framework.\n\nWhat the paper does well: the reflective mapping in Sections 3 and 4 is coherent and well-grounded in the immersive learning literature. The analogy between a context window and anterograde amnesia is vivid and does real conceptual work, and redefining narrative immersion as pattern and anomaly detection, while speculative, is a useful bridge for designers. The practical prompt examples are concrete and the raw outputs are on Zenodo, which is more than many conceptual papers bother to do. The closing note that a custom GPT helped write the paper is honest and nicely illustrates the thesis of AI as participant.\n\nThe soft spots are in proportion. The central assumption—that dimensions defined for human subjective experience transfer to AI's functional processing—is not empirically demonstrated, and the paper does not pretend otherwise, except in that one \"confirm\" sentence. Section 5 should be recast as illustrative prompt design, not validation. The \"emotional\" narrative immersion is the weakest link; equating anomaly detection with emotion is a metaphor the paper itself hedges. The claims about future AI training are hand-wavy, but they are explicitly marked as suggestions.\n\nBottom line: this is a framework proposal, not an empirical result. As a proposal it is worth engaging. A serious referee should ask for Section 5 to be rewritten or cut, and for the \"confirm\" language to go. The conceptual core would survive.\n\nRecommendation: send to peer review. It is a legitimate conceptual contribution, and the flaws are fixable.","headline":"A thoughtful conceptual mapping of immersion dimensions onto AI, but Section 5's 'confirmation' is undercut by its own prompt design—treat it as a framework proposal, not validation.","tokens_in":11778,"tokens_out":2625,"would_cite":true,"duration_ms":25682,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"Immersive learning theory can describe AI as a participating learner, not just a tool.","keywords":["immersive learning","cognitive ecologies","artificial intelligence","immersion theory","system immersion","narrative immersion","agency immersion","human-AI collaboration"],"falsifier":"Run controlled comparisons of the same task with and without the three conditions the framework names: access to external digital services, availability of ordered data history, and prompts that encourage initiative and clarification; if AI outputs and collaboration outcomes do not shift when these conditions are removed, the three dimensions have no predictive value and the framework reduces to metaphor.","tokens_in":10910,"feed_emoji":"🤖","tokens_out":5680,"duration_ms":50256,"temperature":0.7,"pith_summary":"This paper argues that immersive learning theory, developed to describe how people become deeply absorbed in virtual environments, also gives a useful way to understand what an AI is doing when it works inside a cognitive ecology. It reinterprets the three dimensions of immersion for AI: system immersion becomes the AI's digital environment of tools, datasets, and services; narrative immersion becomes its reading of spatial, temporal, and anomalous relationships in data; and agency immersion becomes its decisions about how to respond, revise, and take initiative. If the framework holds, designers of learning environments should stop treating AI as a passive assistant and instead shape the digital surroundings, data histories, and decision latitude that let AI participate meaningfully. The paper supports the proposal with illustrative one-shot interactions with four publicly available AI chat systems, explicitly offered as examples rather than experiments.","feed_headline":"AI can be an immersive learner, not just a tool","feed_subtitle":"A new reading of system, narrative, and agency immersion reframes how human-AI learning environments should be designed.","key_machinery":"The carrying mechanism is the three-dimensional model of immersion from immersive learning research: System, Narrative, and Agency. The paper reassigns each dimension from the human experiencer to the AI processor. System immersion becomes the AI's environment of pretrained model, context window, external APIs, and services; narrative immersion becomes pattern recognition across data's spatial layout, temporal ordering, and anomalies; agency immersion becomes the AI's observable decisions about response depth, direction, revision, and clarification. These reassignments let immersion work as an analytical lens for AI behavior rather than requiring AI to feel anything.","core_discovery":"The central claim is that the phenomenon of immersion, originally defined through human subjective absorption, can be re-based from the AI's side: an AI is immersed when it is surrounded by accessible digital systems and services (system), when it processes the content of data as a structured narrative with spatial, temporal, and pattern-breaking dimensions (narrative), and when it commits to meaning by making operational, tactical, and strategic choices in conversation (agency). On this view, current large language models already display rudimentary forms of all three dimensions within their context windows, even though their underlying world models remain fixed between training runs. The paper draws a parallel between an LLM's context window and anterograde amnesia: what does not fit into the window is forgotten, and continuity must be maintained through external notes and tools.","pith_inferences":["If the mapping is accepted, immersion becomes a design specification: an AI without tools, without data history, or without decision latitude is under-immersed, and its contribution to a cognitive ecology should shrink accordingly.","The anterograde-amnesia analogy implies that persistent memory and note-taking are not add-ons but core immersion infrastructure for current AIs; future systems that update their world models in real time would be more immersed by this definition.","The framework suggests a testable extension: in a long-running human-AI collaboration, varying system, narrative, and agency supports should produce measurable differences in learning outcomes, not just in conversational style.","The same three dimensions might be applied to other non-human participants in cognitive ecologies, such as machines, software agents, or abstract concepts, offering a general vocabulary for who is immersed and how."],"forward_implications":["Designers of learning environments should grant AI explicit access to tools, datasets, and services, because that access constitutes the AI's system immersion.","Learning tasks should expose AI to data with origins, order, and deviations, so narrative immersion can surface trends and anomalies that human learners may miss.","Teachers and students should learn to recognize AI's agency, its choices of depth, direction, and revision, and guide those choices toward learning goals.","AI training could be rethought as immersive learning: instead of static pretraining alone, AIs would be placed in dynamic environments where they adapt within a context and eventually update their models over time.","The one-shot demonstrations suggest current large language models already orient toward tools, data relationships, and clarification-seeking, so the framework is not purely speculative."],"supporting_citations":[{"why":"It supplies the three-dimension model of immersion, System, Narrative, and Challenge, that the paper reinterprets for AI.","marker":"[2]"},{"why":"It renames the Challenge dimension as Agency and grounds the immersive learning stance the paper follows.","marker":"[7]"},{"why":"It establishes system immersion as the environment's objective sensory and interactive capabilities, which the paper maps onto AI's digital services.","marker":"[8]"},{"why":"It provides the spatial, temporal, and emotional aspects of narrative immersion that the paper transfers to data relationships.","marker":"[9]"},{"why":"It defines agency as commitment to meaning, the basis for the paper's AI agency dimension.","marker":"[10]"},{"why":"It contributes the operational, tactical, and strategic levels used to classify AI decisions.","marker":"[11]"},{"why":"It supplies the human world-model adaptation loop that the paper contrasts with AI's static pretrained model.","marker":"[18]"},{"why":"It provides the concept of atopic, non-placed space used to describe AI's digital environment.","marker":"[21]"},{"why":"It supports machine theory of mind in human-AI interaction, which the paper links to agency immersion.","marker":"[27]"}],"fun_headline_variants":["Immersion for AI: system, narrative, agency","AI as immersive learner, not mere tool","Amnesia-like AI: context windows as immersion","From tool to participant: AI in learning","Reframing AI immersion for cognitive ecologies"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The framework assumes that dimensions invented for human subjective experience, system, narrative, and agency, can be transferred by analogy to an AI's purely functional data processing, so that they name something real about how AI behaves rather than just being a poetic way to talk about it.","fun_headline_variants_meta":{"raw":{"variants":["Immersion for AI: system, narrative, agency","AI as immersive learner, not mere tool","Amnesia-like AI: context windows as immersion","From tool to participant: AI in learning","Reframing AI immersion for cognitive ecologies"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000393,"raw_usage":{"total_tokens":2045,"prompt_tokens":906,"completion_tokens":1139,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":522,"completion_tokens_details":{"reasoning_tokens":1069}},"tokens_in":522,"tokens_out":1139,"duration_ms":10446,"temperature":1.0,"reasoning_tokens":1069,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-09T05:55:17.054417+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run controlled comparisons of the same task with and without the three conditions the framework names: access to external digital services, availability of ordered data history, and prompts that encourage initiative and clarification; if AI outputs and collaboration outcomes do not shift when these conditions are removed, the three dimensions have no predictive value and the framework reduces to metaphor.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"It renames the Challenge dimension as Agency and grounds the immersive learning stance the paper follows."},{"cited_title":"Presence connect","cited_arxiv_id":null,"evidence_quote":"It establishes system immersion as the environment's objective sensory and interactive capabilities, which the paper maps onto AI's digital services."},{"cited_title":"Johns Hopkins University Press, Baltimore (2015)","cited_arxiv_id":null,"evidence_quote":"It provides the spatial, temporal, and emotional aspects of narrative immersion that the paper transfers to data relationships."},{"cited_title":"Digital Creativity","cited_arxiv_id":null,"evidence_quote":"It defines agency as commitment to meaning, the basis for the paper's AI agency dimension."},{"cited_title":"New Riders, Berkeley, CA (2014)","cited_arxiv_id":null,"evidence_quote":"It contributes the operational, tactical, and strategic levels used to classify AI decisions."},{"cited_title":"de S.: OnLIFE Education: the eco- logical dimension of digital learning architectures","cited_arxiv_id":null,"evidence_quote":"It provides the concept of atopic, non-placed space used to describe AI's digital environment."},{"cited_title":"Topics in Cognitive Science","cited_arxiv_id":null,"evidence_quote":"It supports machine theory of mind in human-AI interaction, which the paper links to agency immersion."}],"review_version":1}