{"id":"685cc86a-940e-4c54-8499-20505e3098f6","arxiv_id":"2412.12681","paper_version":1,"verdict":"UNVERDICTED","confidence":"HIGH","novelty_score":2.0,"correctness_risk":"unknown","formal_verification":"none","parameter_count":0,"one_line_summary":"This is a call for participation for a workshop on AI-enabled everyday AR, not a research contribution.","lead":"This paper proposes a one-day CHI 2025 workshop on everyday augmented reality (AR) powered by AI. It outlines the workshop's vision, topics, organizers, and planned activities, but presents no new scientific results.","discovery_kind":"unclear","skeptic_critique":{"model":"deepseek-v4-flash","headline":"No significant objection identified; the paper is a workshop call for participation and makes no verifiable technical claim that can be stress-tested.","rationale":"The reader correctly identifies that this is a workshop call for participation rather than a research preprint and correctly marks it UNVERDICTED. The strongest claim in the paper, like most claims in workshop CFPs, is prospective and rhetorical: 'we believe that everyday AR is increasingly feasible.' No evidence is provided, but none is owed in this genre. The reader's weakest assumption—that AI can be integrated into AR for robust, context-aware adaptation—is a real open challenge in the field, but it is not load-bearing because the paper never advances a falsifiable position that depends on that assumption being true. The workshop exists precisely to debate whether and how such integration might be achieved. Therefore, the appropriate stress-test outcome is that no significant objection to the central claim can be raised on technical grounds. I partially agree with the reader because I concur that the document makes no empirical claims, but I do not treat the AI-integration assumption as a weakness of the argument, since the argument does not try to establish feasibility. No change to the reader's verdict is needed.","tokens_in":7911,"tokens_out":1520,"duration_ms":16293,"concrete_test":"Verify that the arXiv document is a workshop call for participation by checking the abstract, the CCS concepts, the ACM Reference Format, and Section 4 ('Call for Participation'), and confirm that the text makes no empirical or theoretical claims requiring falsification. If such a claim is found in a revised version, reassess the verdict accordingly.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's central assertion, that everyday AR is becoming increasingly feasible through AI-in-the-loop approaches, functions as a vision statement rather than a testable hypothesis. It does not specify a mechanism, report an experiment, or derive a formal result. The reader's identified weakest assumption—that recent AI advances can be integrated into AR to provide robust, context-aware adaptation—is genuine, but it is not load-bearing for this document. A workshop call for participation does not require that assumption to be true to satisfy its stated purpose of soliciting discussion, position papers, and community building. The text itself acknowledges open challenges such as privacy, explainability, and sustainability, and the proposed workshop is explicitly designed to address them. There is no internal inconsistency: the document claims only that the vision is worth discussing, not that it has been established. References to prior work and a previous workshop provide context but are not used as evidence for a specific effect size or performance guarantee. Consequently, applying standard research-assessment criteria to this document is category error, and no technical objection can meaningfully attach to its central claim.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"This manuscript is a workshop proposal for CHI EA 2025, authored by Suzuki, Gonzalez-Franco, Sra, and Lindlbauer. It argues that recent advances in AR hardware and AI/ML make 'everyday AR'—always-available, seamlessly integrated augmented reality—increasingly feasible, and that achieving it requires an AI-in-the-loop approach in which digital content and interactions continuously anticipate and adapt to user context. The paper identifies six topics of interest (adaptive and context-aware AR, LLM-powered always-on assistants, AI-assisted task guidance, generative AR content creation, accessible AR design, and real-world-oriented AI agents), describes the planned one-day workshop structure and activities, provides a call for participation, and includes organizer biographies and references. The manuscript contains no experiments, derivations, or datasets; its central feasibility claims are explicitly framed as beliefs ('we believe').","tokens_in":8177,"tokens_out":5171,"duration_ms":47579,"significance":"As a workshop proposal, this paper serves a community-building and agenda-setting purpose rather than making a technical contribution. If the vision of everyday AR were realized, it could indeed constitute a major shift in human-computer interaction, and the paper usefully organizes current research threads such as context-aware AR, generative content creation, and always-on AI assistance. It also explicitly acknowledges open challenges including accessibility, privacy, and explainability, and it draws on the organizers' prior workshop experience at UIST 2023 [24], which lends credibility to the organizational plan. However, the manuscript makes no testable claims and provides no evidence for feasibility; its value depends on the workshop's ability to generate and refine community discourse. For a reader expecting a standard research paper, the contribution is limited; for a workshop proposal, it is appropriate. The stress-test concern about whether AI can handle real-world unpredictability is genuine, but it is not load-bearing for this document because the paper's stated purpose is to invite discussion of that very question.","major_comments":[],"minor_comments":[{"comment":"The phrase 'make it possible to results in a paradigm shift' is ungrammatical; rewrite it as 'make possible a paradigm shift' or 'result in a paradigm shift.'","section":"Section 1, first paragraph"},{"comment":"The phrase 'users every-changing needs' contains a typo; it should be 'users' ever-changing needs.'","section":"Section 1, first paragraph"},{"comment":"The text reads 'participant's lighting talks'; this should be 'participants' lightning talks.'","section":"Section 3.2, Introductions and Lightning Talks"},{"comment":"The caption says 'Around 50 participants,' but Section 3.2 reports 40 participants for the same UIST 2023 workshop; please reconcile these numbers.","section":"Figure 1 caption"},{"comment":"The phrase 'assigned a ‘table. ’' contains stray quotation marks; it should simply read 'assigned a table.'","section":"Section 3.2, Theme Organization and Discussion"}],"recommendation":"minor_revision","confidential_remarks":"This manuscript is a workshop call rather than a technical contribution, and I have evaluated it on that basis. If the venue's acceptance criteria require new technical results, the manuscript would be outside scope; if the venue publishes workshop proposals, it is suitable after the minor fixes listed above. The inconsistent participant counts should be corrected before publication."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nThis is the 'Everyday AR through AI-in-the-Loop' workshop proposal for CHI 2025, not a research preprint. The reader's unverdict is correct, and the stress-test note is also right: applying standard research criteria here is a category error. This is a call for participation, a community-building document, and the second iteration of a similar UIST 2023 workshop.\n\nWhat it does well: it lays out a sensible agenda for AI+AR, touching adaptive and context-aware AR, LLM-powered assistants, generative content creation, accessibility, and real-world agents. The references are current and relevant, and the workshop format is thoughtfully designed with hackathons, design sprints, and mind-mapping activities. It also lists open challenges like privacy, explainability, and sustainability, which shows honesty about the difficulties. As a CFP, it does its job.\n\nThe soft spots are what you'd expect: no empirical evidence, no testable claims, and the central feasibility assertion is explicitly a belief ('we believe that everyday AR is increasingly feasible'). That's fine for a workshop. The only mild criticism I'd offer is that the assumption that recent AI advances can deliver robust, always-on adaptation isn't critically examined, but that's typical for a vision piece and not a load-bearing flaw.\n\nMy verdict: if you're in HCI or AR research, it's worth a skim to see where the community is heading, and it could generate discussion at a reading group. But it doesn't deserve formal peer review as a research paper. A serious editor should desk reject it if it lands on a research track; accept it if it's a workshop proposal. Don't send it to reviewers.\n\nRecommendation: let it be.","headline":"A well-organized workshop call for participation, not a research paper; fine as a vision statement but nothing to referee.","tokens_in":8459,"tokens_out":3550,"would_cite":false,"duration_ms":29264,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":false},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"Everyday AR needs AI in the loop to become feasible","keywords":["Augmented Reality","Mixed Reality","Generative AI","Large Language Models","Human-AI Interaction","Context-aware AR","AI-in-the-loop","Everyday AR"],"falsifier":"A controlled field deployment would settle the claim: give one group an always-on, context-adapting AR assistant and another group a static AR interface for the same everyday tasks; if the adaptive system shows no improvement in completion time, error rate, or perceived load, or if participants disable it because it misreads their context, the premise that AI-in-the-loop is the route to everyday AR is not supported.","tokens_in":7752,"feed_emoji":"🥽","tokens_out":9431,"duration_ms":74354,"temperature":0.7,"pith_summary":"This workshop paper argues that advances in AR hardware and in AI, especially large language models and generative models, make everyday augmented reality increasingly feasible: AR that is always available and seamlessly integrated into daily life, potentially as a general-purpose computing platform. The paper's core proposal is an AI-in-the-loop approach, in which the AR system continuously senses a user's context and adapts content and interactions in real time. A sympathetic reader would care because this reframes AR from a set of specialised productivity or maintenance applications into an infrastructure for daily life, and it gives AI a spatially grounded role: not just answering questions but deciding what the user sees and can do. The paper lays out six research areas that would need to develop together for this vision to hold, from context-aware adaptation to generative content creation to accessible design.","feed_headline":"Everyday AR needs AI in the loop to become feasible","feed_subtitle":"The paper argues that always-on, context-aware AI can make AR a daily computing platform rather than a niche tool.","key_machinery":"The central object is the concept of everyday AR, defined as AR that is always available and integrated into users' daily environments. The mechanism that carries the argument is the AI-in-the-loop model: an AR system in which an AI component continuously senses context, including room geometry, object affordances, user activity, and user state, and uses that understanding to adapt content and interactions without explicit user commands. This mechanism is what the paper says will move AR beyond monolithic applications, because it makes both scene understanding and content generation dynamic. The paper identifies six research thrusts that instantiate the mechanism: adaptive and context-aware AR, LLM-powered always-on assistants, AI-assisted task guidance, generative on-demand content creation, AI-driven accessible design, and real-world-oriented AI agents.","core_discovery":"The paper's central claim is that the combination of recent AI advances and maturing AR hardware enables a new class of experience the authors call everyday AR: always-available, seamlessly integrated digital content that could replace or augment smartphones and desktop computers for many interactions. To get there, the paper asserts, AR must adopt an AI-in-the-loop model, in which digital interactions and content continuously anticipate and adapt to users' changing needs and context. In this model, scene understanding and content generation both become dynamic: large language models support natural, always-on assistant interactions, generative models create on-demand content in real time, and context-aware systems adapt interfaces based on room geometry, object affordances, and user activity. The paper's contribution is a shared research agenda and a call for the community to define the requirements and limitations of this vision rather than a completed system or evaluation.","pith_inferences":["[Editorial inference] If the AI-in-the-loop hypothesis is correct, evaluation of AR systems will need to move from lab-based rendering-quality measures to longitudinal, in-situ measures of context recognition and task benefit, because the central claim is about everyday use.","[Editorial inference] The vision shifts control from the user to the system: users would delegate some interface decisions to AI, which makes trust, privacy, and the ability to override the AI central design problems rather than peripheral ones.","[Editorial inference] A concrete testable consequence is that adaptive AR interfaces using live context models should outperform static, manually configured interfaces on real-world everyday tasks; a field deployment comparing the two would provide evidence for or against the paper's premise.","[Editorial inference] The paper's emphasis on accessible design suggests AI-driven adaptation could make AR more equitable, but that holds only if the underlying context models do not carry the same biases found in their training data."],"forward_implications":["AR would shift from niche applications such as productivity and maintenance to a general-purpose platform for everyday computing, potentially displacing smartphone and desktop interaction for many tasks.","Always-on LLM-based assistants embedded in AR could support users implicitly, understanding context from gaze, gestures, and activities rather than requiring typed questions.","Generative AI could produce interactive AR content on demand, including 3D objects, scenes, and code-generated interactions, removing the need for manual content authoring.","Context-aware adaptation informed by room geometry, object affordances, and user activity would keep augmented interfaces from overwhelming users with irrelevant content.","AI-driven input modalities such as voice, gaze, and facial expression recognition could open AR authoring and experiences to people with motor impairments, who are currently excluded by keyboard-mouse and controller interactions."],"supporting_citations":[{"why":"Supplies the large-language-model capability the paper invokes to argue that natural interaction with virtual agents is now feasible.","marker":"[5]"},{"why":"Establishes the generative-model foundation the paper relies on for on-demand creation of multimodal AR content.","marker":"[13]"},{"why":"Defines context-awareness in pervasive AR, grounding the paper's claim that adaptation to user needs and environment is the path to everyday AR.","marker":"[14]"},{"why":"Provides a computational method for online adaptation of mixed reality interfaces that the proposed AI-in-the-loop model extends.","marker":"[23]"},{"why":"Demonstrates making physical objects interactable in AR, supporting the claim that AI can bridge digital content and the real world.","marker":"[9]"},{"why":"Shows an everyday procedural assistant on a wearable device, supporting the feasibility of always-on AI assistance.","marker":"[3]"}],"fun_headline_variants":["AI in the loop: the path to everyday AR","Everyday AR depends on always-on AI","Context-aware AI makes AR a daily tool","AI-in-the-loop is the key to everyday AR"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The whole vision rests on the assumption that current AI, especially large language models and generative models, can be integrated into AR systems well enough to understand and respond to unpredictable real-world contexts in real time, and that users will accept the always-on sensing this requires.","fun_headline_variants_meta":{"raw":{"variants":["AI in the loop: the path to everyday AR","Everyday AR depends on always-on AI","Context-aware AI makes AR a daily tool","AI-in-the-loop is the key to everyday AR"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000227,"raw_usage":{"total_tokens":1423,"prompt_tokens":851,"completion_tokens":572,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":467,"completion_tokens_details":{"reasoning_tokens":513}},"tokens_in":467,"tokens_out":572,"duration_ms":4689,"temperature":1.0,"reasoning_tokens":513,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-11T13:49:20.860100+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"A controlled field deployment would settle the claim: give one group an always-on, context-adapting AR assistant and another group a static AR interface for the same everyday tasks; if the adaptive system shows no improvement in completion time, error rate, or perceived load, or if participants disable it because it misreads their context, the premise that AI-in-the-loop is the route to everyday AR is not supported.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Establishes the generative-model foundation the paper relies on for on-demand creation of multimodal AR content."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Defines context-awareness in pervasive AR, grounding the paper's claim that adaptation to user needs and environment is the path to everyday AR."},{"cited_title":"PrISM-Observer: Intervention Agent to Help Users Perform Everyday Procedures Sensed using a Smartwatch","cited_arxiv_id":"2407.16785","evidence_quote":"Shows an everyday procedural assistant on a wearable device, supporting the feasibility of always-on AI assistance."}],"review_version":1}