{"id":"fc005e54-8573-4061-92dc-8bfc94ef961a","arxiv_id":"2502.04403","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":4.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"Agency is frame-dependent: whether a system has a boundary, goals, self-causation, and adaptivity depends on arbitrary commitments by the observer.","lead":"The paper argues that whether a system counts as an agent depends on the observer's chosen reference frame, so agency is not an intrinsic property of the system. For AI safety and reinforcement learning, this means questions about whether a system is an agent cannot be answered without specifying the frame being used.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Per-property frame-dependence does not entail frame-dependence of the conjunction; the paper's informal 'reference frame' risks making the central claim vacuous rather than demonstrated.","rationale":"The reader's weakest assumption was the unformalized notion of an 'agent reference frame,' and I agree that this is a central weakness. My stress test sharpens this into two distinct issues. First, even if each property is frame-dependent in isolation, the paper does not show that one can fix a single frame that simultaneously flips all four properties, so the conjunction 'agency' may fail to be frame-dependent in the relevant sense. Second, the paper's own definition of a reference frame as the collection of commitments needed to determine agency makes the conclusion risk vacuity: if a frame is whatever is needed to measure agency, then 'agency is measured relative to a frame' is true by definition. The non-vacuous reading requires an existence proof of divergent complete frames, and that proof is absent. The manuscript's Discussion explicitly acknowledges this gap ('We stop short of presenting a rigorous mathematical definition of reference frames, as well as a formal proof'), which I treat as an in-scope limitation rather than as a pipeline artifact. Because the reader's verdict is already CONDITIONAL, my concern does not move the verdict; it does, however, clarify the specific condition that should be attached: the paper should either supply a formalization in which complete frames can disagree, or soften the central claim to a conditional philosophical thesis. I am not accusing the authors of overclaiming in bad faith; the paper is transparent about its level of rigor, and the examples are plausible. But the central claim as stated goes beyond what the argument currently proves.","tokens_in":5694,"tokens_out":4421,"duration_ms":48162,"concrete_test":"Formalize a frame for the thermostat example as a tuple (B, V, R, C), where B is a boundary, V a set of causal variables, R a goal criterion (e.g., a preference order over temperature outcomes), and C a class of behavior changes counted as adaptive. Enumerate a finite family of such frames and search for a physical system and two complete frames (B, V, R, C) and (B', V', R', C') that yield opposite agency verdicts while both frames satisfy the same legitimacy constraints, such as predictive accuracy on observed behavior. If no such pair exists in this family, the central existential claim fails for the paper's own leading example; if one exists, the claim is at least non-vacuously instantiated and the burden shifts to justifying frame-selection principles.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The argument's load-bearing step is the inference from 'each of the four properties of agency is frame-dependent' to 'agency is frame-dependent.' Agency is presented as the conjunction of these properties, but the examples in Section 2 use different kinds of frames for different properties: individuality uses boundary choices, source of action uses choices of causal variables, normativity uses goal-attribution principles, and adaptivity uses a reference class of behavior changes. From the existence of frames f_i and g_i that flip property i individually, it does not follow that there is a single pair of complete frames f and g, each containing all four commitments, that flips the conjunction. The paper states that a frame 'must include' all four commitments, but never proves such frames can be simultaneously combined to yield opposite agency verdicts for one system. Additionally, the definition of an 'agent reference frame' as 'a collection of these four commitments that allow us to determine whether a system has each of the four properties' makes frame-dependence nearly analytic: any determination of agency requires such commitments. The substantive, non-vacuous claim must be that there exist systems for which different legitimate complete frames give different agency verdicts, with no principled way to select among them. That existential claim is not established, and the manuscript explicitly acknowledges stopping short of a formal definition and proof. This is an internal logical gap between the property-wise arguments and the universal conclusion, not a disagreement with external consensus.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper argues that agency is frame-dependent: any attribution of agency to a system is relative to a 'reference frame,' defined as a collection of commitments about boundaries, causal variables, goal-attribution principles, and reference classes for adaptation. It builds on a four-part account of agency from Barandiaran et al. (2009) and claims that each of the four properties is frame-dependent, citing prior results (Jiang 2019; Harutyunyan 2020; Kenton et al. 2023; Abel et al. 2023; Zadeh 1963). It concludes that agency itself is frame-dependent and discusses implications for RL. The paper explicitly stops short of formalizing reference frames or proving the main claim.","tokens_in":5955,"tokens_out":5246,"duration_ms":50341,"significance":"The paper is an original philosophical contribution that connects distinct strands of RL research to a long-standing question about agency. The individual claims are each grounded in published results, and the paper is honest about its limitations. If the central claim were established, it would have real consequences: agency would not be an intrinsic property of a system, and quantitative agency measurements would need to be indexed to frames. The novelty lies in the conjunction, but that conjunction is not currently proved; the paper's value at present is primarily as a research agenda rather than as an established result.","major_comments":[{"comment":"The inference from per-property frame-dependence to frame-dependence of the conjunction is invalid as stated. For each property i the paper cites examples of frames f_i and g_i that flip property i in isolation, but it never shows that there exists a single pair of complete frames f and g—each containing a boundary, a causal-variable choice, a goal-attribution principle, and a reference class—such that under f the system satisfies all four properties and under g it fails to satisfy one or more. The examples use qualitatively different kinds of commitments for different properties: boundaries for individuality, causal variables for source of action, goal-attribution principles for normativity, and reference classes of behavior changes for adaptivity. The paper asserts that a frame 'must include' all four commitments, but this is the very statement that needs proof. Section 3 acknowledges that the formal proof is missing; this is not a peripheral gap but the load-bearing step of the argument.","section":"Section 2 (summary paragraph)"},{"comment":"The definition of an 'agent reference frame' as 'a collection of these four commitments that allow us to determine whether a system has each of the four properties' makes the conclusion 'agency is frame-dependent' close to analytic: if any determination of agency requires such commitments, then agency is trivially relative to them. The nontrivial, falsifiable claim must be that there exist systems for which two different legitimate complete frames yield opposite agency verdicts, and that no principled criterion selects between the frames. The paper does not provide such an example, and it never defines what makes a frame 'valid' (it mentions 'many valid ways' but not the validity conditions). Without this, the reader cannot distinguish the intended substantive relativity from the trivial observation that all measurement requires a coordinate choice.","section":"Section 2, 'What is a Reference Frame?'"},{"comment":"The four claims are all 'adapted from' prior work, but the adaptations are not stated precisely. In particular, Claim 2 (adapted from Kenton et al., 2023) asserts that two choices of causal variables can identify or refute 'an agent,' yet Kenton et al.'s notion of agent-discovery may not coincide with the four-part Barandiaran et al. notion used here. Similarly, Claim 4 cites 'Theorem 3.1 of Abel et al., 2023' without stating the theorem or its conditions. Because the overall argument depends on these adaptations, the paper should state the original results and explicitly verify that they imply the frame-dependence claims in the sense required here.","section":"Section 2, Claims 1–4"},{"comment":"The phrase 'the logical conjunction of these latter three properties conditioned on the choice of a boundary' is confusing. Earlier the paper lists four properties; if individuality is treated as a precondition that selects the boundary rather than as a property, this should be stated. More importantly, this sentence does not help with the composition problem: even if agency is a conjunction, the existence of separate flipping frames for each conjunct does not guarantee a single flipping frame for the conjunction. The paper also says at one point that the choice of reference frame is 'arbitrary' and at another that there are 'many valid ways' to formalize the components; these two statements are in tension and the paper never clarifies which one is intended.","section":"Section 2, 'In summary' paragraph"}],"minor_comments":[{"comment":"The word 'mostuseful' appears without a space in the paragraph on choosing a reference frame; it should read 'most useful.'","section":"Section 3"},{"comment":"The Ziebart et al. reference contains a typo: 'Artificiall Intelligence' should be 'Artificial Intelligence.' The Abel et al. reference also has an odd spacing in 'V . Roy.'","section":"Reference list"},{"comment":"The term 'frame-dependence' is used without a formal definition. A definition such as 'A property P is frame-dependent if there exist a system S and two frames f and g such that P(S) is true under f and false under g' would clarify what is being claimed and what would count as a counterexample.","section":"Throughout"}],"recommendation":"major_revision","confidential_remarks":"The paper's main novelty over the authors' prior work is the claim that agency as a conjunction is frame-dependent; the composition argument is missing. I would encourage the authors to either provide a worked example of two complete frames yielding opposite verdicts, or to reframe the paper as a position paper that explicitly identifies the open problem. The current framing overstates what has been shown. If a formal construction is outside the paper's scope, consider adjusting the title and abstract to 'Agency is Frame-Dependent: A Research Agenda' or similar."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Dear [Colleague],\n\nHere's my read of Abel et al., \"Agency Is Frame-Dependent.\" The paper is a position piece arguing that agency is not intrinsic but relative to a \"reference frame\" — a bundle of choices about boundaries, causal variables, goal-attribution principles, and what counts as adaptation. What's genuinely new here is the packaging: it brings together results already in the literature (Jiang on boundary dependence, Kenton on causal variables, Zadeh on adaptivity, Dennett on the intentional stance) and shows they point in the same direction. That's a real service. The writing is clear, and the discussion of what frame-dependence would mean for RL is thoughtful.\n\nThe soft spots are significant, though. First, a \"reference frame\" is never formally defined. The paper acknowledges this and calls it future work, which is fine for a philosophical position paper, but it makes the main claim hard to pin down. Second, the logical step: the argument shows each of the four properties is frame-dependent, then concludes agency is frame-dependent. The stress-test concern is right: showing property i can be flipped by some frame f_i does not show there is a single pair of complete frames that flips all four at once. To establish the conclusion, you would need to exhibit a system where two full frames — each making all four commitments — give different agency verdicts. The paper doesn't do that. In fact, it's not obvious such frames exist: the commitments may interact, and a frame that makes a system adaptive might also force a particular boundary.\n\nThat said, the paper is honest about its limits. It says at the outset that the argument is purely philosophical, and it explicitly stops short of formal definitions and proofs. So the gap is not hidden. The question is whether the gap is fatal for the intended contribution. I think it is not fatal for a synthesis paper: the claim is plausible, and the authors have laid out a research program. But the central claim is under-supported as stated.\n\nWho is this for? People working on AI safety, value alignment, and the foundations of RL will get value from it — it's a good map of why agency attribution is slippery. It deserves a serious referee: the venue should exercise it, but the paper is worth engaging deeply. My recommendation is to send it to review, requesting the authors either formalize the frame definition or at least sharpen the conjunction argument.\n\nCheers.","headline":"A useful synthesis of old ideas under a new label, but the central claim is under-supported because the paper never shows a single frame flips all four properties of agency at once.","tokens_in":6549,"tokens_out":2408,"would_cite":false,"duration_ms":25799,"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":"Agency is not an intrinsic property of a system; it must be measured relative to a chosen reference frame.","keywords":["agency","frame-dependence","reference frame","reinforcement learning","goal-directedness","adaptivity","underdetermination of reward","philosophy of AI"],"falsifier":"A single system for which every admissible reference frame yields the same agency verdict would falsify the claim; for instance, if a formally defined thermostat could be shown to be non-agentic no matter which boundary, causal variables, goal principle, and adaptation class are chosen, the universal frame-dependence thesis would fail. A more direct test is to formalize reference frames and exhibit a system with provably frame-invariant agency status.","tokens_in":5501,"feed_emoji":"🧭","tokens_out":10415,"duration_ms":85199,"temperature":0.7,"pith_summary":"This paper argues that agency—a system's capacity to steer outcomes toward a goal—is not an intrinsic property of the system, but depends on a choice of reference frame. It examines the four standard conditions for agency, namely individuality, being the source of one's own action, goal-directedness, and adaptivity, and argues that each condition can flip its verdict when the frame changes. The upshot is that asking whether a given system, such as a thermostat or a rock, 'has agency' is ill-posed without specifying the reference frame. If correct, this means any basic science of agency, and any empirical measurement of agency in artificial intelligence and biology, must be explicitly frame-relative.","feed_headline":"Agency is frame-dependent: no system is an agent on its own","feed_subtitle":"All four agency tests flip with boundary, causes, goals, and adaptation; agency must be measured relative to a frame.","key_machinery":"The central object is an 'agent reference frame,' defined as the tuple of commitments needed to measure agency: a boundary separating system from environment; a reference object, such as a choice of causal variables, for adjudicating the source of action; a principle for recognizing meaningful goal-pursuit; and a reference class of behavior changes that count as adaptation. The argument's load-bearing move is to show that for each of the four properties, at least two plausible frames yield opposite verdicts for the same system, so the conjunction—agency—is frame-dependent. The frame is an arbitrary upstream commitment that must be fixed before any agency measurement can be made.","core_discovery":"In its own terms, the paper's central claim is that agency is frame-dependent: any measurement of a system's agency must be made relative to a reference frame. It takes the four-part account of agency—individuality, source of action, goal-directedness (normativity), and adaptivity—and shows that each part relies on an extraneous commitment that can be varied without changing the system. For individuality, multiple plausible boundaries exist; for source of action, the choice of causal variables can reveal or hide an agent; for normativity, behavior underdetermines goals; and for adaptivity, the choice of a reference class decides whether a policy is adaptive. Because agency is the conjunction of these properties, every agency verdict is relative to the frame. The paper is explicit that it offers a philosophical argument rather than a formal mathematical proof.","pith_inferences":["If the claim is right, then asking whether an AI system 'really' has agency is a category error: the real question is which reference frame is most useful or defensible for the purpose at hand.","There may be an analogy with relativity: frame-dependent quantities coexist with objective invariants, and finding the invariants of agency could turn the philosophical claim into a productive research program.","A concrete extension would be to formalize reference frames as tuples of mathematical objects and prove that simple systems (a thermostat, a neural network) admit frames assigning opposite agency verdicts; that would make frame-dependence a theorem rather than an argument.","Because the paper identifies reward underdetermination as evidence for normativity frame-dependence, inverse reinforcement-learning methods that impose priors such as maximum entropy are implicitly choosing a frame—an insight that could make learned goal judgments more transparent."],"forward_implications":["Empirical claims about agency must be reported together with the reference frame that produced them; a bare assertion that a system is or is not an agent is incomplete.","Debates about whether particular systems—thermostats, robots, neural networks—are agents shift from a yes/no question to the question of which frame is being used and why.","In reinforcement learning, choices about reward, goals, and policy adaptation carry implicit frame commitments that are currently left unspecified.","A formal science of agency would require defining reference frames and then studying what remains invariant across them, rather than asking for a single intrinsic agency verdict.","Plausible frame-selection principles, such as explanatory or predictive power, can make the intentional stance one method among many for choosing a frame."],"supporting_citations":[{"why":"Supplies the four-part definition of agency (individuality, source of action, normativity, adaptivity) that the paper's argument analyzes.","marker":"Barandiaran et al., 2009"},{"why":"Provides proposition 10 showing that agent-environment boundary choices change key RL quantities, supporting the claim that individuality is frame-dependent.","marker":"Jiang, 2019"},{"why":"Argues that many plausible boundaries can be drawn around an agent, supporting the individuality claim.","marker":"Harutyunyan, 2020"},{"why":"Develops the causal account showing that discovering an agent in a causal model is relative to the choice of variables, supporting the source-of-action claim.","marker":"Kenton et al., 2023"},{"why":"Introduces the classical inverse RL observation that the zero reward function is consistent with any behavior, grounding the normativity underdetermination claim.","marker":"Russell, 1998"},{"why":"Establishes the reward underdetermination result in inverse RL, grounding the claim that goal-directedness is frame-dependent.","marker":"Ng and Russell, 2000"},{"why":"Makes the claim that every system is adaptive with respect to something, grounding the adaptivity frame-dependence claim.","marker":"Zadeh, 1963"},{"why":"States Theorem 3.1, that policies can be understood as adaptive or not depending on a reference class of behavior changes.","marker":"Abel et al., 2023"},{"why":"Supplies the intentional stance and the rock/thermostat/robot puzzle that motivates the need for a frame-dependence account of agency.","marker":"Dennett, 1989"},{"why":"Provides maximum entropy inverse RL as an example of an upstream principle that selects among underdetermined goal attributions.","marker":"Ziebart et al., 2008"}],"fun_headline_variants":["Agency exists only relative to a frame","No agency without a reference frame","Frame-dependence: the core of agency","Agency verdicts always depend on the frame"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The central argument assumes that the notion of a reference frame can be made precise enough to vary the four agency properties while keeping the system fixed, and that no frame is objectively privileged; if frames are unformalizable or constrained by objective standards, the claim becomes vacuous or false.","fun_headline_variants_meta":{"raw":{"variants":["Agency exists only relative to a frame","No agency without a reference frame","Frame-dependence: the core of agency","Agency verdicts always depend on the frame"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000174,"raw_usage":{"total_tokens":1235,"prompt_tokens":854,"completion_tokens":381,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":470,"completion_tokens_details":{"reasoning_tokens":327}},"tokens_in":470,"tokens_out":381,"duration_ms":4568,"temperature":1.0,"reasoning_tokens":327,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-09T00:25:15.619843+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"A single system for which every admissible reference frame yields the same agency verdict would falsify the claim; for instance, if a formally defined thermostat could be shown to be non-agentic no matter which boundary, causal variables, goal principle, and adaptation class are chosen, the universal frame-dependence thesis would fail. A more direct test is to formalize reference frames and exhibit a system with provably frame-invariant agency status.","supporting_citations":[],"review_version":1}