PaperlayerView exact revision on arXiv

Source-linked AI summary

What Can Analogy Tell Us About Artificial Consciousness?

Keith J. Holyoak, Martin M. Monti

arXiv:2610.01002v1cs.AI

TL;DR

The paper asks whether current AI could plausibly have subjective experience, a question with ethical and legal implications if AI is conscious. It develops a causal framework for evaluating analogical evidence and concludes that human-based analogy currently offers only limited support for consciousness in AI.

  • Problem

    The paper asks whether current AI work provides a plausible route to subjective experience, a question with potential moral and legal implications if AI systems are conscious.

    Source: Introduction · p. 1

  • Method

    The paper develops a general framework for evidential analogy that weights similarities by causal relevance while accounting for unknown causes, disabling differences, and alternative routes.

    Source: Evidential Analogy · pp. 1–3, Five questions for evaluating claims… · p. 4

  • Results

    The human-based analogy provides only limited support for consciousness in present AI because its behavioural similarity to humans is not matched by strong correspondence to candidate mechanisms of consciousness.

    Source: Conclusion · p. 8

  • Takeaways & Limitations

    Assessing consciousness claims requires identifying evidence for causal mechanisms, including principled evidence for routes that differ from the human case.

    Source: Conclusion · p. 8

    The analysis is not definitive: new knowledge of biological consciousness could change the analogy, while support for non-biological routes depends on prior beliefs about their possibility.

    Source: What Does the Human Analogy Imply for… · p. 7

Abstract

from arXiv · show

Who or what is conscious? Because subjective experience is directly accessible only in the first person, judgments about consciousness in other entities depend partly on analogy. Historically, such inferences have focused on nonhuman animals, but advances in artificial intelligence have raised the possibility of conscious AI. Here we develop a causal framework for evaluating such evidential analogies. The key distinction is between similarities in factors plausibly involved in generating consciousness and similarities in downstream behavioural or cognitive effects. Our framework weights source-target similarity by causal relevance while allowing for unknown causes, disabling differences and alternative routes to consciousness. Applied to biological systems, it explains why analogical support generally weakens with increasing causal distance from humans. Applied to contemporary AI, it suggests that behavioural similarity provides only limited evidence for consciousness because relevant causal correspondences remain poorly established. The framework also clarifies what evidence would strengthen claims of artificial consciousness.

Introduction

The paper asks whether current or near-future AI could have subjective experience, focusing on what makes an entity feel anything at all. The question also carries ethical and legal implications if AI systems are conscious.

Introduction

  • Recent advances in AI, especially large language and multimodal models, have prompted speculation that current or near-future systems may be conscious.The paper examines whether current AI work offers a plausible route to subjective experience.Source: Introduction · p. 1
  • The paper focuses on subjectivity—what makes an entity feel some way at all—rather than phenomenal character or access consciousness.Phenomenal character concerns the specific way experience feels; access consciousness concerns information availability for reasoning, language, and action control.Source: Introduction · p. 1
  • If AI systems are conscious, they might have moral value or rights, raising questions about personhood, property, and political participation.The passage notes that conscious beings, particularly those capable of pain and pleasure, are generally recognized by moral philosophers as having some moral value.Source: Introduction · p. 1

Evidential Analogy

The paper frames judgments about consciousness in animals and AI as evidential analogies to humans, but argues that similarity matters only insofar as it tracks plausible causes. Its framework accommodates unknown contributors, target-specific disruptions, and alternative routes, then spells out how analogical evidence should be assessed.

Evidential Analogy

  • An evidential analogy supports attributing a property to a target when it resembles a known source in relevant ways, with human consciousness serving as the source for evaluating animals and AI.The paper also considers how consciousness may be distributed across animal species because those criteria bear on AI assessment.Source: Evidential Analogy · p. 1
  • Similarity in plausible causes supports expecting similar effects, but resemblance in downstream behaviour or cognition alone does not establish shared consciousness-generating mechanisms.The framework gives greater weight to causally relevant similarities than to similarities in downstream effects.Source: Evidential Analogy · pp. 1, 2, 4
  • Near analogies are especially informative when causal mechanisms are poorly understood, because close source-target similarity makes sharing unknown causal relations more likely.As causal knowledge improves, similarity on independently established causal factors matters more than overall resemblance.Source: Evidential Analogy · p. 2, Five questions for evaluating claims… · p. 4
  • The human mechanism is modeled as a soft INUS set: candidate causes can vary in importance and need not each be strictly necessary.This avoids treating consciousness as dependent on a single, fully known conjunction of necessary causes.Source: Evidential Analogy · p. 3
  • The framework allows for unknown causal contributors, target-specific differences that disrupt a source-like mechanism, and consciousness arising through an alternative mechanism.Its formalism is intended to organize evidence, not assign precise numerical probabilities.Source: Evidential Analogy · pp. 3–4, Five questions for evaluating claims… · p. 4

Five questions for evaluating claims of consciousness

  • Causal mismatches weaken an analogy in proportion to their relevance, and differences that interfere with a shared mechanism can reduce its support.An incidental mismatch matters less than the absence of a factor with high causal weight.Source: Five questions for evaluating claims… · p. 4
  • Failure of a human-based analogy does not establish that a target lacks consciousness, but shifts the evidential burden to independent support for an alternative mechanism.The framework does not assume that the human route is the only possible route to consciousness.Source: Five questions for evaluating claims… · p. 4
  • If a hypothesis depended partly on an analogy, evidence of weaker causal overlap should reduce confidence in that hypothesis.The framework provides this directional guide even when exact parameter values are unavailable.Source: Five questions for evaluating claims… · p. 4

Consciousness in Humans

The section argues that analogical evidence for consciousness should track similarities in causally relevant mechanisms, not just outward resemblance. Animal comparisons show how support can rest on convergent biological evidence, while AI’s behavioral similarities to humans have not established comparable causal mechanisms.

Consciousness in Humans

  • Consciousness is associated with interacting arousal and thalamocortical systems, yet complex cognition, language, and behavior can occur without it or be impaired while consciousness remains.The section cites anesthesia, disorders of consciousness, and preserved awareness despite severe cognitive impairments as evidence that cognitive architecture alone carries little causal weight.Source: Consciousness in Humans · pp. 4–5

Animals as a Calibration Case for Evidential Analogy

  • Analogical support across animals rests on convergent structures and processes plausibly related to consciousness, and generally weakens as causal and evolutionary distance from humans increases.Evidence includes extensive mammalian homology, partial bird homology, and independently evolved cephalopod nervous systems with functional correspondences.Source: Animals as a Calibration Case for… · pp. 5–6

What Does the Human Analogy Imply for AI Consciousness?

  • Implementations of access-consciousness theories and symbol-like LLM representations have not supplied evidence of subjective experience.The cited global-workspace simulations captured neural and behavioral findings, but their researchers did not conclude that the systems had inner experience.Source: What Does the Human Analogy Imply for… · p. 6
  • Human-level intellectual abilities are not a sufficient basis for inferring consciousness: conscious birds and octopuses lack them, and consciousness appears to predate such abilities biologically.The section characterizes the assumption that high intelligence alone generates experience as the “intellectualist fallacy.”Source: What Does the Human Analogy Imply for… · p. 7
  • AI systems produce human-like behavior through engineered, nonhuman mechanisms, with little evidence that mechanisms implicated in human consciousness play a corresponding causal role.The section notes that AI training uses human-generated data, while behavioral competence is treated as a downstream effect with low causal weight.Source: What Does the Human Analogy Imply for… · pp. 6–7
  • Further assessment requires identifying biological causes, improving comparative inference, and testing both human-like and alternative mechanisms in artificial systems.The analysis is not definitive; biological components might be replicated in future systems, while physical-substrate differences remain an uncertainty for current hardware.Source: What Does the Human Analogy Imply for… · p. 7

3. What would count as evidence for AI consciousness?

Behavioral competence alone cannot determine whether an AI system is conscious. Stronger evidence would connect system properties to mechanisms plausibly involved in generating subjective experience.

3. What would count as evidence for AI consciousness?

  • Candidate computational or architectural properties need evidence of a causal role in consciousness, rather than merely in intelligent behavior.The research questions distinguish mechanisms that play a causal role from those that only support intelligent behavior.Source: §3 · p. 8
  • Mechanistic evidence must be distinguished from behavioral imitation and successful prediction of human reports.Source: §3 · p. 8
  • Intervention-based tests could assess whether candidate AI mechanisms play roles analogous to consciousness-related mechanisms in biological systems.Source: §3 · p. 8
  • Evidence should also clarify what would justify greater confidence in consciousness when an AI system differs substantially from human biology.Source: §3 · p. 8

4. Alternative routes and emerging systems

The section frames alternative routes to consciousness as possible but requiring independent evidence. It identifies causal tests, substrate-neutral predictions and evaluation of emerging biological–artificial systems as open research questions.

4. Alternative routes and emerging systems

  • Weak evidence for a human-like mechanism does not exclude consciousness through another causal route, but such claims require independent evidence.Source: §4 · p. 8
  • The section asks what empirical findings would support a genuinely non-biological route to consciousness.Source: §4 · p. 8
  • It asks how substrate-neutral theories could generate predictions distinguishing conscious from non-conscious systems.Source: §4 · p. 8
  • It asks which properties of neuromorphic, organoid and hybrid biological–artificial systems matter most for evaluating consciousness.Source: §4 · p. 8
  • It asks when an emerging system might become a new source case rather than remain a distant analogue of humans.Source: §4 · p. 8
  • Progress requires moving from behavioural resemblance to causal tests of biological mechanisms, cross-species generalization and evidence for genuinely alternative routes.The proposed agenda includes identifying mechanisms that generate consciousness in known biological systems and specifying evidence for alternative routes to conscious experience.Source: §4 · p. 8

Conclusion

The conclusion argues that analogical evidence depends more on similarities in plausible causes of consciousness than on downstream behavioural or cognitive resemblance. This distinction supports animal-consciousness inferences more strongly than current-AI inferences, while leaving alternative routes possible but in need of evidence.

Conclusion

  • Analogical force depends primarily on similarity in factors plausibly involved in generating consciousness, not downstream behavioural or cognitive effects.Source: Conclusion · p. 8
  • Shared evolutionary continuity, physiology and functional organization support animal-consciousness analogies more than current-AI analogies, whose candidate-mechanism correspondence is weaker.The human-based analogy therefore provides only limited support for attributing consciousness to present AI.Source: Conclusion · p. 8
  • Weak support from the human analogy is not proof of impossibility; progress requires clarifying human causal mechanisms, comparing animal taxa and specifying evidence for alternative routes.Artificial consciousness could in principle arise through a different mechanism, substrate or organizational route.Source: Conclusion · p. 8
Loading 2610.01002v1…