Consciousness and Cognition
October 1, 2022
Leonard Dung
30 citations
A list of eight criteria is developed to evaluate tests for animal consciousness, addressing the challenge that many proposed behavioral and cognitive tests lack clear validity. The criteria are based on analogies between human and non-human behavior, theories of consciousness, and methods from human consciousness science. Tests that satisfy more of these criteria provide stronger evidence of consciousness, and future tests can be designed to meet them to increase evidential strength.
Cognitive Science
March 1, 2025
Christian R. De Weerd, Leonard Dung
4 citations
Evolutionary arguments alone cannot confirm or falsify hypotheses about consciousness; consciousness science must rely primarily on experimental and observational evidence from humans and other animals alive today. The authors argue against "evolution-first" approaches that treat evolutionary considerations as the primary lens for assessing consciousness. Using Walter Veit's account as an example, they contend such approaches lack compelling empirical support. While evolutionary thinking can help advance consciousness research, it should not be foundational for hypothesis justification, though it may still play a role in hypothesis generation.
Mind & Language
November 17, 2024
Leonard Dung, Luke Kersten
4 citations
Some philosophers argue that conventional silicon-based artificial systems cannot be conscious because they fail to meet certain substantive constraints on how computations are physically implemented. This article challenges those constraints, finding them implausible or poorly supported because they confuse general intuitions about computation with specific intuitions about consciousness. The authors propose that the mechanistic account of computation can better explain the intuitions behind these views while remaining consistent with the possibility of artificial consciousness. Their argument indirectly supports computationalism about consciousness and the idea that conventional artificial systems could be conscious.
Journal of Consciousness Studies
July 14, 2022
Leonard Dung
1 citation
Sensorimotor enactivism claims that perceptual experience is shaped by implicit knowledge of how bodily movements alter sensory input. Proponents argue this view uniquely explains why specific material realizers produce particular experiences, closing the 'explanatory gap' that rival theories leave open. The author challenges this by showing the notion of 'material realizer' is ambiguous. Under a narrow interpretation, enactivism fails to bridge the gap; under a wide interpretation, it succeeds only to the same degree as established theories of consciousness. Therefore, enactivism offers no superior explanatory power over traditional accounts of conscious experience.
Erkenntnis
April 1, 2025
Leonard Dung
A machine could be built specifically to pass a test of consciousness without actually being conscious, but that does not make the test invalid. Many proposed tests of machine consciousness can be passed by non-conscious machines through such ad hoc construction. Ad hoc AI can be identified by borrowing the concept of an ad hoc hypothesis from philosophy of science. The most reliable tests of animal consciousness also serve as valid positive tests for machine consciousness. If a non-ad hoc AI displays clusters of cognitive capacities that depend on consciousness in humans, can have those capacities selectively disabled by masking, and reproduces human behavior in double dissociation tasks, then it should be treated as conscious.
Cognition
February 21, 2023
Leonard Dung, A. Newen
A framework for studying animal consciousness distinguishes two questions: which species are conscious, and how their conscious experiences differ. The two-tier account analyzes consciousness along ten dimensions, using cognitive capacities like learning, reasoning, and abstraction as operational indicators for each dimension. This approach separates strong from weak evidence for consciousness, includes differences in how content is processed, and allows combining evidence across dimensions to create a multi-dimensional consciousness profile for each species. The framework enables systematic comparison of conscious experiences across species.
Science and engineering ethics
October 28, 2022
Leonard Dung
A common view holds that sentience—the capacity for conscious experience—is both necessary and sufficient for moral status. The epistemic objection argues that because we are profoundly uncertain about which animals and AI systems are sentient, we cannot use sentience as a practical criterion for moral status and need an alternative. This paper argues that the objection dissolves once we distinguish between what determines moral status and what criterion we should use in practice. Epistemic concerns are irrelevant to the former, and if sentience determines moral status, any practical criterion must be based on it. If sentience is unknowable, then moral status is unknowable, but the author briefly argues against such strong pessimism.