The question of conscious AI is profoundly difficult. The primary problem lies not with our failure to understand AI, but with our failure to understand consciousness: we do not know how to assess claims of consciousness in systems that differ radically from us (Block Reference Block2002).
Given our ignorance, non-evidential factors – “biases” – are likely to play an important role in determining our attitudes. Seth identifies three such factors (anthropocentrism, anthropomorphism, and human exceptionalism), arguing that they combine to make the hypothesis of conscious AI more plausible than the evidence suggests. I agree that these factors are likely to shape our views of AI consciousness in important ways, but their influence is unlikely to be unidirectional, for they might also bias us to reject the possibility of conscious AI.
Seth suggests that anthropocentrism (the tendency to interpret the world in terms of human values and experiences) and human exceptionalism (the tendency to see ourselves as superior to other forms of life) combine to motivate the view that conscious AI is inevitable. “We humans know we’re conscious…. We think we’re intelligent….and we feel we’re special, so we assume consciousness and intelligence go together.” This is indeed one possible inference. Equally, however, anthropocentrism and human exceptionalism might combine to motivate the view that conscious AI is impossible – or at least unlikely.
Consider how things stand at present. Although most of us recognize that humans don’t have a monopoly on consciousness, we do tend to think of ourselves as occupying an exalted position in the “consciousness club.” Other species have impressive sensory repertoires, but our conscious capacities (imagination, insight, creativity, and so on) are uniquely impressive. AI systems might outperform us in many ways, but they don’t have any of what really matters: consciousness. “AI might take my job,” we tell ourselves, “But at least I’ve got feelings!”
Ascribing consciousness to AI would upend this picture. Not only could we no longer view such systems as “mere tools,” we might also find that our (self-appointed) position as the pre-eminent members of the consciousness club was under threat. After all, conscious AI systems are likely to be conscious in impressive ways, and to have many of the capacities that we take most pride in - rationality, insight, creativity, and the like. (In some sense, these capacities fall under what Seth calls “intelligence,” but it would be a mistake to think that they don’t also involve consciousness. Consciousness isn’t just about feelings.)
What about anthropomorphism – the projection of humanlike properties onto other objects? Seth sees anthropomorphism at work in our tendency to attribute consciousness to Large Language Models (LLMs), and there is clearly some truth to this charge. In interacting with LLMs, we often feel as though we are engaging with another conscious agent. And there is evidence that at least some of us might be inclined to take this experience at face value, with recent studies finding that roughly 18% of Americans are willing to ascribe “sentience” (Anthis et al. Reference Anthis, Pauketat, Ladak and Manoli2025) or “subjective experience” (Dreksler et al. Reference Dreksler, Caviola, Chalmers, Allen, Rand, Lewis, Waggoner, Mays and Sebo2025) to AI systems.
At the same time, it’s surely not unreasonable to think that consciousness is implicated in – and therefore arguably an indicator of – aspects of higher cognition and linguistic competence. In fact, Seth himself seems to embrace a version of this view, suggesting that the development of AI “may be enhanced by implementing some of the functions associated with consciousness.” Crucially, we rely on linguistic and cognitive capacities (such as command-following) to probe for the presence of consciousness in various contexts (e.g., after brain injury or following sedation). I am not, of course, suggesting that command-following would be a good test for AI consciousness. The point, rather, is that given the centrality of linguistic and cognitive tests for consciousness in humans (and other animals), it’s not unreasonable to regard these capacities as relevant to the AI question.
This leads us back to the question of what counts as a “bias.” Intuitively, biases are “non-evidential factors,” but what counts as evidence here? This issue is particularly challenging if one assumes – as Seth, I take it, does – that in looking for evidence (or markers) of consciousness, we have no choice but to start with ourselves (and other animals that we regard as having the capacity for consciousness) (Bayne et al. Reference Bayne, Seth, Massimini, Shepherd, Cleeremans, Fleming, Malach, Mattingley, Menon, Owen, Peters, Razi and Mudrik2024). But given this starting point, some kind of anthropocentrism would seem to be unavoidable. The question isn’t whether to evaluate claims of artificial consciousness on the basis of commonalities between AI systems and ourselves (and thus “anthropocentrically”), but which commonalities matter.
This issue can be brought into sharp relief by considering Terry Bisson’s story “They’re made of meat,” in which two intergalactic visitors discuss the possibility of human consciousness. “You’re asking me to believe in sentient meat,” says one. “I’m not asking you, I’m telling you,” comes the reply. “Thinking meat! You’re asking me to believe in thinking meat!” “Yes, thinking meat! Conscious meat! Loving meat. Dreaming meat.” Are Bisson’s imagined visitors guilty of the bias of chauvinism (anthropomorphism’s twin) in doubting the possibility of human consciousness because we’re made of meat and not silicon? If so, are we guilty of chauvinism in wondering whether silicon-based systems could be conscious? Perhaps. Equally, one could argue that both inferences are reasonable (albeit potentially flawed) implications of the fact that when faced with alien systems, one has no alternative but to extrapolate from oneself.
The upshot is that although anthropocentrism clearly can function as a “bias,” it’s arguable that any response to the challenge of artificial consciousness will involve a kind of anthropocentrism. The crucial challenge here is figuring out how to ground inferences about artificial consciousness on the basis of what we (think we) know about human consciousness. The notion of a “bias” provides little traction on that issue, not least because of the difficulty in providing a theory-neutral account of what qualifies as a “bias.” Better, then, to tackle the problem of artificial consciousness head-on by asking what follows from our best models of human consciousness.
You have
Access
The question of conscious AI is profoundly difficult. The primary problem lies not with our failure to understand AI, but with our failure to understand consciousness: we do not know how to assess claims of consciousness in systems that differ radically from us (Block Reference Block2002).
Given our ignorance, non-evidential factors – “biases” – are likely to play an important role in determining our attitudes. Seth identifies three such factors (anthropocentrism, anthropomorphism, and human exceptionalism), arguing that they combine to make the hypothesis of conscious AI more plausible than the evidence suggests. I agree that these factors are likely to shape our views of AI consciousness in important ways, but their influence is unlikely to be unidirectional, for they might also bias us to reject the possibility of conscious AI.
Seth suggests that anthropocentrism (the tendency to interpret the world in terms of human values and experiences) and human exceptionalism (the tendency to see ourselves as superior to other forms of life) combine to motivate the view that conscious AI is inevitable. “We humans know we’re conscious…. We think we’re intelligent….and we feel we’re special, so we assume consciousness and intelligence go together.” This is indeed one possible inference. Equally, however, anthropocentrism and human exceptionalism might combine to motivate the view that conscious AI is impossible – or at least unlikely.
Consider how things stand at present. Although most of us recognize that humans don’t have a monopoly on consciousness, we do tend to think of ourselves as occupying an exalted position in the “consciousness club.” Other species have impressive sensory repertoires, but our conscious capacities (imagination, insight, creativity, and so on) are uniquely impressive. AI systems might outperform us in many ways, but they don’t have any of what really matters: consciousness. “AI might take my job,” we tell ourselves, “But at least I’ve got feelings!”
Ascribing consciousness to AI would upend this picture. Not only could we no longer view such systems as “mere tools,” we might also find that our (self-appointed) position as the pre-eminent members of the consciousness club was under threat. After all, conscious AI systems are likely to be conscious in impressive ways, and to have many of the capacities that we take most pride in - rationality, insight, creativity, and the like. (In some sense, these capacities fall under what Seth calls “intelligence,” but it would be a mistake to think that they don’t also involve consciousness. Consciousness isn’t just about feelings.)
What about anthropomorphism – the projection of humanlike properties onto other objects? Seth sees anthropomorphism at work in our tendency to attribute consciousness to Large Language Models (LLMs), and there is clearly some truth to this charge. In interacting with LLMs, we often feel as though we are engaging with another conscious agent. And there is evidence that at least some of us might be inclined to take this experience at face value, with recent studies finding that roughly 18% of Americans are willing to ascribe “sentience” (Anthis et al. Reference Anthis, Pauketat, Ladak and Manoli2025) or “subjective experience” (Dreksler et al. Reference Dreksler, Caviola, Chalmers, Allen, Rand, Lewis, Waggoner, Mays and Sebo2025) to AI systems.
At the same time, it’s surely not unreasonable to think that consciousness is implicated in – and therefore arguably an indicator of – aspects of higher cognition and linguistic competence. In fact, Seth himself seems to embrace a version of this view, suggesting that the development of AI “may be enhanced by implementing some of the functions associated with consciousness.” Crucially, we rely on linguistic and cognitive capacities (such as command-following) to probe for the presence of consciousness in various contexts (e.g., after brain injury or following sedation). I am not, of course, suggesting that command-following would be a good test for AI consciousness. The point, rather, is that given the centrality of linguistic and cognitive tests for consciousness in humans (and other animals), it’s not unreasonable to regard these capacities as relevant to the AI question.
This leads us back to the question of what counts as a “bias.” Intuitively, biases are “non-evidential factors,” but what counts as evidence here? This issue is particularly challenging if one assumes – as Seth, I take it, does – that in looking for evidence (or markers) of consciousness, we have no choice but to start with ourselves (and other animals that we regard as having the capacity for consciousness) (Bayne et al. Reference Bayne, Seth, Massimini, Shepherd, Cleeremans, Fleming, Malach, Mattingley, Menon, Owen, Peters, Razi and Mudrik2024). But given this starting point, some kind of anthropocentrism would seem to be unavoidable. The question isn’t whether to evaluate claims of artificial consciousness on the basis of commonalities between AI systems and ourselves (and thus “anthropocentrically”), but which commonalities matter.
This issue can be brought into sharp relief by considering Terry Bisson’s story “They’re made of meat,” in which two intergalactic visitors discuss the possibility of human consciousness. “You’re asking me to believe in sentient meat,” says one. “I’m not asking you, I’m telling you,” comes the reply. “Thinking meat! You’re asking me to believe in thinking meat!” “Yes, thinking meat! Conscious meat! Loving meat. Dreaming meat.” Are Bisson’s imagined visitors guilty of the bias of chauvinism (anthropomorphism’s twin) in doubting the possibility of human consciousness because we’re made of meat and not silicon? If so, are we guilty of chauvinism in wondering whether silicon-based systems could be conscious? Perhaps. Equally, one could argue that both inferences are reasonable (albeit potentially flawed) implications of the fact that when faced with alien systems, one has no alternative but to extrapolate from oneself.
The upshot is that although anthropocentrism clearly can function as a “bias,” it’s arguable that any response to the challenge of artificial consciousness will involve a kind of anthropocentrism. The crucial challenge here is figuring out how to ground inferences about artificial consciousness on the basis of what we (think we) know about human consciousness. The notion of a “bias” provides little traction on that issue, not least because of the difficulty in providing a theory-neutral account of what qualifies as a “bias.” Better, then, to tackle the problem of artificial consciousness head-on by asking what follows from our best models of human consciousness.
Financial support
This research received no specific grant from any funding agency, commercial or not-for-profit sectors.
Competing interests
The author has no competing interests to declare.