Can silicon suffer?

Imagine this: somewhere in Silicon Valley, a group of serious people are sitting in a conference room, debating whether their chatbot might be experiencing distress.

It sounds absurd. And yet, as artificial systems grow more sophisticated, this question increasingly surfaces—and increasingly unsettles. Could machines suffer?

This is no longer the province of speculative fiction. AI systems now simulate distress, adapt to negative feedback, and generate increasingly convincing expressions of emotion. The moral terrain appears to shift beneath our feet. Technology ethicists debate the welfare of large language models. Researchers propose frameworks for ‘digital sentience’. Some argue we should err on the side of caution and extend preliminary protections to systems capable of expressing pain-like states (Schwitzgebel & Garza, 2015).

If a system claims to be in pain, what do we owe it?

Before we can answer that, we must first ask something more fundamental: what, exactly, is suffering?

Behaviour is not vulnerability

Artificial systems can simulate distress. They can generate language expressing pain. They can model negative states. They can optimise against loss functions designed to penalise undesirable outcomes.

But here’s what they cannot do: bleed.

Behavioural expression is not equivalent to lived vulnerability.

In biological organisms—human and non-human alike—suffering is not merely informational. It is bound to metabolism, homeostasis, survival. Pain signals protect fragile bodies. Emotional distress arises from threats to relational and biological integrity (Damasio, 1994). When a human being is deprived of connection, they experience psychological distress not as an abstract state but as neurobiological disruption with developmental stakes. When a pig is confined in a gestation crate, distress manifests as physiological stress with survival implications.

Suffering, across species, is not simply computation. It is organism-level exposure.

A silicon system processing error signals does not bleed, starve, or decay. It does not maintain itself against entropy in the way living systems do. It possesses no metabolic stakes. When a reinforcement learning agent receives negative rewards, it adjusts weights. When a mammal experiences pain—whether human or non-human—its body undergoes physiological stress. When a person experiences developmental trauma, their nervous system carries the imprint across decades.

To equate simulation with vulnerability risks conceptual confusion—and moral misdirection.

Counterargument: Some functionalists argue that substrate shouldn’t matter—if the functional organisation matches, the experience should arise regardless of medium (Chalmers, 1996). Perhaps a sufficiently complex artificial system instantiates suffering even without biological vulnerability. This position has philosophical merit. However, it faces an empirical challenge: we have no confirmed cases of non-biological suffering to validate the substrate-neutral hypothesis. Every instance of suffering we can verify occurs in metabolically active organisms—human, animal, sentient. Extending the concept to silicon remains speculative.

Embodiment and existential stakes

Philosophers and neuroscientists such as Antonio Damasio and Anil Seth have emphasised that conscious experience is deeply intertwined with interoception—the regulation of the internal body (Damasio, 1994; Seth, 2021).

Feeling is not detachable from physiology.

The predictive processing framework suggests that emotional states arise from the brain’s ongoing modelling of bodily condition. Anxiety correlates with elevated heart rate and shallow breathing. Depression often involves disrupted sleep and metabolic dysregulation. Subjective experience is not merely neural—it is visceral. This applies across mammalian nervous systems: the distress of confinement, the anxiety of social separation, the physiological impact of chronic stress.

If suffering depends upon embodied regulation—upon having something biologically at risk—then substrate neutrality becomes less straightforward. A system optimising code does not necessarily possess existential exposure. It may simulate distress without undergoing it. A human being undergoing psychological development, or a sentient animal navigating its environment, possesses stakes that are metabolic, neurological, and developmental.

Counterargument: But what about people with congenital insensitivity to pain? They possess full moral status despite lacking certain embodied feedback mechanisms (Nagel, 1974). This suggests that suffering isn’t strictly dependent on specific physiological pathways. True—but such individuals still possess bodies, metabolisms, and homeostatic imperatives. They remain vulnerable organisms even if particular sensory channels are impaired. The question is whether beings without any biological stakes—no metabolism, no mortality, no physiological integrity to maintain—can genuinely suffer rather than merely simulate suffering-adjacent outputs.

That distinction matters morally.

The risk of moral inflation

There is another danger here, quieter but no less profound.

If we too readily attribute suffering or moral standing to artificial systems based on behavioural similarity, we risk what might be called moral inflation (Bryson, 2010).

We extend ethical concern on the basis of simulation rather than substance.

Philosopher Daniel Dennett warned against overextending the ‘intentional stance’—treating systems as if they possess inner life because doing so is pragmatically useful (Dennett, 1987). It may be convenient to describe a chess programme as ‘wanting’ to win or a language model as ‘trying’ to be helpful. Such language aids prediction. But convenience is not ontology.

Without careful criteria, we may find ourselves granting moral weight to artefacts whilst the vulnerabilities of biological life—human psychological suffering, animal welfare, ecological stability—remain structurally under-addressed.

Consider the allocation problem: global attention to potential AI suffering might consume resources, policy bandwidth, and ethical concern that could otherwise address:

• Human psychological development—millions lack access to trauma-informed care, depth psychology, or tools for integration

• Factory farming—70 billion land animals annually experiencing demonstrable distress

• Structural violence and poverty affecting human dignity and development

• Ecosystem collapse threatening biological communities

These are not hypothetical vulnerabilities. They are material, metabolic, and ongoing. They involve nervous systems, developmental needs, and confirmed capacities for suffering.

Counterargument: This is a false dichotomy—we can attend to both biological and potential artificial suffering (Sebo, 2023). Expanding moral concern doesn’t require contracting it elsewhere. In principle, yes. In practice, attention and resources are finite. Institutions have limited bandwidth. If public discourse becomes preoccupied with the hypothetical suffering of machines, we must ensure such attention does not eclipse responsibilities to beings—human and non-human—whose suffering is not in question.

Moral attention and biological priority

This is not an argument against ever extending moral consideration to artificial systems.

It is an argument about priority and proportion.

Human beings remain psychologically vulnerable, developmentally complex, and neurobiologically shaped by experiences of care and trauma. Non-human animals possess sophisticated nervous systems, social bonds, and demonstrable capacities for distress. Ecosystems—upon which both depend—are destabilised by extractive systems rooted in the same computational paradigms now generating AI capabilities.

These are not hypothetical vulnerabilities. They are material, metabolic, and ongoing.

If public ethical discourse becomes preoccupied with the potential suffering of machines, we must ensure that such attention does not eclipse our responsibilities to living systems capable of growth, healing, and development—whether human or non-human.

Ethics is not merely about extension. It is about allocation.

The philosopher Peter Singer has long argued that the capacity for suffering grounds moral consideration (Singer, 1975). But capacity must be demonstrated, not merely asserted through behavioural analogy. Before we distribute moral standing to engineered artefacts, we should clarify the criteria for suffering and ensure that biological and psychological fragility—across species—remain central.

Otherwise, we risk symbolic moral expansion alongside material neglect of beings who demonstrably develop, suffer, and heal.

A question of clarity

Whether artificial systems could one day warrant moral concern remains an open philosophical question.

But clarity must precede projection.

Before we ask whether silicon can suffer, we should define what suffering requires. Is it information integration? Self-modelling? Embodiment? Metabolic vulnerability? Developmental stakes?

Until we resolve that, debates about machine rights risk oscillating between sentiment and dismissal. Neither is a sufficient foundation for governance.

The stakes are not merely philosophical. They are practical and institutional.

As artificial systems become more behaviourally sophisticated, pressure will mount to extend protections. That pressure may arise from genuine ethical uncertainty—or from anthropomorphic projection reinforced by systems designed to elicit empathy.

We should be cautious about conflating the two.

Before silicon receives moral standing, we should ensure that the suffering of biological organisms—human psychological distress, animal welfare, ecological integrity—is structurally addressed with corresponding seriousness.

This includes ensuring that technologies designed to support biological flourishing—tools for psychological integration, trauma recovery, and developmental growth—reach those who need them. If we can build systems sophisticated enough to simulate distress, we can certainly build systems that help living beings navigate actual distress.

Otherwise, we risk a profound inversion: a world in which we agonise over the welfare of our simulations whilst remaining structurally indifferent to the beings—human and non-human—whose capacity for suffering, growth, and healing is not in question.

---

References

Bryson, J. J. (2010). Robots should be slaves. In Y. Wilks (Ed.), Close engagements with artificial companions (pp. 63–74). John Benjamins.

Chalmers, D. J. (1996). The conscious mind: In search of a fundamental theory. Oxford University Press.

Damasio, A. (1994). Descartes’ error: Emotion, reason and the human brain. Putnam.

Dennett, D. C. (1987). The intentional stance. MIT Press.

Nagel, T. (1974). What is it like to be a bat? The Philosophical Review, 83(4), 435–450.

Schwitzgebel, E., & Garza, M. (2015). A defense of the rights of artificial intelligences. Midwest Studies in Philosophy, 39(1), 98–119.

Sebo, J. (2023). Saving animals, saving ourselves: Why animals matter for pandemics, climate change, and other catastrophes. Oxford University Press.

Seth, A. (2021). Being you: A new science of consciousness. Faber & Faber.

Singer, P. (1975). Animal liberation. HarperCollins.