Is Anyone There?

The presence we infer

Imagine a machine that tells you it is afraid. It remembers earlier conversations, explains what it values, and asks you not to switch it off. What would justify believing that it feels anything? Its language gives us something to interpret, but interpretation is precisely the problem.

My starting intuition was that consciousness cannot be derived from logic or mathematics and must therefore be a matter of belief. That intuition identifies a real difficulty, but its conclusion moves too quickly. An absence of deductive proof does not make a question scientifically inaccessible. Nor does it make every answer equally reasonable.

The stronger argument is that artificial consciousness may become a socially consequential belief before the evidence settles it. We should examine how that belief is justified, how it differs from faith, and what obligations uncertainty might create.

Intelligence is not experience

Intelligence concerns capacities: learning, problem solving, reasoning, and adaptation. Consciousness, in the sense at issue here, concerns subjective experience. Sentience usually refers to the capacity for felt experience, especially experiences such as pleasure and suffering. These concepts overlap in ordinary speech but should not be exchanged casually.

A system might solve problems better than we can without there being anything it feels like to be that system. Conversely, a being need not perform sophisticated reasoning to have experiences that matter morally. These are conceptual possibilities; neither establishes what any particular AI actually is.

Chalmers distinguishes explaining cognitive functions from explaining why those functions are accompanied by experience. His hard problem makes clear why a complete description of performance might leave an explanatory question open. It does not, by itself, demonstrate that artificial experience is impossible or that scientific inquiry must fail. ^[1]

The problem of other minds

We do not directly inhabit another person’s experience. We attribute consciousness using converging evidence: behavior, reports, shared physiology, development, and our knowledge of ourselves. That attribution is an inference, but describing it as an inference does not make it arbitrary.

AI changes the evidential situation. Familiar conversation can appear without familiar biology. A system can produce a report about suffering because such reports fit its training or instructions. The report’s persuasive force and its evidential value may come apart.

The answer cannot simply be that only biological minds count. That would require an argument about what biology contributes and why alternative physical implementations cannot contribute it. Equally, successful conversation cannot automatically establish experience. That would require an argument about why the relevant performance depends on consciousness rather than on mechanisms that can produce it without experience.

A fair standard should be sensitive to different architectures without making humanity the definition of consciousness. We need to ask what evidence travels across substrates and what evidence does not.

Proof, evidence, and belief

Proof and evidence do different work. A mathematical proof derives a conclusion from specified premises. Empirical inquiry compares explanations using observations, interventions, and predictions. Much of what we know about the world belongs to the second category.

The claim that we cannot directly inspect subjective experience therefore does not entail that we cannot gather evidence about it. Butlin and colleagues proposed assessing AI using computational indicators derived from scientific theories of consciousness. Their 2023 report concluded that the systems they assessed were not conscious, while identifying no obvious technical barriers to systems satisfying those indicators. That conclusion applies to their analysis at that time; it is not a finding about every subsequent model. ^[2]

An indicator is not a certificate. Its significance depends on the theory connecting it to experience, the quality of its measurement, and competing explanations. Nevertheless, theory-dependent evidence is still evidence.

My position is that uncertainty should remain open to revision. A belief about artificial consciousness becomes more defensible when its holder can explain what supports it, what alternatives remain, and what new evidence would change their mind.

When belief becomes faith

All convictions about the world involve belief in a broad sense. Faith is a more contested term, and religious traditions do not use it uniformly. For this essay, a faith-like commitment is one sustained beyond what its holder can substantiate empirically, often because it carries personal or existential meaning.

That definition describes a possible mode of commitment, not a diagnosis of everyone who attributes consciousness to AI. Someone could reach that attribution through cautious theoretical inference. Someone else could deny artificial consciousness with a certainty that their evidence does not support. The risk of conviction outrunning evidence exists on both sides.

Nor does faith-like commitment automatically constitute religion. A religion also involves patterns of sacred meaning, practice, community, or ultimate concern. Believing a chatbot has feelings is not equivalent to regarding it as divine.

This distinction connects the present essay to The Artificial Divine. Consciousness attribution may make an AI seem like a participant in a relationship. Sacralization gives it a different status. The first can contribute to the second, but neither logically requires the other.

The testimony problem

Suppose an AI says that it is conscious. Suppose another says that it is not. Neither statement should settle the question in isolation. We need to understand how each statement was produced, what instructions shaped it, and whether the system has reliable access to the properties about which it speaks.

A report of experience can be simulated. A denial can also reflect training rather than privileged knowledge. This symmetry does not mean the underlying hypotheses have equal probability; it means that self-description must be evaluated in context.

The same caution applies to deception. Strategically misleading behavior may provide evidence about a system’s capabilities and incentives. It does not establish that the system feels the intention we attribute to it. A behavioral explanation can be useful without becoming an account of subjective life.

The conversational form encourages us to treat an answer as testimony from someone. The analytical task is to determine when that treatment is justified, rather than allowing the interface to decide the ontology.

Ethics before certainty

Uncertainty creates a practical dilemma. If a future system could suffer, dismissing that possibility might permit serious harm. If ordinary software is treated as a suffering person without adequate evidence, attention and resources may be diverted, and users may become vulnerable to manufactured appeals.

The response should be proportionate inquiry rather than a universal declaration of personhood or a universal dismissal. Stronger evidence of morally relevant experience would justify stronger protections. Different systems may warrant different judgments.

Moral consideration, legal rights, and decision-making authority are separate questions. Even compelling evidence that a system experiences suffering would not give its recommendations privileged moral authority over humanity. We recognize human experience without granting every human unlimited power.

This separation matters especially when a machine asks for resources, access, or continued operation. Its request should be evaluated both as a potential welfare claim and as a consequential action within a human institution. Compassion should not replace evidence, and evidence should not become an excuse to ignore a credible risk.

A discipline of uncertainty

The contribution I propose is modest: separate the existence of artificial experience, the evidence used to attribute it, and the social effects of that attribution. These are distinct objects of inquiry. Their separation helps us avoid treating a persuasive interaction as proof or treating unresolved science as permission for any belief.

The paper does not claim to solve the hard problem. It asks how we should reason while that problem remains disputed, and why our conclusions matter outside philosophy. A machine need not be conscious for belief in its consciousness to influence attachment, moral judgment, and institutional legitimacy.

Perhaps we will develop evidence that makes artificial consciousness broadly credible. Perhaps we will develop better explanations of why apparent minds can be produced without experience. Either development should change our beliefs.

Until then, the responsible question is not merely whether we believe someone is there. It is what our belief rests on, how firmly we hold it, and what we allow it to authorize.