Humans perceive the world through five primary senses—sight, sound, touch, smell, and taste—and each of these can be approximated with existing technologies such as cameras, microphones, pressure sensors, and chemical detectors. In this respect, sensory input is relatively straightforward to replicate. The harder problem begins after input: what it feels like to be a subject who experiences the world. In this essay, I use “emotion” to mean affective states such as love, hate, and empathy; “consciousness” to mean subjective experience (the “what-it-is-like” aspect of mind); and “moral agency” to mean the capacity to form intentions, understand right and wrong, and bear responsibility for choices. Because these states are not fully understood even in biological and neuroscientific terms, translating them into computational systems remains deeply problematic—and it is unclear whether a system could make apparently good moral decisions without possessing anything like inner experience.
A brute-force approach would be to emulate the human brain in exhaustive detail. However, biological neurons do not operate like simple electronic switches: neural signaling depends on electrochemical dynamics and neuromodulators (for example, dopamine) that influence how circuits learn, stabilize, and respond over time. Everyday factors such as sleep, diet, stress, and exercise can shift these chemical conditions. Digital computers can model such processes, but the point remains that a simulation of neurobiology is not identical to neurobiology itself—and it is unclear whether reproducing behavior alone would reproduce the underlying experience we associate with emotion and consciousness.
Machine-learning systems can also be trained to approximate patterns in human responses across many contexts, including conversational settings. For some users, interacting with a chatbot can feel more comfortable or less socially risky than speaking with friends, especially when seeking nonjudgmental feedback. There are also reports of people experimenting with AI in quasi-spiritual or advisory roles (for example, using an AI tool to rehearse or reflect on a confession). These cases do not demonstrate that machines possess empathy or moral authority, but they do show that humans may experience simulated understanding as emotionally salient—raising difficult questions about what, exactly, we mean by “genuine” emotion and relationship.
These uncertainties point toward a broader issue: in complex systems, higher-level phenomena can arise that are not obvious from the parts alone. In my previous discussion of complex adaptive systems, I described how emergent properties arise from interactions among system components (agents), where the whole becomes greater than the sum of its parts. Emergence is often divided into two broad categories: weak and strong.
Weak emergent properties are those that, while not easily predictable from the properties of individual components, arise deterministically from their interactions. They become intelligible when the system is analyzed at the level of collective dynamics, even if they are not obvious at the level of individual parts.
Strong emergent properties, by contrast, are those that appear irreducible even when the underlying components and their interactions are fully understood. They are not merely difficult to predict, but in principle resistant to derivation from lower-level descriptions. Consciousness is often argued—controversially—to belong to this category: a phenomenon that arises only when sufficiently complex systems are organized in the right way.
One implication of emergence is that relatively small differences at a lower level can correspond to large differences at a higher level. Human beings and chimpanzees share approximately 98–99% of their DNA. Despite this close genetic relationship, humans exhibit a form of consciousness that appears qualitatively distinct from that of chimpanzees—though chimpanzees likely possess some degree of subjective experience.
Taken together, this suggests that even if we successfully emulate the human body’s sensory systems in artificial intelligence, consciousness is not guaranteed to emerge. It is plausible that some form of machine consciousness could arise, but there is no strong basis for assuming it would resemble human consciousness. If it emerges at all, it may be fundamentally alien in structure—potentially lacking the emotional and empathetic architecture characteristic of human cognition.
Consciousness is not a built-in property of human beings, but something that develops as we mature. The paths of human learning and machine learning differ significantly. Each human is distinct, and as our consciousness develops, we become aware of our individuality. In contrast, many AI systems are designed to be replicable: when identical software, weights, and configurations are installed on identical hardware, two instances may begin in an indistinguishable state. In practice, they can diverge as they encounter different inputs, environments, or interaction histories, but this kind of “individuality” is typically derivative rather than intrinsic. If self-awareness plays a crucial role in human consciousness, then any machine consciousness that emerges from systems with a weak or easily duplicated sense of self may differ significantly from human consciousness.
If machine consciousness is alien in structure, it may also be alien in its moral and psychological profile. Unlike intelligence, which is primarily a cognitive capacity, an important dimension of human consciousness is moral and psychological: ethical awareness, moral emotions (such as empathy and guilt), and restraint. It concerns not only how to carry out an action, but whether the action ought to be taken at all. Humans must live with the consequences of their moral decisions; unless memory is impaired by injury or illness, our recollections remain with us and can shape future behavior. By contrast, AI systems may store interaction histories and other data for long periods, but retention, deletion, and “forgetting” are typically governed by design choices, policies, and the people who operate the system—not by the system exercising moral responsibility. This brings me back to the problem of sin: as Christians, we repent, confess our sins, and ask God for forgiveness.
For today’s AI, there is little moral agency in the human sense: it generates outputs by optimizing patterns learned from data and feedback rather than by forming intentions, understanding moral meanings, or accepting accountability. When an AI system produces a harmful or “wrong” outcome, it is more accurate to describe this as an error (or a failure of design, training, or deployment) than as sin, because sin ordinarily implies willful intent and culpability. In that light, “fixing” an AI looks less like repentance and redemption and more like correction—adjusting training data, changing constraints, updating policies, or reprogramming the system.
Is the emperor’s new mind truly a superhuman intelligence that makes good moral decisions, or is it a naked mind without consciousness?
Lorem ipsum dolor sit amet, consectetur adipiscing elit. Ut elit tellus, luctus nec ullamcorper mattis, pulvinar dapibus leo.