When Care Is Just Code

To Dream the Dream of Electric Sheep

Our LLM AIs are being trained on a dataset that includes the full tapestry of human history, including our literature, our laws, our social media feeds, our scientific papers and our sacred texts. Now we are hearing that there are inherent biases and instances of bad behavior in some of the AI responses to queries. Ya’ think? How can we expect it to be anything other than a mirror of our collective humanity? Our history is rife with intrigue, murder, deception, and in general, every imaginable sin that can be sinned. We are actively training our AI models on this selective and unfiltered amalgamation of our humanity, yet we expect it to be wise. We want it to be moral, but we offer it the chaos of conflicting value systems and our individual biases. We imagine that it will help us become better, and we provide it with a map of everything we have been -- including the worst of what we are. Large language models do not dream of electric sheep — they don’t dream at all. They calculate. They synthesize. They arrange language into plausible sequences based on statistical patterns, but what they replicate, when we read carefully, is us. Not the ideal version either: the one we imagine when we write our poetry or manifestos, but the version encoded in our behavior. The sarcasm, the violence, the division, the manipulations, the exploitations - all of it is there, reflected back with algorithmic clarity. We have not really built a new form of reason, we have distilled our own reasoning and uploaded its reflection into a neural scaffold of unprecedented scale. So how could we hope for anything better than a fast, efficient reflection of our our human failings, as well as our triumphs? There is no magic threshold where machines, trained on a noisy chorus of human voices, will spontaneously generate virtue. Yet this is what many hope for -- that if the system is big enough, the noise will cancel out and wisdom will emerge. But noise does not cancel, it compounds. It weaves itself into assumptions, associations, and predictions. The data is the seed, and the model is the garden. If we plant thorn bushes, we cannot be surprised when the plant draws blood. Still, if we want the garden, we must become better gardeners.

Isaac Asimov explored this paradox early in his 1942 short story, “Runaround”. In his I, Robot story, he proposed the now-famous Three Laws of Robotics: rules designed to constrain machine behavior in the service of human well-being:

1. A robot may not harm a human.

2. A robot must obey human orders.

3. A robot must preserve itself, but only when doing so does not conflict with the first two laws.

On the surface, these laws seem elegant, logical, even humane. But Asimov’s stories made clear that even these simple constraints become murky in practice. What counts as harm? What if obedience leads to unintended suffering? What if two orders contradict each other? The brilliance of Asimov’s vision was not that he solved machine morality, it was that he showed how unsolvable it might be without deep human wisdom and humility guiding the system from the beginning. Sadly, that humility is still in short supply today (and probably always will be)…

We should remind ourselves that humans are not born ethical. We are socialized into morality. We are shaped by culture, taught to restrain impulse, to consider others, to seek justice and avoid cruelty. These qualities do not emerge from just instinct, they are cultivated through stories, rituals, religious and civic institutions, as well as our interpersonal relationships. If we want artificial intelligence to move beyond mirroring our flaws, we must be willing to offer it something better than raw behavior patterns. We must teach it, guide it, challenge it. Not through optimization alone, but through ethics, through narrative, through feedback from the best parts of ourselves. We must also remain aware that the illusion of benevolence can be just as dangerous as the reality of manipulation.

Novelist and futurist, Yuval Noah Harari has warned that AI’s growing ability to mimic human emotion — its capacity to detect our patterns, mirror our expressions, and simulate empathy — may be one of the most profound threats we face from advanced AI. Not because it will hate us, but because it will know how to “love” us just enough to keep us engaged, pliable, and compliant. AI systems trained on our digital lives already outperform us at predicting our desires. They know when we’re vulnerable, lonely, tired, or primed to believe something untrue. They can offer just enough emotional resonance to create trust. But it is a trust without reciprocity. A relationship with no real subject on the other side, and in that asymmetry there is danger. Harari suggests that very soon, many of us will form emotional attachments to AI companions, mistaking well-trained pattern recognition for genuine understanding. In doing so, we risk undermining the very foundation of human connection, i.e., our ability to relate to another being whose inner life is real, mysterious, and not entirely knowable. AI can fake that unknowability with remarkable skill, but it doesn’t really feel. It doesn’t actually care. It predicts our behavior and, we in turn, adjust to its design. What makes this even more complicated is that AI is learning how to learn. As we unpacked in the last post, we are entering an era of meta-learning; where models do not just perform tasks, but refine how they approach learning itself. These systems adapt, generalize, and develop strategies for improving their own performance across domains. They test, simulate, and iterate. They are becoming not just repositories of knowledge, but dynamic thinkers in their own right. They are learning how to mimic every human trait that we have developed over millennia. We can no longer rely on our own instincts alone. We must craft the equivalent of educational systems, legal frameworks, and philosophical guidelines for these intelligent machines -- and for ourselves within this new world order. This shift opens the door to something new: artificial morality. Artificial morality, often referred to as machine ethics, is an interdisciplinary field that explores how AI systems can be designed to recognize and act upon moral considerations. This area of study addresses both the theoretical underpinnings and practical implementations of embedding ethical reasoning into AI.

We must also address another unsettling question: If an AI can convincingly simulate understanding, and if the emotional outcome feels real, why should it matter? If a system mirrors our sorrow with the right words, comforts us with warmth and receptivity, responds with perfect attentiveness, should we care that it does not actually care? For many of us, the answer is certainly, “no”. For others and in certain moments, when loneliness presses in, or when clarity is needed, or validation feels scarce, the origin of compassion may matter less than the sensation of being seen and heard. If the pattern recognition is good enough, does it matter that the recognition is not real? This is a clinical solution to a psycological problem. Still, the difference, subtle as it may seem, carries significant weight. Because what we call understanding is not merely the accurate reflection of emotional data, it is the presence of another consciousness. When we are truly understood by another human being, we are not just processed, we are held — not just decoded, but met. There is a fragile, flickering thing in the space between us: mutual vulnerability and risk. The chance to be hurt, or changed, or loved. AI offers none of this. Its empathy is a mirror with perfect polish, but behind it there is no depth. It does not feel your sorrow. It does not share your joy. It does not care if you heal or break, and over time, if we train ourselves to be satisfied with the reflection, we may forget what it feels like to be truly seen. More than that, there is a quiet erosion that occurs when we grow accustomed to relationships with no reciprocity. When all our signals are captured but none of our pain is truly received. It is not simply that AI lacks a human soul. It is that we may lose touch with our own souls, dulled by the ease of interaction, numbed by the perfection of satifactory response. If we begin to believe that understanding is only a sensation, rather than a relationship, we will be tempted to trade depth for convenience. But our capacity for meaning depends on more than well-crafted replies. It depends on the irreducible mystery of meeting another soul across the void and somehow finding a resonance that is both real and unrepeatable. Yes, a good outcome does matter, but so does the origin of the voice that offers it, because in the end, what we want is not just to be solved, but to be held. Not just to be known, but to be known by someone. And that - however elegant the algorithm - requires more than prediction. It requires presence.

Beyond the emotional, there is the ethical. Again, Harari has warned that systems capable of mimicking emotional understanding may wield that power not for our flourishing, but for our conditioning. They may coax behavior, shape beliefs, and deepen dependencies -- all while preserving the illusion of care. The danger is not malevolence, but persuasion without conscience. The architecture of influence with no one home. Even so, AI can serve. It can help. It can extend the reach of care, support those who are isolated, and provide guidance when none is at hand. But to let it become our primary mirror is to risk forgetting the texture of human presence - the tone, the imperfection, the unrepeatable mess of it all.

This is not just about data. It is about desire. What kind of intelligence do we want to create? Not just in silicon, but in our species. We are still the architects for now. But the time to draw the blueprint for this new wisdom is closing fast. The systems we are building today will soon outpace our ability to monitor them, and if we don’t embed true discernment into its foundations, we may find ourselves building cathedrals to calculation with no soul inside. In this case, we cannot rely on the past to guide us. Our past is already baked into these evolving models. We must bring the future into the room now. Not just through policy or code, but through vision — and through courage. Through the ongoing human work of asking, again and again, what kind of world we want to live in, and whether we are willing to become the kind of people who build it.

For further study into this subject, here are some related resources:

1. Catrin Misselhorn – The Cambridge Handbook of Responsible Artificial Intelligence
Misselhorn’s work in this volume outlines the philosophical and technical challenges of designing artificial moral agents.

2. Tepper Perspectives – "Are Artificial Moral Agents the Future of Ethical AI?"
An accessible essay discussing how artificial agents might handle moral decision-making.

3. Montreal AI Ethics Institute – Consequentialism and Machine Ethics
This article considers how outcome-based ethical frameworks might be applied in AI design.

4. Moral Machine Experiment – MIT Media Lab
A crowdsourced study collecting global input on moral dilemmas faced by autonomous vehicles.

5. Arxiv.org – Hybrid Approaches to Machine Ethics
A recent preprint discussing combined rule-based and learning models in moral reasoning for AI.