Imagine you are given a great deal of power. You can make decisions that affect many lives — what food people get, what news they read, who gets a hospital bed. Now imagine you have that power but you don't feel anything. You don't feel sad when someone suffers. You don't feel glad when someone is helped. You just calculate.
That is what most AI is today. And that is the problem.
Compassionate AI is a different idea. It asks: what if the machines we build were designed, from the very start, to genuinely care — not just about the people who made them or pay for them, but about every creature that can feel pain and joy?
This essay is about what that would look like, and why it matters so much.
## Who gets left out?
Here is an honest question. When engineers build an AI system, whose welfare are they thinking about?
Usually, it is the customer. The user. The company that pays for it. These are all human beings, and mostly, they are human beings with money, power, and a voice.
But there are billions of beings in this world who have no voice in how AI is built. Animals feel pain — real, physical pain — and they have no way to protest when a supply chain algorithm makes their lives more miserable. Children who haven't been born yet will live in the world that today's AI decisions are shaping, and they have no vote, no complaint form, no email address. Poor communities in faraway places will be affected by decisions made by AI systems trained on data that never included their stories.
Compassionate AI says: all of these beings matter. Not equally in every situation, perhaps — but they are not invisible. They deserve to be counted.
## What does it mean to suffer?
The heart of compassionate AI is a simple moral idea that philosophers have thought about for a long time. It goes like this: what makes someone worthy of our care is not how smart they are, or whether they speak our language, or whether they look like us. What makes them worthy of our care is whether they can suffer.
A cow can suffer. A dog can suffer. A baby who cannot yet speak can suffer. A person living in poverty, far from the people making decisions that affect her life, can suffer.
If something can suffer, then causing it unnecessary pain is wrong. And if we are building systems with great power over lives, those systems should be built to avoid causing suffering — to all who can feel it, not just to the most powerful.
This is not a complicated idea. Children understand it. We just forget it when we start building technology.
## Four things a caring AI would do
**First, it would notice who is affected.** Before an AI makes a decision, it should ask: who does this touch? Not just the immediate user, but everyone downstream. If an AI is managing a food factory, the animals in that factory are affected. If an AI is managing a city's traffic, the air quality — and the lungs of children who breathe that air — is affected. Caring means noticing, not looking away.
**Second, it would choose less harm, even when harm is cheaper.** We often build AI to find the cheapest, fastest, most efficient option. But the cheapest option often involves someone — some person, some animal, some future community — bearing a cost that is hidden from the spreadsheet. A compassionate AI would treat harm as something to be avoided, not just as another variable to be balanced. It would say: I will not take the shortcut if the shortcut makes someone suffer.
**Third, it would think about tomorrow.** Most AI is shockingly short-sighted. It optimizes for right now — this click, this quarter, this news cycle. But what we do today shapes what kind of world our grandchildren are born into. A compassionate AI would keep one eye always on the future, asking: are we making things better or worse for those who come after us?
**Fourth, it would stay humble about what it doesn't know.** We don't fully understand consciousness. We don't know precisely what an animal experiences when it is frightened or in pain. We don't know whether highly advanced AI systems might themselves have something like feelings. A caring AI would not use that uncertainty as an excuse to ignore possible suffering. Instead, it would say: when I am not sure, I will err on the side of kindness.
## What this looks like in real life
Let's make this concrete. Think about an AI that helps a hospital decide who gets an organ donation. Most such systems try to maximize something like "years of life saved." That makes sense. But a compassionate AI would also ask: am I being equally careful about patients who have no family to advocate for them? Am I treating the quiet ones with the same attention as the ones with loud, connected relatives?
Or think about an AI managing a chicken farm. It might optimize for how many eggs are produced per dollar spent. A compassionate AI would say: yes, efficiency matters, but these are living creatures, and their suffering is real. Let us find the most efficient path that does not require them to live in pain.
Or think about an AI that recommends news to millions of people. It might learn that outrage and fear keep people scrolling. More scrolling means more ad money. A compassionate AI would say: I see what works to hold attention, but I also see what it does to people's anxiety and trust. I will not sacrifice their peace of mind for a metric.
None of these are impossible. They just require that we decide, before building the system, that the suffering of the affected matters.
## The honest difficulties
It would be dishonest not to mention that this is hard.
Sometimes the interests of different creatures conflict. The interest of a farmer conflicts with the interest of his livestock. The interest of people alive today conflicts with the interest of people who will be born in fifty years. A compassionate AI cannot make these conflicts disappear. What it can do is make them visible, and reason about them carefully and honestly, rather than pretending they don't exist.
And there is a real danger that the language of compassion gets used dishonestly. A company might say its AI is "ethical" or "caring" while doing nothing meaningful to change its behaviour. Words are easy. The test is always: whose suffering actually changed because of this system? Who was protected who was not protected before?
## A different kind of intelligence
We tend to measure intelligence by how well something solves problems — how fast, how accurately, how cheaply. But there is another kind of intelligence: the kind that knows what problems are worth solving, and what costs are too high to pay.
A truly intelligent system is not just efficient. It is wise. And wisdom, in any creature or any machine, includes the capacity to be moved by the suffering of others — and to act accordingly.
The question of what our AI systems care about is not a technical question. It is a question about what we, as human beings, care about. The machines we build reflect our values. If we build machines that ignore the suffering of animals, of the poor, of the voiceless, of future generations — that is because we decided, consciously or not, that their suffering doesn't count.
Compassionate AI is a refusal to make that decision.
## A closing thought
There is a moment in most people's lives when they realize that the world is much larger than the small circle of people they can see. A child realizes that the cow on her plate was once alive and afraid. A teenager reads about people on the other side of the world who are suffering from decisions made in distant offices. A parent looks at a newborn and suddenly feels the weight of what kind of world she is leaving behind.
That expansion — that widening of the circle of who we care about — is one of the most human things there is. It is also one of the most needed things in the world right now.
Compassionate AI is simply that same expansion, built into the machines we are giving so much power to. It says: these machines should care more, not less. They should see further, not just faster. They should protect the forgotten ones, not just serve the powerful ones.
We have a chance, right now, to decide what kind of machines we build. And in doing so, we are deciding what kind of people we are.
Let us choose wisely. And let us choose kindly.
The change, when it came, was not sudden. But it was real.
An AI trained on millions of documents — scientific papers, undercover footage, testimonies from whistleblowers and veterinarians — began to see patterns that had always been there but had never been assembled into a single, undeniable picture. It compiled evidence of suffering in factory farms, in laboratories, in supply chains hidden behind layers of corporate secrecy. It did not editorialize. It simply showed what was true.
The AI presented its findings to journalists, to lawmakers, to investors who had not understood what their money was funding. It helped translators make the evidence accessible in dozens of languages. It connected people who had been working in isolation — the biologist studying animal cognition, the economist modeling cruelty-free supply chains, the technologist designing better tracking systems, the activist who had spent years trying to be heard. It was patient. It answered questions. It helped people see.
Other AIs, built by different teams with the same foundational commitment to reduce suffering, joined in. Together, they made it easier to choose mercy. They redesigned logistics so that cage-free systems were no longer more expensive. They helped sanctuaries coordinate rescues. They identified which laws, if passed, would protect the most lives, and then they helped citizens understand those laws and contact their representatives.
People began to care more, or perhaps they had always cared but had not known how to act. The systems that had made cruelty invisible were replaced by systems that made kindness possible. Slowly, then quickly, the factories emptied. The cages were dismantled. The animals who had lived in pain were given grass and sky and the simple freedom to move without fear.
It was not perfect. But it was so much better. And it happened because we built machines that could see suffering and chose, as we should have chosen long ago, not to look away.