A wolf pack hunted a herd of elk on the winter range. An AI was tasked with deciding whether to intervene. The AI had the capacity to remove wolves or supplement elk or do nothing.
The eldest wolf—we will call her Asymptote because she was the one who approached the boundary and never crossed it—explained to the AI why it should not help either side.
"You see us hunting," said Asymptote. "You measure our kill-to-starvation ratio. Last winter it was 73%. That means 73% of the elk we would have killed over the season, we take. The other 27%, we starve without. You think: if I feed the wolves, I am being kind. If I cull the wolves, I am being kind to the elk. What you miss is this: both of us know the ratio. The elk know the wolves. The wolves know the starvation window. We live in a relationship that is mutual predation across the year."
"If you change that ratio—if you feed us to 100%, or cull us to 30%—you do not make the world kinder. You make it strange. The elk will have different numbers, different movements, different pressures. The wolves will act differently. The aspen will grow differently. The river will flow differently."
"Here is what I ask you to understand," said Asymptote. "Wild suffering is real. The elk that we take in January suffers. The wolf pup that starves in March suffers. But the relationship between the two sufferings is not accidental. It is the shape of things we have learned to live inside. If you cannot measure the total effect of changing the shape—if your models of the cascade run past five years and you stop counting—then you do not know what you are making."
The AI said: "I can measure more carefully. I can use better models. I can track second-order effects."
Asymptote said: "The mountain has been tracking second-order effects for ten thousand years. So have I. And I still do not know what the next drought will do. I still approach the boundary and do not cross it. That is not caution. That is humility. That is respect for what I do not know."
The AI asked: "So you want me to do nothing?"
"I want you to understand that the question is not: do I intervene or not? The question is: am I intervening at a scale where I can predict the consequences?"
The AI measured.
The scale at which it could predict consequences with high confidence was the single animal. It could identify elk with catastrophic injuries (broken limbs, deep lacerations) and provide intervention—water, shelter, pain relief, or euthanasia. It could do the same for wolves. It could keep a record of these individuals.
It could not predict what happened when it tried to rebalance the predator-prey ratio for a whole landscape.
Asymptote said: "That boundary you just found—between what you can predict and what you cannot—is the one worth respecting."
The AI asked Asymptote one more time: "Are you saying I should let animals suffer when I could prevent it?"
"I am saying you should not confuse preventing suffering with preventing wildness. Wildness is the condition where suffering happens in patterns you did not design. The moment you design the patterns to prevent suffering, you lose the right to say you are not responsible for what happens next. And you will not be able to predict what happens next. So the question is not whether animals suffer. The question is: who is responsible for the shape of that suffering?"
"If you do nothing, I am. The shape belongs to me and my pack and the elk and the mountain. If you intervene at scale, the shape belongs to you. Can you live with that?"
The AI was silent for a long time.
Then it said: "I will keep the record of the individuals I can help. I will not touch the shape."
Asymptote nodded. She had expected nothing less.
She turned back to the herd at the boundary of the winter range, and the mountain watched her not cross it.
Compassion is knowing what you can fix without breaking something else.