When We Can Hear Them
AI could help us understand animals’ calls. But better insight does not decide how we should treat them — or who that understanding should benefit.
In a research project, researchers collected and analysed 7,414 calls from 411 pigs. The recordings covered different situations from birth to slaughter. The researchers investigated whether a machine could distinguish between calls from situations they judged to be positive or negative for the animals.
It could, with considerable accuracy in the experiment.
The researchers did not translate the grunts into Danish sentences or obtain answers about what the pigs thought of their lives or their approaching deaths. But the study, involving the University of Copenhagen, showed that more information could be extracted from the animals’ calls. It opens up the possibility that, in the longer term, AI could help us monitor pigs’ welfare more closely.
Some of the animals AI could help us understand are also the ones we eat.
There is something strange about following research into how we might understand a being better when that same being is also something we produce for food. We can be excited by a technological possibility and suddenly discover that it leads into a question better technology cannot answer.
The pigs in the recordings did not acquire lives only when the algorithm found a pattern in their calls. They already had them. What AI can change is our access to understanding the animals’ lives.
Who is there in the sound?
A person unfamiliar with a language can hear a conversation without understanding a single word. We nevertheless know that the sounds mean something to those speaking. Something is happening between them that we cannot access.
It is tempting to carry that experience out into nature. Whales click, birds call, pigs grunt. What are we missing?
The comparison has a limit. We cannot assume that every animal sound contains a sentence simply waiting to be translated. Communication can concern contact, danger or a bodily state without resembling human language. Other animals also live through senses and relationships that a human translation could easily impoverish.
Even so, the research shows that there may be more to investigate than our first listening reveals. In a study from 2025, researchers used machine learning to classify positive and negative contact calls in seven species, including cows, sheep, pigs and goats. Results varied between species, and the researchers called for further testing. The model did not measure emotions directly. It identified acoustic differences between calls the researchers had associated with positive or negative states on the basis of the situation and earlier behavioural studies.
Classifying a call is therefore different from measuring an emotion directly. A machine can give us an indication of how an animal is doing without giving us the whole explanation. But we do not need the whole explanation before such an indication becomes valuable. If a call can help us detect poor welfare, that is different from merely hearing noise from a barn.
In AI Finds the Connections We Overlook, I wrote about the machine’s ability to find patterns and test connections that people had not previously seen together. Here, that possibility comes close to home. The pattern we discover may be an indication of how another living being is doing.
An ocean full of something we do not understand
Project CETI investigates sperm whale communication using AI among other methods. In an analysis of 8,719 click sequences, researchers found a rich interplay of rhythm, tempo and variations that depended on the surrounding sequences. The analysis revealed more structure than earlier descriptions had captured.
The meaning of many of the sequences is still unknown. A pattern in sounds is not a dictionary, and a dictionary would not, in itself, be a conversation either. The work also rests on many years of observations and recordings. The AI models analyse material that biologists have collected and described through long-term fieldwork.
Still, imagine that one day we get further and can connect particular exchanges with something the whales do together. If that connection also holds when researchers test it, a sound recording will take on a different character for us. We will not just hear that whales are present; we will also be able to begin following some of what happens between them.
It need not end with a whale answering our questions. The discovery of just how much we have not understood may itself broaden our view of the ocean.
We search for signs of intelligent life beyond Earth and devote enormous effort to building intelligence in machines. Meanwhile, we are surrounded by living beings whose communication we can only partly access. It would be a remarkable detour if the work on artificial intelligence made more of them understandable to us.
The animal and the meat
Sperm whales are far removed from most people’s everyday lives. Pigs are close in a different way: we can encounter them as animals in research recordings and as meat in the supermarket’s refrigerated display.
There need not be any secret between those two images. We know where pork comes from. But we can know that connection without thinking about the pigs’ lives when we choose dinner. The recordings and the shopping bring different things into the foreground.
If AI makes us better at understanding how pigs are doing, that connection may become harder to push into the background. That is one possible consequence. There is also reason to ask whether we will instead find new ways to keep it at a distance.
In three psychological experiments published in 2012, Brock Bastian and his colleagues investigated the tension between eating meat and not wanting to harm beings with mental lives. Participants could attribute fewer mental capacities to animals when the animals were associated with food and impending meat consumption. The researchers interpreted this as a way of reducing the internal conflict.
The study is not about AI and does not tell us how all meat eaters think. But it challenges the comfortable notion that more knowledge automatically leads to more consideration. We are also capable of adjusting our beliefs to what we already do.
A technology that makes animals more understandable therefore encounters people with habits, meals, economic interests and ideas about which lives count in which ways. Some will change their diets. Others will demand better conditions for animals or argue that good welfare and meat eating can be reconciled. Being able to understand more does not settle that disagreement.
Better insight into animals’ lives may, however, change the basis of that disagreement. An abstract concern for animals may become more concrete if we can better follow their reactions to the conditions we give them.
Those who understand can also control
The pig research itself points towards a practical application: helping farmers monitor their animals’ welfare. If they obtain a system that can warn of poor welfare, problems may become easier to detect and address. Better conditions would be a real improvement for the animals, even if they continued to be raised for slaughter.
The same insight can therefore both challenge meat production and become part of it. Some may use knowledge of animals’ signals as a reason to stop eating meat. Others may use that knowledge to document that a particular product comes from animals kept in better conditions. Knowledge of animals can change production methods without changing the purpose of production.
If we also become better at producing the signals ourselves, we gain another form of influence. A sound we play to test a hypothesis may be a real message to the animal hearing it. The animal may respond by looking for the source of the sound or becoming unsettled, even though we regard the playback merely as an experiment.
The philosophers Marriah Alcantara and Kristin Andrews investigate precisely that difference in a paper on research ethics and AI-based animal communication. Better understanding can help animals, but recording and playback can also disturb them and their relationships. The researchers also raise the question of whether animals may have an interest in privacy that could become harder to protect if we gain access to their communication.
Animals communicate with one another. Their messages are not necessarily addressed to us.
We can discover something we ought to take into account without thereby acquiring the right to use our knowledge as we please. Listening is not the same as being invited. And knowing an animal’s responses may make us better at helping it or better at getting it to do what we want.
They do not have to pass our test
We should not require a grunt to be translated into an articulate protest before we take the animal into account. Otherwise, we once again make our own way of existing the yardstick for everyone else.
There is a difference between being able to communicate, being intelligent and being able to experience pain or pleasure. A being can have something to lose even if it cannot explain the loss to us. If AI makes only some species understandable, that does not mean the others are empty.
In The Last Insult, I explored what happens to humanity’s self-understanding when our intelligence can no longer serve as proof of superiority. Animal communication takes the question in another direction. We do not need animals to surpass us before we have reason to reconsider our relationship with them. Understanding a little more of their lives may be enough.
That development may arrive without a single great breakthrough. A better indicator in the barn. A previously overlooked connection in whales’ clicks. Small shifts in what we can perceive and what we must then consider.
So we need not imagine that one day the pigs will say something no one can dismiss and the world will then change. They already make sounds, and we have knowledge of their behaviour and needs. AI can give us more to work with, but the choice of how to use that knowledge remains ours.
The 7,414 calls were collected in a study covering situations from birth to slaughter. The algorithm could distinguish between the sounds. That result does not, in itself, say anything about whether slaughter should take place.
If the next generation of tools makes us even better at understanding animals, it will be worth watching whether anything also changes beyond the recordings. In the barn. In the refrigerated display. In what we accept.
Research continues to provide new answers to the question of how much AI can help us understand animals. What we do with that understanding is a question we must answer ourselves.