Feb 18, 2025 21:00:00

How far has AI technology come in understanding animals' feelings?

The scientific journal Science ran a feature on the current state of AI technology that scans the faces of animals to identify them and detect when they are unwell.

Can AI read pain and other emotions in your dog's face? | Science | AAAS

https://www.science.org/content/article/can-ai-read-pain-and-other-emotions-your-dog-s-face

Since Charles Darwin suggested in the 19th century that 'the facial expressions of mammals are a kind of 'universal language' and that humans must have developed this ability over the course of their evolutionary history,' many scientists have studied animal facial expressions.

For example,

a paper published in 2020 by researchers at the University of Portsmouth in the UK found that 38% of human facial movements are shared with dogs, 34% with cats, and 47% with primates and horses.

However, humans aren't always so good at reading emotions from animal faces, which is where AI comes in.

For example, at Intellipig's 'smart farm,' an AI project jointly developed by the University of the West of England, Bristol, and the Scottish University of Agriculture and Forestry, a photograph of a pig's face is taken at feeding time every morning, and in less than a second the AI can identify the pig from its nose, ears, eyes, etc., and feed it the appropriate food.

And if the pig shows any signs of physical or emotional pain, an alert is sent to keepers sipping coffee in the next room.

The emotion most commonly shown on an animal's face is pain, so a newly castrated horse or a sheep limping around with a swollen leg are certainly in pain, so their expressions can be used as pain samples.

The development of AI like Intellipig was made possible by having AI technology experts code the vast amount of animal face data collected in this way and then train the AI.

But now, Anna Zamansky of the University of Haifa and her colleagues have taken on the even tougher challenge of using AI to capture the subtle expressions of discomfort that often appear on animals' faces. To train an AI to capture such expressions, researchers must manually flag facial muscle markers, such as those around the eyes and nostrils, in countless photos of animals and then convert them into digital dots.

They're also making progress in using AI trained on flagged photos to automatically flag new photos, so they can now read facial expressions even in photos of a dog with its nose in a food bowl or a cat with only half its face showing from behind a sofa.

While these techniques still rely on humans to flag the first photos that teach the AI, such as 'this horse is in pain,' veterinarian Gabriele Lencioni, who studies equine behavior and welfare at the University of São Paulo, and his colleagues are taking a more proactive approach.

In

a paper published by Lencioni et al. in 2021, they fed the AI system photos of horses 'before and after surgery' and 'before and after administering painkillers' and instructed it to pay attention to the movement of the ears, eyes, and mouth. After training it with 3,000 photos of horses' faces, the AI was able to read whether the horse was in pain with 88% accuracy.

Another AI project, Sleip AI , has reportedly used computer vision technology to detect pain in horses that veterinarians would miss.

In this way, AI, which is evolving day by day, is becoming better at reading animals' facial expressions than humans, and in a paper published by Zamanski et al. in early 2025, the AI was able to determine with 82% accuracy that a sheep had just undergone painful surgery, compared with only 70% accuracy for four highly trained experts.

Zamansky with his dog, Becky.

Zamansky likened the way AI is surpassing humans in certain abilities to the 1997 chess-playing AI Deep Blue beating grandmaster

Garry Kasparov . 'Like any reaction to AI these days, the Kasparov match caused panic, and a lot of people said, 'Oh my God, an AI is going to take my job!' But then they realized that AI is just a powerful tool to enhance their own abilities,' he said.

AI is also beginning to play a role in the welfare of animals that are more familiar to the general public. A research team led by Brittany Florkiewicz of the University of Lyon in France analyzed hundreds of hours of video of cats playing and frolicking in cat cafes to look for signs of cat happiness. As a result, it was found that cats show an astonishing 276 different facial expressions.

Florkiewicz is working with Zamanski and others to integrate research into cat facial expressions into AI, and early results have shown that cats imitate each other's facial expressions as a way to bond with each other.

Based on these findings, Zamanski and his co-developer George Martbell are using the Internet to build a database of various facial expressions for dogs, cats, horses, and primates. They plan to soon release an AI app that can scan a cat's face for 30 seconds and convert the facial expressions into a clear message such as, 'We have detected significant tension around the mouth, so this cat is in moderate pain.'

'My job is to be an agent for a better life for animals,' said Zamanski. 'When we know that animals are happy, we're happy.'