The promise and perils of using facial recognition technology on animals.By Laura Bridgeman Jun 12, 2021, 8:00am EDT
Share this story
- Share this on Twitter
- gnition technology is rapidly becoming ubiquitous, used in everything from security cameras to smartphones. But in the near future, humans may not be the only ones to be digitally captured. Researchers are training forms of artificial intelligence to recognize individual animals by their faces alone — and even discern their emotional state just by reading their expressions.
Much of the research into animal facial expressions has focused on species like dogs and horses. But some of the most cutting-edge work is aimed at an unlikely subject: the farmed hog.
The typical hog factory farm employs a small number of workers to oversee hundreds, or even thousands, of pigs — too many for the people running the facility to tell which ones might be in distress. Researchers at the Centre for Machine Vision at the University of West England, where pig emotion recognition work is being conducted, envision this technology could be used to help farmworkers more readily identify sickness and injury. If AI can routinely scan the pigs’ faces and alert workers to particularly stressed-out animals, treatment can come sooner and suffering can be reduced.
Twice a week, we’ll send you a roundup of the best ideas and solutions for tackling the world’s biggest challenges — and how to get better at doing good. Sign up here.
There is even potential for the technology someday advancing to the point of detecting “happiness” in pigs — a holy grail for animal ag.
But while the idea of learning more about what animals are feeling is self-evidently enticing — why wouldn’t we want to learn more about them? — some animal welfare advocates question the very premise of this research. While the bulk of the funding is from the UK government, one reason for the skepticism is that the research is partly supported by companies in the meat and agriculture industry, including a pig genetics company that has availed its farms for the study. It’s not hard to see that industry’s interest in this work: Keeping more pigs alive under intensive conditions would be a financial boon, as would being able to advertise how “happy” the animals were — something the Centre’s website suggests could be possible.
And that all leads to a deeper question: Just how comfortable — let alone happy — can a pig be on a factory farm? In the US, nearly all pigs raised for meat are kept in unnatural, highly mechanized, and crowded conditions, given no access to the outdoors. Conditions are similar in much of the European Union, and factory farming is on the rise in low- and middle-income countries as global demand for meat increases. These environments are so difficult to endure that, by some estimates, up to 35 percent of US-raised pigs die before ever reaching the market.
The project of discerning the emotional state of pigs — and the meat industry’s larger push to invent new technology that promises to improve animal welfare — illustrates the fine line between meaningful efforts to reduce animal suffering and so-called “humane-washing,” where animal welfare is portrayed as being better than it actually is.
There is a growing body of research that shows what changes farms could make today to reduce the suffering of farmed animals, like eliminating extreme confinement, ending breeding practices that make animals grow too big too fast, and providing outdoor access and enrichments designed to mimic experiences they would normally enjoy if left to their own devices. All of which raises the question: Who is this new technology really for — the pigs, or the humans who raise, slaughter, and eat them?
How to identify stressed-out pigs
The most cost-effective methods of raising animals tend to cause the most harm. Animals’ bodies become levers on which a balancing act is performed: expending the fewest resources (such as living space) while keeping animals alive and productive. Economic considerations often outweigh welfare, resulting in the inhumane conditions that are a hallmark of intensive animal agriculture.
On the side of animal well-being are researchers like Melvyn Smith, director of the Centre for Machine Vision, for whom improving animal welfare is a big motivator in his quest to use AI to identify stressed-out pigs. “If we could understand how the animal is feeling, if the animal can tell us this itself, then that gives us an opportunity to tailor treatment and care for individual animals,” he told me.
To try to understand how an animal is feeling, he and his colleagues, in partnership with Scotland’s Rural College, are building on past facial recognition research. They have already trained a form of deep-learning AI that is tailored to analyzing images, known as a convolutional neural network (CNN), to distinguish between individual pigs just by analyzing photos of their faces.
This new project — aimed at recognizing emotions — adds a layer of nuance to this research by training the CNN to recognize the difference between stressed and unstressed pigs.
THE MOST COST-EFFECTIVE METHODS OF RAISING ANIMALS TEND TO CAUSE THE MOST HARM
Like other deep learning algorithms, the Centre’s CNN learns by being exposed to data sets — in this case, thousands of photographs of pig faces that are likely to be experiencing stress or not. Cameras affixed just above the water spigot where pigs drink allow for close-up and relatively uniform images of each pig every time they take a sip. The CNN then analyzes each photograph, searching for subtle variations in the pigs’ faces around the eyes, the position of the ears, and other features.
To observe whether pigs are stressed, the animals are placed in situations known to be either mildly stressful or preferable. Pigs kept in pens with multiple generations tend to experience stress (particularly true of younger pigs), whereas relatively stress-free environments can be created by giving pigs essentially an all-you-can-eat buffet. Saliva and blood can be measured to determine cortisol levels, a chemical associated with a stress response.
With the three-year project about halfway complete, the results so far are impressive: The CNN is able to distinguish between pigs’ stressed and unstressed facial expressions more than 90 percent of the time.
By helping AI recognize expressions related to core emotional states in pigs, farmworkers could be alerted to individuals that are experiencing discomfort, allowing for swifter medical attention or alterations to the pigs’ living environment.
Caring for farmed animals as individuals is becoming increasingly difficult due to intensive animal agriculture operations. On smaller-scale farms, workers are able to spend far more time with individual pigs, getting to know animals’ personalities and watch out for suggestions that they may be unwell. But most pigs live on factory farms, where just a few workers can be responsible for the care of thousands of animals.
And factory farms are ramping up around the world: In the US, where factory farming has become the norm for animal agriculture generally, nearly 130 million pigs were raised and slaughtered in 2019 alone. The UK saw intensive pig farming increase 26 percent between 2011 and 2017. In China, a “hog hotel” factory farm consisting of a collection of buildings reaching 12 stories into the sky clocks in as the biggest multi-story hog farm on the planet, with the capacity to house upward of 1,000 pigs per floor.
It is no easy task to keep pigs alive within the crowded indoor conditions of factory farms. According to the Iowa Pork Industry Center at Iowa State University, about one in three pigs die before reaching the market due to factors like stillbirth, sow crushing, infectious diseases, and poor air quality. Not only does this figure represent massive economic losses for the industry, it also demonstrates the sheer scale of health problems pigs on factory farms must regularly contend with, many of which can cause chronic physical and psychological pain even when they are not ultimately fatal.
Identifying negative emotions like stress could help reduce the suffering of farmed pigs. But the research won’t end there: The next goal is detecting subtler emotions, including happiness.
Can animals have a good life on a factory farm?
Interest in animal facial expressions seems to be growing within the scientific community. Facial coding systems are being developed for species like horses and dogs, where expressions related to pain or frustration are being mapped out. Dogs have been observed to make “cute” faces at humans, while rats and chimps are perceived to smile and laugh when they are tickled.
But is happiness something that can be measured by facial expression?
Smith’s team wants to find out. Once the current study on pig stress is complete, the next stage will be seeing whether the CNN can detect other, more nuanced emotions, perhaps one day giving “farmers and their prospective customers an idea of how happy their pigs are,” as the Centre’s website notes.
But technology capable of detecting happiness and more subtle or complex emotions is not without controversy. When it’s applied to human beings, critics warn of the inaccuracies arising with a one-to-one mapping of prototypical expressions to emotions. A scowl doesn’t always mean anger; a furrowed brow doesn’t always denote concentration.
Further complicating the matter is that happiness is a philosophically elusive concept even when it comes to Homo sapiens, since there remains a lack of consensus over what exactly constitutes happiness. Fleeting moments of pleasure, joy, or contentment, along with longer-term experiences of an engaged, meaningful life, are thought to be among the ingredients associated with states of happiness in people.
While the constituents of happiness probably look different depending on the species, certain conditions are more likely to guarantee the suppression of happiness regardless of the kind of animal.
“Pigs can never be happy in factory farms,” says Lori Marino, director of the Kimmela Center for Animal Advocacy and an expert in animal behavior who co-authored a study on pig cognition and emotion. To Marino, “a CAFO [concentrated animal feeding operation] is so far from what a pig needs to thrive that it could not be a place that would make a pig happy or content. They are not designed for pig happiness.”
“I also worry that these companies will only share data that are self-serving and the data will be biased toward convincing people that pigs are happy in CAFOs,” she continued.
These concerns may be well-founded. People and businesses that use animals often state that the animals under their control are happy, like the California Milk Advisory Board’s “Happy Cow” campaign or Elon Musk’s “totally happy” lab monkey.
Such claims of animal happiness can be dubious given the mounting science revealing the extent to which animals can be harmed in captivity. One of Marino’s other studies looks at how captivity can cause brain damage in some animals, impairing cognitive functions such as memory and decision-making.
Other researchers conducted a study that found horses that were confined within stalls emitted brain waves associated with states like depression and anxiety, whereas horses allowed to roam in herds on pastures showed brain waves associated with feelings of calm. Pregnant pigs kept in gestation crates, cages that are barely bigger than their bodies, are known to become unresponsive over time — behavior that has been linked to depression. Much is already known about the emotional state of animals in captivity without state-of-the-art tech telling us.
How Iowa’s largest hog producer courted power, turned farming into a numbers game, and transformed the American heartland.
Smith’s inquiry into whether pigs are happy on farms may find they’re not, but that doesn’t deter him. He says he is interested in switching the longstanding emphasis within the animal research community from detecting simply an absence of negative emotions to detecting positive emotions, and that this might lead to a better understanding of what contributes to higher quality of life and happiness for pigs.
But given that the current project is partly supported by industry stakeholders, including the farming technology company AgSense (owned by Valmont Industries), JSR Genetics Ltd. (a pig breeding company), and Garth Pig Practice (a veterinary consulting service), skepticism about the uses of this technology is in order. (AgSense, Valmont Industries, and Garth Pig Practice did not respond to requests for comments for this article.)
Moving the needle on animal welfare
The intensive animal agriculture industry is facing increasing scrutiny of operations that not only harm animals but give rise to a litany of damaging consequences, from perpetuating environmental racism — especially in North Carolina, where hog farms disproportionately pollute predominantly Black communities — to accelerating climate change. Demands to abolish factory farming altogether are growing louder.
Still, improvements in farmed animal welfare are worthwhile since it’s unlikely factory farming is going away anytime soon. Once implemented, the Centre’s CNN may quantifiably improve the welfare of pigs on factory farms, even if incrementally.
But while there’s still much to learn about animal welfare, there’s even more that we already know. If the pork sector were concerned with animal thriving, practices known to cause chronic stress — such as gestation crates — would already be eradicated.
There exists abundant evidence of the pain male piglets endure when they are castrated without anesthesia, yet these mutilations continue to be widespread.
Confining pigs indoors within crowded, barren pens on concrete flooring can lead to abnormal biting behaviors that can devolve into cannibalism — something that can be addressed by giving pigs additional space and covering floors with natural materials like peat or compost.
AI technology may one day yield deeper insights into farmed animals’ emotional states. And there’s some genuine value in research diving into what animals are feeling. The question that looms over the use of such tech in a factory farming context is whether we already know enough anyway.