The first thing I ever said to my dog was, “Would you like to come home with me?” He was six pounds and ten weeks old. He leaned his head forward and sniffed my mouth.
In the four years since, I have continued to bombard him with questions he cannot answer. I ask him what he’s up to, if he wants to go for a walk, if he feels sleepy. When he is sick, I ask him what is wrong; when another dog growls at him, I pull him aside to ask if he’s okay. He does what he can to communicate his thoughts to me: he barks; he sighs; he scratches at the door.
But of course we never spoke to each other, not really. Some 15,000 years after humans first domesticated the wolf, scientists have learned that different types of barks mean different things (for example, dogs use a lower, longer bark for strangers), but our understanding of canine communication remains fairly limited. (Researchers call it cautious communicationnot languagebecause no animal has been shown to have the same complexity of verbal systems as humans.)
While barking at a squirrel is easy enough to decipher (I’ll eat you!), people have more trouble determining whether a whine is just a dog having random feelings on Tuesday, or something much more serious. Dog owners often joke about how they would give up years of their lives for the chance to talk to their pet for an hour or a day. Meanwhile, peddlers posing as dog whisperers and pet psychics have happily taken their money by claiming they can help them translate their dogs’ inner thoughts.
Now, amid a wave of broader interest in artificial intelligence applications, some dog researchers are hoping AI can provide answers. In theory, the technology is very suitable for this. AI is essentially a pattern recognition machine. ChatGPT can respond in language that appears human because it is trained on massive writing datasets, which it then mimics in its responses. A similar premise applies to other generative AI programs; large language models identify patterns in the data they receive, map the relationships between them, and produce output accordingly.
Researchers work with the same theory when it comes to dogs. They send audio or video of dogs to a model, along with text descriptions of what the dogs are doing. They then see whether the model can identify statistical patterns between the animals’ observed behavior and the sounds they make. In fact, they try to ‘translate’ barking.
Researchers have been using similar approaches to study communication between dogs since 2006, but AI has recently become much better at processing massive amounts of data. However, don’t expect to discuss Immanuel Kant’s philosophy with Fido over coffee anytime soon. It’s still in its infancy, and researchers don’t know what kind of breakthroughs AI could deliver – if at all. “It has enormous potential, but the gap between potential and reality is not yet fully visible,” Vanessa Woods, an expert on dog cognition at Duke University, told me.
Right now, researchers have a big problem: data. Modern chatbots are trained on large collections of text – trillions of words – giving them the illusion of language fluency. To create a model that can translate, say, dog barking into English (if such a thing is even possible), researchers would need millions, if not billions, of neatly cataloged fragments. This barking should be thoroughly labeled by age, breed and situation, distinguishing a 10-year-old male labradoodle barking at a stranger, and a six-week-old bichon frize puppy playing with his littermate.
Such a catalog does not currently exist. This is one of the great ironies of the project: dogs are all around us, constantly captured by phones, doorbell cameras and CCTV. You don’t have to look Planet Earth to see the dog living in its natural habitat; the internet is filled with more clips of dogs than anyone could watch in a lifetime. And yet all of this media has never been cataloged in any serious way, at least not on the scale that would be necessary to better understand what their barking means.
Perhaps the best catalog in existence comes from researchers in Mexico, who have systematically recorded dogs in their homes in specific situations, getting them to bark by, for example, knocking on a door or squeaking a favorite toy. A research team from the University of Michigan took some of the 20,000 recordings in the dataset and fed them into a model trained to recognize human speech. They barked at the model and then had it predict what they were barking at based on sound alone. The model was able to predict with an accuracy of about 60 percent which situation preceded the barking. That’s far from perfect, but still better than chance, especially considering the model had more than a dozen barking contexts to choose from.
The same approach of using AI to decipher dog barks is happening with other animals. Perhaps the most promising work is whale chatter, as my colleague Ross Andersen has written. Other researchers focus on pigs, bats, chimpanzees and dolphins. One foundation is offering up to $10 million in prize money to anyone who can “crack the code” and have a two-way conversation with an animal using generative AI.
Dogs are unlikely to be the animals that help scientists win the prize. “I don’t think they necessarily use words, sentences, paragraphs,” Rada Mihalcea, co-author of the Michigan study, told me over Zoom. (Of course, in the middle of our conversation, a stranger knocked on my door, causing my foster dog to start barking.) As much as dog owners like me would like something akin to Google Translate for dogs, Mihalcea starts with much more limited ambitions. She hopes that this line of research “can help us understand what exists as a language system in the first place – if such a system exists.”
Another research group, led by Kenny Zhu of the University of Texas at Arlington, is taking a different approach. His team pulls huge amounts of dog videos from YouTube. But the data is extremely noisy – literally. The researchers have to isolate the barking from all the other sounds happening in the background of the videos, which makes the process tricky. Zhu’s team has preliminary findings: They ran their algorithms through the sounds of six different breeds (huskies, Shiba Inus, pit bulls, German shepherds, Labradors and Chihuahuas) and believe they found 105 unique phonemes, or sound units. covering all races.
Even if researchers are eventually able to obtain a perfect data set, they will encounter another problem: There is no way to be sure that whatever observations the AI makes are accurate. When training other AI models in human languages, a native speaker can verify that an output is correct and help refine the model. No dog will ever be able to verify the AI’s results. (Imagine a dog sitting in an academic research lab and nodding solemnly: Yes, that’s right. ‘Ruff-ruff’ means ‘Give me the chicken’.The dream of AI as an intermediary between humans and dogs faces a fundamental bias: it is human researchers using human-made AI models and human language ideas to better understand dogs. No matter how good the technology becomes, there will always be uncertainties.
The focus on better understanding dogs’ verbal sounds can obscure how much we already know about them. Dogs have evolved to communicate better with humans: their barking has changed and their eyes have become more expressive. Wild dogs and wolves bark less than domestic animals, suggesting that humans are a major reason why our pups make noise. “The whole thing about dog geniuses is that they can communicate with us without talking,” Woods told me. “We can also read them very clearly, which is why we are so in love with them.”
I know what she means. During a heat wave this summer, I decided to buy heat-resistant dog boots to protect my pup from the scorching sidewalk. You put them on by stretching them over your dog’s paws and snapping them into place. The first time I put them on my dog, he stared at me. When I tried to take him out later that week, he was thrashing around in the grass and running around chaotically. He didn’t want to wear the boots. And I didn’t need an AI to know that.
