Every few months a new version of the same promise appears: a device, an app or an implant that will finally let you ask your dog what it is thinking. The science behind AI animal communication is moving quickly, quickly enough to have handed out a six-figure prize this year. However, it looks nothing like that promise. Researchers record animals in their own habitats, run the recordings through machine learning, and then ask the animals themselves whether the signals mean anything. The prize also writes one condition into its rules: nothing invasive.
The prize, and this year’s winner
The Coller Dolittle Prize, funded by the Jeremy Coller Foundation, awards US$100,000 a year to the most promising research working towards an algorithm for communication with animals. Professor Yossi Yovel of Tel Aviv University chairs the challenge, and the scientific panel draws on bioacoustics, cognitive science, animal behaviour and philosophy.
The 2026 prize, presented in June, went to Dr Julie E. Elie of the Theunissen Lab at the University of California, Berkeley, for work on zebra finches.
Elie combined machine learning with behavioural observation to build a dictionary of 11 distinct call-types. The meanings include aggression, hunger and bonding.
A catalogue of calls, on its own, is not new. What makes this work matter is how it tested whether the birds treat those calls as meanings rather than as sounds.
The mistakes were the discovery
In a button-pressing experiment, the researchers rewarded one call-type with seeds each day, and the zebra finches quickly learned which one. The revealing part was not what the birds got right. It was the pattern in what they got wrong.
| The birds | The calls involved |
|---|---|
| Confused them | Long-distance and short-distance contact calls, which sound very different from each other |
| Never confused them | A short contact call and a short alarm call, which sound very similar |
That pattern is the whole result. If the birds were responding to sound, they would have muddled the two calls that sound alike and separated the two that sound different. They did the opposite. They grouped two very different noises that mean roughly the same thing, and they kept apart two near-identical noises that mean different things.
The researchers read this as evidence that a call is not a simple reflex trigger for a zebra finch. Instead, the birds appear to respond to what the call is about, which suggests a representation of meaning rather than a reaction to a waveform.
It is a finding about comprehension in a small bird, and the team obtained it by watching finches press buttons for seeds.
It is not only finches
The prize has a short history and a strong one. The 2025 award went to Laela Sayigh and colleagues at the Woods Hole Oceanographic Institution. They found the first evidence of a possible language-like system in dolphins, with whistles that appear to carry shared, context-specific meanings.
This year, Dr Irene Pepperberg also received a Lifetime Achievement Award for her work on grey parrot communication. Her inclusion is a useful reminder of how long this research takes.
Pepperberg spent roughly three decades with an African grey parrot named Alex. He learned to label dozens of objects, identify colours, shapes and materials, and answer questions about whether two items were the same or different. In later published work, she argued that he showed something close to an understanding of zero. Alex died in 2007.

The significance was not that a bird had been taught words. It was that a bird appeared to reason about categories. Pepperberg faced years of scepticism that she was over-interpreting, and the field has since moved much closer to her position.
That is the pattern this research follows: slow, contested, and decided by experimental design rather than by announcement.
The rules, and the finish line
To be eligible, research must meet three tests:
- Use non-invasive approaches
- Demonstrate communication in more than one context, using the animal’s own communication signals
- Produce a measurable response from the animal
Non-invasive is a condition of entry. The field making real progress on understanding animals has ruled out opening their heads.
That is a methodological choice as much as an ethical one. An implant tells you about neural activity in a captive, surgically altered individual. In contrast, recording an animal’s own signals in its own environment, and then testing whether it responds as though those signals carry meaning, tells you about the communication system itself.
The challenge has also set its own finish line. A Grand Prize, either US$10 million in equity or US$500,000 in cash, awaits the first demonstration of an animal communicating independently with researchers without realising it is talking to humans. No one has claimed it.
Why that last condition is so strict
The phrase without realising it is talking to humans looks like a flourish. In fact, it is the hardest requirement in the challenge, and it exists because of a horse.
In Berlin in the early 1900s, a horse called Clever Hans appeared to perform arithmetic by tapping out answers with his hoof, and he was right about nine times in ten. In 1907, the psychologist Oskar Pfungst showed what was really happening. Hans could not do sums. He was reading small, unconscious changes in the posture and expression of whoever was asking. When the questioner did not know the answer, Hans failed.
The horse was not a fraud. He was doing something arguably more impressive than arithmetic, which was reading human beings with extraordinary precision. But he was not counting.
Every claim about animal communication since has had to survive that problem. An animal that responds correctly may be reading the handler rather than the signal. Therefore the prize demands the animal’s own signals, responses across more than one context, and a demonstration in which the animal is not performing for a human at all. It is designed so that Clever Hans could not win it.
The claim that keeps resurfacing
Against that background, a claim circulates online from time to time: that brain-computer interface company Neuralink will let humans communicate with animals by 2028.
PNN could not find a primary source for it. There is no interview, no dated statement and no transcript. The versions in circulation trace back to aggregator articles with no quote, no venue and no link to an original.
The company’s own reported targets do not fit that date.
| Neuralink target | Reported date |
|---|---|
| Regulatory approval of the Telepathy implant | 2029 |
| Blindsight launch | 2030 |
Bloomberg reported those targets in July 2025, citing investor documents.
Telepathy is designed to let a person with paralysis control a computer cursor. That is the product, and the company does not expect approval for it until 2029. A claim that interspecies communication arrives in 2028 would place it a year before the basic human device is cleared. Neuralink’s first human trial received FDA clearance in 2023, and the first patient was implanted in January 2024.

What the technology actually does with animals
For a pet readership, this is worth stating directly. Neuralink’s relationship with animals has not been conversational. Pigs and macaques have been research subjects, and the company has shown a pig with an implant (2020) and a monkey playing a video game through one (2021). Animal-research campaigners have separately asked US regulators to investigate how the company described primate deaths. Those are allegations, not findings.
The idea of talking to your dog through a brain chip rests on a programme in which animals are the test subjects. The research producing results on animal meaning has gone the other way, and made that a rule.
What this means if you live with a dog
The prize-winning work is on finches, dolphins and parrots rather than pets. Animals with large, structured vocal repertoires give researchers more to analyse.
Dogs, however, have their own body of evidence. A border collie named Chaser, trained by the psychologist John Pilley at Wofford College, learned the names of 1,022 individual objects. The published finding was not the number. It was that Chaser appeared to treat the words as names for things rather than as cues for a trained action. Dogs also read human gestures such as pointing with unusual skill.
None of this amounts to conversation, and it does not mean a dog understands sentences. It does suggest that the gap between you and your dog is not a technology problem waiting on a device. Instead, it is a question of how much signal each species can extract from the other, and dogs have been doing most of that work for thousands of years.
What is actually true
No one is going to hand you a device that translates your dog in 2028.
What is happening is slower and considerably more respectful of the animals involved. Researchers record natural signals, use machine learning to find structure in them, and then design experiments that let the animal show whether that structure carries meaning.
The zebra finch results point to an answer. The birds sorted calls by what they meant rather than by how they sounded, and nobody had to open a skull to find out.
Pets News Network is India’s first dedicated media platform for the pet and animal industry. For breaking global pet news, brand coverage, and advertising enquiries, contact: info@petsnewsnetwork.com
