The Dog Doesn’t Need a Voice
Neuralink, the second body and the mistake of imagining animal communication as human language
A post circulating this week makes a remarkable claim: Elon Musk says Neuralink will allow humans to communicate with animals by 2028. The screenshot offers a tidy explanation of how this might work: read neural signals, use algorithms to interpret them, translate intentions, and turn them into a conversation.
And I am reminded of Thomas Nagel’s question about what it might be like to be a bat, as another social post - this time of an interview with Nagel - popped up on my instagram feed during the week. One of the early essays I had to read as a first year Philosophy student at University. And in all the work I currently do around re-perceiving reality, it is one I never forgot.
The original post gives no link to Musk actually saying the above, and no Neuralink announcement and no named study. Its headline states a deadline; its caption retreats to “could eventually”. I can find no primary source for either Musk’s supposed 2028 promise or a Neuralink animal-communication programme. But Neuralink’s publicly described human research is aimed at decoding intended movement so people with paralysis can control computers and other devices. That is extraordinary work. It is not a system for translating one species’ experience into another’s language.
However, The claim is compelling because it folds several real developments into one imagined breakthrough. Brain-computer interfaces can record neural activity and use computational models to associate patterns with particular intentions. AI can help researchers process huge collections of animal sounds, images and behavioural data. Neuralink has also conducted research involving animals. Put these facts together and it is easy to picture a universal translator arriving next.
A neural signal is not a sentence
Yet a brain signal is not a sentence waiting to be read. Neural activity has to be interpreted in relation to a task, a body, an environment and a particular individual. Neuralink’s current human use case is comparatively bounded: identify an intention, such as moving a cursor, and map it to an action. Understanding a dog would demand a far more ambitious chain of inference. Researchers would have to establish what a neural pattern means to that dog, how it relates to the dog’s sensory world and behaviour, and whether any proposed translation corresponds to what the dog is communicating.
A dog’s world is not ours
That last step is where the human fantasy of “talking to animals” can get ahead of the science. We hear the word communication and picture words, questions and answers. But communication does not have to be organised around language as humans know it. Dogs read and respond to posture, gaze, movement, sound, scent and context. Research on canine emotion perception has found that dogs attend more to bodies than to faces, while humans tend to look first at the head.
Smell is not a side channel in a dog’s world. It is one of the principal ways the world becomes available to them. A review of canine olfaction describes sniffing and searching as natural, species-typical behaviour, important to dogs’ welfare as well as their perception.
The mismatch is not simply that dogs lack our words. It is that we may be asking the wrong question when we ask how to make them speak ours. A dog’s communication is expressed through an embodied relationship: where it turns, what it approaches, what it avoids, how it moves, what it smells, and how those actions change in response to another being. A device that outputs “I’m hungry” or “I’m sad” might be useful in some circumstances. But unless those labels are grounded in careful observation and validated against the dog’s behaviour, they could be human interpretations dressed up as the dog’s own voice.
Understanding is already relational
The opening of my BBC Dough: Pets conversation gives me a better starting point than the promise of a chip. Greg Foot asks whether I talk to my dogs and want to know what they are saying back. I answer: “I think I do understand quite a lot of what they’re saying. Mostly it’s around food.” The joke carries an argument. Understanding is already happening, but it is practical, embodied and situated: appetite, routine, attention, movement, the familiar signals between these particular dogs and me. The programme frames the question through AI, cloning and the future products that might reshape life with pets. Those technologies may make animals more measurable and more legible to us. But greater visibility is not the same as mutual understanding. A sensor can record a dog’s movement; an algorithm can classify a pattern. Neither automatically tells us what that experience means to the dog.
What technology could help us see
There may still be a modest, valuable future for technology here. Machine learning could help connect a dog’s vocalisations to situations, bodily cues and subsequent behaviour. Wearables could reveal changes in activity or stress. Better research might help owners recognise needs they currently miss. Dogs can learn associations between human words and objects, and can be remarkably attentive to human gestures. A future interface might therefore support a practical exchange: a system notices a pattern, offers a tentative interpretation, and checks it against what the animal does next.
That would be a form of communication. It would not be telepathy, and it would not mean that a human had acquired direct access to a dog’s private thoughts. A responsible system would have to preserve uncertainty: “This pattern may indicate discomfort” is a different claim from “Your dog says it is in pain.” The first is a hypothesis that can be tested and acted on; the second risks converting a complex animal into a character in a human conversation.
The second body and the animal mind
This question connects to my work on The Second Body and on Me:chine. Neural interfaces make the body newly available to systems: signals can be sensed, classified and translated into action. For humans, this raises questions about agency and authorship: who initiates an action, who interprets it, and who controls the interface through which it travels? With animals, those questions become sharper because the animal cannot consent to the device, approve the categories used to interpret it, or correct a translation that gets its experience wrong.
This also connects to The Mind Becomes Capital, where I examine how cognition, once measured and turned into data, may be treated by institutions and markets as a resource to capture and improve. The same temptation applies to animal minds: make them legible, quantify their states, and build products around the promise of a clearer connection. But the value of understanding an animal should not be measured only by the new services or human conveniences it creates. The first test should be whether the technology improves the animal’s life, and whether it respects the animal’s own ways of sensing and acting.
The dog already has a voice
The 2028 deadline in the post is unsupported. The deeper possibility is more interesting than the headline. We may build systems that help us interpret animal behaviour with greater care. We may learn to recognise patterns we have overlooked. We may even discover that some animal communication is more structured or surprising than we presently understand. But a translation into fluent human speech is not the natural finish line of that research.
Dogs already communicate. The challenge is not to give them a human voice. It is to become better readers of the voices, bodies, scents and behaviours they already have, and to remember that understanding another creature may require us to meet it in its world, rather than pulling it into ours.
Sources and notes
Neuralink, Telepathy: device control clinical trial. Neuralink’s public description focuses on device control through decoded neural signals.
Neuralink, PRIME Study progress update. The company describes its initial indication as restoring digital autonomy for people with quadriplegia by decoding movement intentions into actions such as cursor movement.
Correia-Caeiro, Guo and Mills, “Bodily emotional expressions are a primary source of information for dogs, but not for humans”, Animal Cognition (2021).
Berg et al., “Olfaction in the canine cognitive and emotional processes”, Neuroscience & Biobehavioral Reviews(2024).
Amy Martin, “Using AI to ‘talk to animals’ might make us feel clever – but what, if anything, does it do for them?”, The Guardian (22 September 2026).
BBC Sounds, Dough: The Future of Pets.