Stephanie King has spent thousands of hours listening to dolphins in Shark Bay, Australia, as co-lead of one of the longest-running wild dolphin research projects in the world. She uses AI in her work. She does not believe AI will let us 'speak dolphin,' and she thinks the question itself is doing damage. "I think we're doing the animals a disservice by focusing on that question," she says. The core tension is between two visions of what AI can do for animal science. One — the one getting funded and headlined — is the Dr. Dolittle fantasy: crack animal language, build an interspecies translator, win the Coller Dolittle Challenge prize. The other, held by researchers like King and elephant expert Joyce Poole, is more modest and more honest: use AI to process vast datasets of animal vocalizations, movements, and behaviors to better understand communication systems we barely grasp. Dolphins process echolocation through brain regions connected to touch, not vision. They may be 'feel-hearing' their environment. That's not something a large language model can decode. The hype pipeline is already distorting incentives. King says researchers are "biasing themselves towards the most exciting discovery" and describing findings in sensational terms to attract tech funding and media coverage. The framing — that we're about to 'crack the code' of whale song or elephant rumbles — creates expectations that outrun the science by years or decades. Poole, who has built one of the world's richest annotated archives of elephant vocalizations at ElephantVoices, is blunt: "I get quite exasperated by this goal of being able to talk to the animals. I think it's kind of gimmicky." The risk calculus is not hypothetical. César Rodríguez-Garavito, founding director of NYU's More Than Human Life (Moth) program, acknowledges AI's potential to give non-human species a voice in marine-protection policy or environmental law — efforts already underway in multiple countries. But he immediately pivots to the threats: trophy hunters using AI-powered tools to lure animals, whale-watching companies attracting cetaceans toward boats, national park visitors provoking wildlife through apps. "The nightmare scenario is one where we interrupt their communications and their lives at will, just for our entertainment," he says. Technologies are not beholden to the intentions of their makers. Moth recently published the PEPP framework (Prepare, Engage, Prevent, Protect), an ethical guideline set for non-human animal communication technologies. But Rodríguez-Garavito concedes that whether and how it will be adopted is uncertain — a downstream effect of the broader absence of legal and ethical boundaries around AI deployment in wildlife contexts. The regulatory vacuum means the most commercially viable applications, not the most scientifically responsible ones, will likely move fastest. The structural problem is familiar from other AI domains: the people who know the most about the subject (field researchers with decades of observational data) have the least control over the narrative, while the people who control the narrative (tech funders, media, prize organizers) have the least understanding of what the science actually shows. Dolphins have been studied for decades and we still don't know what most of their sounds mean. The gap between that reality and the 'we're about to crack the code' headline is where extraction happens — attention and capital flow toward spectacle while the slow, unglamorous work of actually understanding animal cognition gets crowded out. What's being sold is a mirror, not a window. The fantasy of talking to animals flatters human intelligence and human centrality. The researchers closest to the work are asking a harder question: can we learn to understand these beings on their own terms, even if their experience is fundamentally inaccessible to us? That question doesn't win prizes or generate clicks, but it's the one that might actually serve the animals.
