We Can Finally Talk To Animals

Summary

The video describes how AI is being used to study animal communication by analyzing animals’ own sounds and behavior, rather than teaching animals human words. It emphasizes that recognizing patterns is not the same as understanding what those patterns mean: none of the examples establishes a full, two-way conversation.

Studies and methods described

• Zebra finches: In a 2026 study, researchers reportedly trained AI on more than a million finch calls. The system generated new calls in real time, and nearby finches responded to them more than to a recording played on a loop. The birds appeared sensitive to the timing and sound of the calls, but researchers still do not know what the calls mean.

• Earlier attempts to teach animals human communication:

  • In the 1960s, Margaret Howe Lovatt and dolphin Peter attempted English lessons. Peter could imitate a sound resembling “hello,” but this did not demonstrate understanding.
  • Nim Chimpski learned many signs, but Herbert Terrace concluded that he often copied his teachers rather than constructing sentences.
  • Koko became known for using many signs, though critics questioned how reliably people around her interpreted them.

The video contrasts these examples with AI’s approach: analyzing animals’ own signals instead of teaching them ours.

• Elephants: A team analyzed about 500 recorded African elephant rumbles collected over more than 30 years, along with information about who called and who was being addressed. AI identified patterns associated with particular recipients. In playback tests, elephants responded more to calls originally directed at them than to calls meant for other elephants. Researchers interpret this as evidence that elephants use name-like calls.

• Dolphins: Google’s DolphinGemma model, developed with Georgia Tech and the Wild Dolphin Project, predicts patterns in dolphin sounds and can generate new sounds. The video also describes experiments with dolphins’ individual “signature whistles”: dolphins tended to respond to playback of their own whistle.

In another project, divers use an underwater computer called CHAT to play invented whistles associated with objects such as toys. Dolphins have imitated some of these sounds, but researchers have not shown that they intentionally use them to request the associated objects. DolphinGemma is presented as a research tool, not a translator.

• Sperm whales: Project CETI researchers used AI to analyze thousands of sperm whale vocal patterns. The video says they found variation in click timing and rhythm, including extra clicks, and features likened to vowels. A generative system called Wham was trained on recorded whale sounds and can produce new, whale-like “codas.” These sounds have not been played to whales, and their meanings remain unknown.

• Dogs and pigs: A study using recordings from 74 dogs reportedly found that AI could distinguish playful from aggressive barking and estimate some characteristics, with about 70% accuracy. A separate study analyzed more than 7,000 pig grunts and squeals from over 400 pigs; its model classified the emotional context of sounds with 92% accuracy. The video suggests such tools could eventually help monitor animal welfare, while noting that consumer “pet translation” apps generally lack scientific validation.

• A test of AI’s limits: Researchers applied similar pattern-finding methods to recordings of human toddlers, whose likely intentions can be partly checked. The AI’s sound groupings often did not match what the children appeared to mean. The video uses this example to stress that acoustic patterns alone do not establish meaning.

• Testing claims about animal language: The video notes that a 2025 bonobo study proposed that bonobos combine calls to produce meanings, but says a later analysis found similar patterns in random, fictional data. This is presented as a warning that statistical structure can be mistaken for evidence of language.

• Connecting sounds to behavior: Researchers are recording animals’ sounds alongside video, tags, or observations of what the animals are doing, so AI can link vocalizations with context. A robotic honeybee at the Free University of Berlin reportedly performed a food-location dance that real bees followed, suggesting that machines may be able to convey information through animal signals.

• Ethical and practical issues: The video describes a $10 million prize for demonstrating two-way communication with an animal, with smaller annual awards for progress. It also raises concerns about how people should respond if an animal communicates pain or distress, and whether inappropriate signals could confuse or disrupt animals.

Researchers and sources featured

Named people

• Margaret Howe Lovatt — dolphin language project with Peter
• John Lilly — led the dolphin project
• Herbert Terrace — chimpanzee sign-language research
• Mikey Pardo — Colorado State University elephant study
• David Gruber — Project CETI
• Elodie Briefer — pig vocalization research
• Jeremy Koller — funder of the communication prize

Organizations, institutions, and projects

• Earth Species Project
• Colorado State University
• Save the Elephants
• Elephant Voices
• Google
• Georgia Institute of Technology
• Wild Dolphin Project
• University of St Andrews
• Project CETI
• Massachusetts Institute of Technology (MIT)
• University of Michigan
• University of Copenhagen
• University of California, Berkeley
• Free University of Berlin
• Columbia University
• NASA
• A research team in Israel and a collaborating team in Mexico, not named in the subtitles

Publications and named initiatives

• Science — the bonobo study and subsequent analysis are discussed
• Call of the Collar — the animal-communication prize described in the video
• DolphinGemma and CHAT — the underwater dolphin playback system
• Wham — the sperm-whale sound-generation system described in the video