Nature Speaks. Can AI Help Us Listen?

Aza Raskin, Co-founder of the Earth Species Project, delivered a keynote at SXSW 2026 arguing that AI has unlocked an unprecedented ability to decode and eventually translate animal communication — and that doing so may be one of the most important steps humanity can take toward surviving on a finite planet. Raskin frames the core challenge facing civilization not as a lack of technology, but as a problem of disconnection: from each other, from our systems, and from the natural world. Earth Species Project's mission is to use AI to break the interspecies understanding barrier as a mechanism for reconnection and paradigm shift.

The talk walks through mounting scientific evidence that the natural world is saturated with complex communication. Parrot parents whisper unique names to their chicks; dolphins use signature whistles as proper names and reference individuals in the third person; elephants, belugas, and marmoset monkeys all use individual names. Between 60–70% of crow communication consists of quiet, intimate calls made while flying — a category entirely unknown to science before AI analysis surfaced it. Dolphin communication is so poorly understood that scientists label roughly half of it simply 'non-signature whistles' — analogous, Raskin quips, to cataloguing all of human language as 'names and non-names.' Plants are not exempt: a 2019 University of Tel Aviv study showed primrose flowers producing sweeter nectar within three seconds of hearing a bee's approach, and stressed tomato and tobacco plants emit ultrasonic signals at human conversational volume. A South American vine appears capable of visually mimicking artificial plants, suggesting perception without a central nervous system.

The technical core of the keynote explains how AI makes cross-species translation plausible. A 2017 deep learning breakthrough showed that embedding spaces — geometric representations of language where relationships between concepts become distances and directions — are structurally universal. English, Spanish, Japanese, Finnish, and Aramaic all share the same underlying 'shape,' enabling translation without a Rosetta Stone. Crucially, this universality extends beyond language: the same technique maps image-to-text, fMRI brain activity to visual imagery, and in principle DNA sequences, robot motion, and animal vocalizations. A 2023 study already reconstructed what humans were seeing purely from fMRI data. The 'Platonic Representation Hypothesis' — a recent AI research result — finds that regardless of model or modality, these learned shapes are converging toward a common structure, suggesting AI is approaching a representation of reality independent of any single sensing modality.

Raskin describes several active Earth Species projects. Working with University of Leon biologists, AI mapped crow vocabularies and revealed the 60–70% of communication science had never catalogued. In a 48-hour zebra finch experiment — compared to the prior state-of-the-art of 10-second playback trials, a 17,000x extension — female finches were found to alter their calls to match the AI's communication style, overturning the assumption that their songs are innate. A pilot with Raincoast off British Columbia is attempting to infer orca behavior directly from audio, to unlock two decades of unannotated recordings. Beluga whales communicate up to 120 kHz (vs. human hearing at 20 kHz) in modem-like information packets, yet even with whalable sensor backpacks, researchers can disentangle only about 3% of useful data — making beluga communication a larger unexplored frontier than the ocean floor itself.

The talk addresses ethical dimensions forthrightly. Raskin warns that the ability to communicate fluently will arrive before genuine understanding does — ChatGPT-style pattern mimicry may produce responses that seem meaningful without true comprehension. Playing a deceased matriarch's call to an elephant herd caused excitement that turned to grief. Synthetic whale songs could go viral in humpback culture, disrupting 34 million years of cultural evolution. The risk of bad incentives (e.g., factory farms discovering emotional states that affect yields) demands that laws, norms, and international treaties precede economic exploitation. A global study with the Collective Intelligence Project found 60–70% of respondents across 30–40 countries said animals should participate in democratic governance if they could be understood. Raskin concludes with the analogy of the Hubble Deep Field: pointing the telescope at what appeared to be empty sky and discovering more galaxies than had ever been seen. Earth Species is building the equivalent instrument for nature — and the discovery it expects is that humanity, like Earth, is not the center.