Earth Species Project
The Earth Species Project is a nonprofit research organization Aza Raskin co-founded in 2017 with Britt Selvitelle, dedicated to using machine learning — specifically the transformer architectures that power large language models — to decode the communication systems of nonhuman animals. The organization has released NatureLM-audio, the first foundation model trained on animal vocalizations; presented research at NeurIPS; and secured seventeen million dollars in grants heading into 2025. Its significance in Raskin's broader framework lies in the demonstration that the same technology he warns could hack the operating system of civilization can be redirected toward the expansion of human understanding rather than the capture of human attention.
In the AI Story
The project's intellectual framing distinguishes between super human and extra human applications of AI. Super human applications accelerate what humans already do — more code, more features, more productive output along existing dimensions. Extra human applications expand what humans can perceive and understand — making categories of knowing available that the pre-AI world did not make visible. Decoding whale song is the paradigmatic extra human application, because it reveals patterns in nonhuman communication that human perception cannot detect and human analysis cannot process.
The organization's technical work has produced genuine contributions to computational bioacoustics. NatureLM-audio applies self-supervised learning techniques — the same methods that enabled large language models to acquire linguistic competence from text — to the far larger and less-organized corpus of recorded animal vocalizations. Early results have identified previously unrecognized structural patterns in crow communication, orca dialects, and cetacean song.
The project's significance in Raskin's broader argument is structural. His critique of engagement-optimized AI cannot be dismissed as technophobia because he is simultaneously building with the same technology. His advocacy for humane design cannot be dismissed as impractical because he is demonstrating, in his own practice, that the alternative applications exist and can produce results. The Earth Species Project is the working proof that the dual-use technology can serve purposes other than the ones the market currently rewards.
Raskin's signature reframing — AI is like the invention of the telescope, and when we invented the telescope, we learned that Earth was not the center. I've been thinking a lot about the implications of what happens when AI teaches us that humanity is not the center — emerges directly from this work. The decentering is not metaphorical. It is what happens when the technology reveals that other species possess communicative structures humans had assumed were uniquely human.
Origin
The project was co-founded in 2017 by Raskin, Britt Selvitelle, and Katie Zacarian. Its initial focus was applying machine learning techniques developed for natural language processing to animal vocalizations — a research program that anticipated, by several years, the broader recognition that transformer architectures could be adapted across domains beyond text.
Key Ideas
Extra human applications. The organization's operating premise that AI's highest value lies in expanding human understanding along new dimensions, not accelerating human productivity along existing ones.
Dual-use reality. The same technical infrastructure that enables engagement extraction also enables interspecies communication research — the design and incentive structure, not the technology, determine which purpose prevails.
NatureLM-audio. The first foundation model for animal vocalizations, released in 2024, representing a genuine technical contribution to computational bioacoustics.
Appears in the Orange Pill Cycle
Further reading
- Earth Species Project, NatureLM-audio technical report (2024)
- Aza Raskin, What if we could talk to animals? (TED, 2023)
- Karen Bakker, The Sounds of Life (2022)