Google announced DolphinGemma, a language model built to interpret dolphin vocalisations, on the same day the Earth Species Project (ESP) reported that its bird dataset had grown to 150,000 recordings. Both developments were disclosed in a September 8 2026 article by The Decoder, a German technology outlet.
Scale of data collection
The Decoder notes that the ESP dataset now comprises “150 000 Aufnahmen, einer der größten Vogel‑Datensätze überhaupt.” The figure is presented as of September 2026 and is described as one of the largest bird collections worldwide. No earlier baseline is given in the packet, so the claim stands alone as a snapshot of current scale.
Google’s DolphinGemma model
In the same piece, Google is reported to have “kürzlich … DolphinGemma ein Sprachmodell für Delfinlaute vor.” The model is positioned as a species‑specific AI tool, extending Google’s portfolio of large‑language models into marine bioacoustics. Google’s corporate details – chief executive Sundar Pichai, headquarters in Mountain View, United States, and an employee headcount of 47,756 – are drawn from Wikidata (Q95). The packet flags the Wikidata figures as background only, but they are the only available source for company background in this brief.
The gap to two‑way communication
Both the dataset expansion and DolphinGemma are framed against a broader claim: “no AI system has achieved verified two‑way communication with a non‑human animal.” The Decoder cites Jeremy Coller’s June forecast, reported via Bloomberg, that a genuine two‑way breakthrough could arrive by 2030. The forecast is presented as a forward‑looking statement, not a current fact, and underscores that the goal remains unrealised as of September 2026.
Sector implications and ethical concerns
The rapid growth of animal‑sound datasets and the launch of species‑specific models raise several industry‑wide questions. First, the scale of the ESP bird collection suggests that AI researchers now have a richer acoustic training base than ever before. Larger datasets typically improve model performance, which could accelerate research into decoding animal signals.
Second, Google’s entry signals that major U.S. tech firms see commercial or scientific value in extending language‑model technology beyond human text. While the packet does not disclose a pricing model or projected revenue, the move may prompt other AI players to pursue similar niche models, potentially creating a new sub‑segment within the broader AI market.
Third, the ethical dimension is foregrounded in The Decoder’s commentary. The outlet warns that “je besser Maschinen Tiere verstehen, desto größer wird die Gefahr, dass dieses Wissen gegen sie verwendet wird,” highlighting concerns that improved decoding could be weaponised or used without animal consent. The article mentions calls for ethical guidelines and even animal privacy rights, indicating that regulatory scrutiny could follow as the technology matures.
What remains unknown
The packet does not provide a timeline of past dataset growth, so the rate of increase cannot be quantified. It also lacks details on DolphinGemma’s architecture, training data, or performance metrics, leaving investors and analysts without a basis for valuation. Finally, the exact definition of “verified two‑way communication” is not spelled out, and no independent third‑party verification is cited.
For now, the sector’s narrative is one of rapid data accumulation and model diversification, tempered by the acknowledgement that true two‑way dialogue with non‑human animals is still a future milestone. Stakeholders—from biotech investors to animal‑rights groups—will be watching both the technical progress and the emerging policy debate closely.