Overview
Interspecies communication covers the scientific, technological, and artistic work of understanding how other organisms signal and, in some cases, of signalling back. Ethology built the foundations over a century, but the field changed after 2015, when cheap recorders and self-supervised models made it possible to find structure in millions of hours of animal sound without first labelling them. Whether animals want to be understood, and what is owed to them in the process, remains an open question.Examples
- Project CETI's analysis of 8,719 sperm whale codas recorded off Dominica, published by Sharma et al. in Nature Communications (2024)
- Earth Species Project's NatureLM-audio (2024), an audio-language model trained on recordings across thousands of species
- The Wild Dolphin Project's CHAT wearable and Google's DolphinGemma (2025), trained on 40 years of Atlantic spotted dolphin recordings
- Gunnison's prairie dog alarm calls, which specify a predator's species, size, and colour (Slobodchikoff)
- Arachnophilia, which amplifies spider web vibrations into concerts and installations (Saraceno)
Signals
The honeybee waggle dance encodes direction in the angle of the run relative to the sun and distance in its duration, a symbolic code that took a decade of marked-bee experiments to read (von Frisch, 1967). Humpback whale song is built from themes and phrases that an entire population revises from year to year, and recordings of it released in 1970 sold over 100,000 copies and helped shift public opinion toward whale conservation (Payne and McVay, 1971). Both established that a nonhuman signalling system could be read closely, and that reading it changed how the species was treated.
Alarm Calls
Captive language studies with parrots and bonobos remain disputed in their interpretation. Field studies of natural signals have been easier to verify. Gunnison's prairie dog alarm calls distinguish coyote from dog from human and encode the size, shape, and colour of the intruder, down to the shirt a researcher wore, and playback experiments show that the colony responds differently to each (Slobodchikoff, 2009). African elephants address specific individuals with name-like rumbles, and the addressed elephant responds when its call is played back while others ignore it (Pardo et al., 2024). Bottlenose dolphins do something similar with signature whistles, each animal developing its own in its first year and answering when it is copied. Individual names appear to exist in at least two species besides humans.
Plants and Fungi
Carbon moves between trees through shared mycorrhizal fungi (Simard et al., 1997), though the evidence for forest-wide "communication" is much thinner than the popular account (Karst, Jones, and Hoeksema, 2023). Pea roots grow toward the sound of running water (Gagliano et al., 2017). Tomato and tobacco plants emit ultrasonic clicks between 20 and 100 kilohertz, dozens per hour when cut or dehydrated and roughly one per hour when healthy, and a classifier can tell the two conditions apart from the sound alone (Khait et al., 2023). On this evidence, signalling occurs across ecosystems and is not limited to animals, which is one premise of ecological intelligence.
Listening at Scale
The main change of the past decade is the volume of recordings. Autonomous recorders, hydrophones, and animal-borne tags produce archives that no one can listen to, and machine intelligence trained on those archives can cluster and segment calls without labels. Digital bioacoustics is the name for this pairing of sensors and models (Bakker, 2022), and it applies the methods of planetary computation to the living world. The method was demonstrated on roughly 15,000 Egyptian fruit bat vocalizations, classified by context: quarrels over food, sleeping position, mating (Prat, Taub, and Yovel, 2016).
Sperm Whales
Sperm whale codas are the click sequences whales exchange at the surface. An analysis of 8,719 codas from the clans around Dominica found that they vary in rhythm, tempo, rubato, and ornamentation to produce at least 156 distinct types, a combinatorial system compared to a phonetic alphabet (Sharma et al., 2024). Earlier catalogues had counted around 20 coda types. The recordings come from a network of moored hydrophones, drones, and suction-cup tags that Project CETI has maintained around the island since 2020, dense enough to attribute each coda to a known individual. The study does not claim that whales have language. It shows that their communication carries more structure than earlier catalogues had measured.
Foundation Models
NatureLM-audio (2024) is an audio-language model trained on recordings across thousands of species that identifies species, counts individuals, and describes calls in plain language (Earth Species Project). DolphinGemma (2025) is a roughly 400-million-parameter model trained on 40 years of Atlantic spotted dolphin recordings, small enough to run on a phone in the field and paired with the CHAT wearable, through which a diver exchanges synthetic whistles with dolphins. The Coller Dolittle Challenge (2024) put a 10 million dollar prize on demonstrable two-way communication with another species. The models are good at finding structure. Whether that structure carries meaning is a separate question that the models cannot answer on their own.
Art and Design
Artists and designers have approached the exchange as an experiment in shared attention rather than as a decoding problem. The Interspecies Internet (2013) imagined networked interfaces through which animals could communicate across distance. Arachnophilia amplifies spider web vibrations, in Arachnid Orchestra (2015) in Singapore and in On Air (2018) at the Palais de Tokyo, where visitors, spiders, and airborne dust shared the exhibition (Saraceno, 2018). Refuge for Resurgence (2021), shown at the Venice Architecture Biennale, set a banquet table for humans, foxes, pigeons, rats, and fungi, with a place setting designed for each (Superflux, 2021). In multispecies design, these projects push toward interfaces addressed to nonhuman users directly, from pollinator sensors to bat-responsive building envelopes. In each case the interface is designed around the other species' sensory range.
Ethics and Open Questions
Decoding a signal is different from having access to an experience. A bat's sonar world stays closed to a human however completely its calls are catalogued (Nagel, 1974). Animals cannot consent to being tagged, recorded, or addressed with synthetic calls. The search for a "phonetic alphabet" in whales may also be anthropocentric, looking for a version of human language rather than a system that is unlike it. There is also a question of ownership. Recordings and the models trained on them can be sold, while the species contributing the data receives nothing. One argument holds that animals already act politically, by refusing, leaving, and negotiating, and that the gap lies in the ability of human institutions to recognise this (Meijer, 2019). Legal personhood for New Zealand's Whanganui River (2017) and constitutional rights for nature in Ecuador (2008) create parties that must be spoken for, and interspecies research is being drafted as evidence. Whether that speaking amounts to representation is a central problem for more-than-human design. The practical outcome of the field so far is a set of tools for paying closer attention to other organisms, and a set of unresolved questions about when and how to use them.