Beyond Words: Using AI to Communicate with Animals
For decades, researchers have been fascinated by the complex communication systems of animals, with African elephants being a prime example. Joyce Poole, who has studied these majestic creatures for over 50 years, observed something intriguing in the 1980s: elephants seemed to direct their calls to specific individuals. However, at that time, the tools to confirm her hypothesis didn’t exist. Fast forward to today, and advances in artificial intelligence are beginning to decode these intricate animal communications, revealing a world of possibilities for interspecies communication.
Poole’s observations, in collaboration with Mickey Pardo, laid the groundwork for a breakthrough study. They recorded nearly 500 elephant calls, meticulously noting the context and intended recipient of each call. By feeding this data into a statistical model, they discovered that the model could predict the recipient of a call with remarkable accuracy. This finding suggests that elephants might use specific “names” to communicate with one another, a discovery that has shifted our understanding of animal intelligence and communication.
The use of AI doesn’t stop with elephants. Machine learning is now being applied to tackle the challenges of animal communication research, such as the “cocktail party problem,” where multiple animals vocalise simultaneously in a noisy environment. AI models, like Deep Karaoke, originally designed for human speech recognition, have been adapted to separate and identify individual animal calls from mixed recordings. This advancement allows researchers to study the vocalisations of animals like macaque monkeys in unprecedented detail.
Researchers have applied AI models, similar to those used in human speech recognition, to whale vocalisations. These models are capable of singling out individual whale calls from a cacophony of sounds, allowing scientists to study specific communications with unprecedented precision. This approach not only helps identify which whale is vocalising but also provides insights into the context and purpose of these calls. By unravelling the complexities of whale communication, AI is paving the way for a deeper understanding of these majestic creatures, potentially enabling us to interact with them in ways that were once unimaginable.
In addition to separating calls, AI is also being used to generate unique animal vocalisations based on existing recordings. This approach, known as “supervised learning,” relies on training models with large datasets labelled by humans. However, this method is limited by the extent of our knowledge about animal communication. As a result, some AI researchers are turning to “self-supervised learning” models, like those used in natural language processing, which can learn from vast amounts of unlabeled data. These models have the potential to uncover patterns and structures in animal communication that humans might overlook.
The Earth Species Project, co-founded by Aza Raskin, is one of several organisations working to develop AI models for animal communication. Raskin believes that just as AI models have learned to translate between human languages by identifying similarities in linguistic structures, they could do the same for animal communication. By aligning the “shapes” of different communication methods, AI might help us translate animal vocalisations without needing direct examples or a “Rosetta stone” for each species.
However, this ambitious project faces significant challenges. One major concern is validation — how do we ensure that the AI models are accurately interpreting and translating animal communication, especially when these forms of communication are so different from our own? Additionally, there is the question of whether we might overestimate the potential for meaningful conversations with animals, given the vast differences in our experiences and perceptions of the world.
Despite these hurdles, the potential benefits of AI-driven research in animal communication are profound. Around the world, researchers are collecting vast amounts of data — audio, video, and spatial information — on various species to feed these AI models. The ultimate goal is to achieve a deeper understanding of nonhuman species, which could lead to greater appreciation and protection of the animals with whom we share the planet.
As we continue to explore the possibilities of AI in decoding animal communication, we are reminded that humans are not the only beings capable of complex thought, communication, and emotional connections. By bridging the communication gap, we can gain insights into the minds of other species, affirming their right to exist and their unique place in the world. The journey toward interspecies communication is just beginning, and time will tell how far AI can take us in understanding and protecting the diverse life forms that inhabit our planet.
