In 2025 we will see AI and machine learning used to make real advances in understanding animal communication, answering a question that has puzzled humans for as long as we have existed: “What do animals say to each other?” one?” The new one Coller-Dolittle Prizewhich offers cash prizes of up to half a million dollars for scientists who “crack the code” is an indication of a strong belief that recent technological advances in machine learning and large language models (LLMs) put this purpose in our understanding.

Many research groups have been working for years on algorithms to understand animal sounds. Project Ceti, for example, decodes the click the trains of sperm whales and the songs of humpbacks. These modern machine learning tools require a large amount of data, and until now, such a quantity of high-quality and well-annotated data has been lacking.

Consider LLMs like ChatGPT with training data available to them that includes general text available on the internet. Such information on animal communication has not been available before. Not only are human data corpora many orders of magnitude larger than the kind of data we have access to for animals in the wild: More than 500 GB of words were used to train GPT-3, compared to just over 8,000 “codas” (or vocalizations) for a recent Project Ceti analysis of sperm whale communication.

Moreover, when working with human language, we already know what was said. We even know what constitutes a “speech,” which is a huge advantage in interpreting animal communication, where scientists rarely know when a particular wolf is howling. , for example, means different from another wolf howling, or even if wolves consider a howl to be somehow similar to a “word” in human language.

However, 2025 will bring new advances, both in the amount of animal communication data available to scientists, and in the types and power of AI algorithms that can be applied to the data. Automatic recording of animal sounds has been put within easy reach of every scientific research group, with inexpensive recording devices like the AudioMoth exploding in popularity.

Large datasets now come online, because recorders can be left in the field, listening to the calls of forest gibbons or forest birds, 24/7, over long periods of time. There are times when such large datasets are impossible to handle manually. Today, new automatic detection algorithms based on convolutional neural networks can sift through thousands of hours of recordings, selecting animal sounds and clustering them into different types, according to their natural acoustics. characteristics.

When big animal data becomes available, new analytical algorithms become a possibility, such as using deep neural networks to find the hidden structure of sequences of animal vocalizations. , which may be similar to the meaningful structure of human language.

However, the fundamental question that remains unclear is, what exactly are we hoping to do with these animal sounds? Some organizations, such as Interspecies.io, set its goal as clear as, “transfer signals from one species to the same signal for another.” In other words, at Translate animal communication to human language. But most scientists agree that non-human animals have no actual language of their own—at least not in the way that we humans have language.

The Coller Dolittle Prize is a little more sophisticated, looking for a way “to communicate or interpret the communication of an organism.” Deciphering is a slightly less ambitious goal than translating, considering the possibility that animals may not, in fact, have a language to translate. Today we do not know how much information, or how little, the animals express themselves. By 2025, humanity has the potential to leapfrog our understanding of not just how much animals talk but what exactly they say to each other.