Competition to translate animal sounds into human language

In 2025, we will see artificial intelligence and machine learning being used to make real progress in understanding animal communication, answering the question that has baffled humans for as long as we have existed: “What do animals say to each other?” The latest Coller-Dolittle Awardoffering up to half a million dollars in prize money to scientists who “crack the code” is an indication of our high confidence that we are realizing this goal of recent technological advances in machine learning and large language models (LLM).

Many research groups have been working on algorithms for years to make sense of animal sounds. Project Jet, for example, decrypts click on trains of sperm whales and songs of humpbacks. These modern machine learning tools require extremely large amounts of data, and so far there is no such high-quality and well-interpreted data.

Consider LLMs such as ChatGPT, which have training materials that include all the text available online. Such information on animal communication has not been available in the past. It’s not just that the corpus of human data is much larger than the kind of data we can get for animals in the wild: more than 500 GB of words were used to train GPT-3, compared to more than 8,000 “codes.” ” (or sounds) for Project Ceti’s latest analysis of sperm whale communication.

In addition, when working with human language, we already to know what is said We even know what a “word” is, which is a huge advantage over the interpretation of animal communication, where scientists rarely know whether a particular wolf howl means something different from, say, another wolf howl, or even whether wolves howl. we rarely know whether he thinks so. it’s kind of like a “word” in human language.

Nevertheless, 2025 will bring new advances in both the amount of animal communication data available to scientists and the types and power of AI algorithms that can be applied to that data. Automated recording of animal sounds has been placed within easy reach of every scientific research group, with inexpensive recorders such as the AudioMoth gaining popularity.

Massive datasets are now available online as recorders can remain in the field for long periods of time listening to the calls of gibbons in the jungle or birds in the forest 24/7. There have been cases where it was impossible to manage such massive data sets manually. Now, new automatic detection algorithms based on convolutional neural networks can comb through thousands of hours of recordings, picking out animal sounds and grouping them into different types based on their natural acoustic properties.

Once these large animal data sets are available, new analytical algorithms become possible, such as using deep neural networks to find hidden structure in sequences of animal sounds that may be analogous to meaningful structure in human language.

But the main question that remains unclear is what exactly are we going to do with these animal sounds? Some organizations, such as Interspecies.io, state their goal very clearly as “transmitting signals from one species into coherent signals for another.” In other words, for translate animal communication to human language. However, most scientists agree that non-human animals do not have their own language – at least not that we humans do.

The Coller Dolittle Prize is a little more complicated, looking for a way to “communicate with or decipher an organism.” Deciphering is a slightly less ambitious goal than translating, given the likelihood that animals do not actually have a translatable language. Today, we don’t know how much or how little information animals communicate among themselves. In 2025, humanity will have the potential to understand not only how much animals say, but also what they say to each other.