Monkeys, ChatGPT, and Us
What do monkeys, ChatGPT, and us have in common? According to Quentin Gallot, a PhD candidate at the University of Neuchâtel and researcher at NCCR Evolving Language, they all possess extraordinary predictive abilities that manifest in similarities in their communication systems. In a new study published in the journal iScience, Quentin and his team analysed the vocalisations of olive colobus monkeys, monkeys from West Africa, and discovered an unexpected grammatical system. This discovery provides fresh insight into the evolution of human language and its parallels with artificial intelligence.
An olive colobus. © Clémentine Bodin.
Olive colobus monkeys (Procolobus verus), discreet inhabitants of the tropical forests, have a simple communication system made up of only two call types: a short, low call (A) and a long, high-pitched call (B). These primates combine these calls in sequences that can reach up to 24 consecutive calls—combinations that are far from random.
In a previous study, Quentin Gallot had already shown that this species follows specific rules to combine its calls, depending on the type of threat. For example, a sequence starting with “A” signals the presence of an eagle, while a sequence starting with “B” indicates the detection of a leopard or the imminent fall of a tree. Building on these findings, the researchers now aim to understand how information flows within these long call sequences.
When every call counts
In the face of a sudden and unpredictable danger, every call counts. “To maximise the chances of survival, essential information must be transmitted and understood from the very first alert calls of a sequence,” says Quentin Gallot, first author of the study. To test this hypothesis, the researcher analysed thousands of vocalisations of olive colobus monkeys recorded in Taï National Park, Côte d’Ivoire. To decode these call sequences, he used statistical models typically reserved for the study of human language.
In his study, Quentin reveals that information about the danger type is transmitted gradually. The very first call already provides a reliable clue about the level of threat, distinguishing, an eagle attack—which requires an immediate response—from other less urgent disturbances. The third call then further specifies the nature of the threat, differentiating a leopard from a falling tree. Subsequent calls can then adjust or refine the information, confirming or correcting the initial message.
At the heart of artificial intelligences
“We humans don’t wait until the end of a sentence to react or respond because our brains are very good at predicting the next word, and our grammar is adapted for it,” explains the biologist. This principle is also at the core of generative artificial intelligences like ChatGPT, which, thanks to algorithms capable of predicting the next word, have become surprisingly adept at speaking like humans in just a few years.
But is this ability unique to our species? Probably not. Olive colobus monkeys possess a form of grammar that, like humans, allows them to gradually understand a message by predicting the upcoming calls. “Our results suggest that this ability, via grammar, to predict and interpret information throughout a vocal sequence, would have emerged long before the advent of human language,” he concludes.
So far, Quentin Gallot has focused on alert situations, but these monkeys also communicate during their daily interactions. “Their vocal production seems even more complex, with longer call sequences and more acoustic variations,” the researcher explains. These elements are what his team plans to explore in a forthcoming study. This could perhaps reveal further aspects of the communication system of these monkeys, and offer new perspectives on the emergence of our own language.
