Talking to Whales

Look to the Oceans Instead of the Stars

Humanity has spent decades searching the cosmos for signs of extraterrestrial life. However, we may be overlooking a potential form of intelligent life right here on Earth: whales and dolphins. These marine mammals possess traits that suggest sophisticated cognition, such as large brains and advanced communication skills. They have large, intricate brains with expansive neocortices. They live in tight-knit social groups with specialized roles, working cooperatively to herd fish or corral krill. Different families even exhibit distinct dialects in their songs and vocalizations.

AI: A Solution

Now, advances in artificial intelligence may finally allow us to decode the communication of these marine mammals. A new project called CETI (Cetacean Translation Initiative) aims to use natural language processing to parse the meanings behind sperm whale codas. Led by researchers from institutions like MIT and Harvard, the team will deploy advanced recording devices around Dominica to capture whale clicks matched with behavioral data. Algorithms will then analyze millions of codas seeking patterns that may reveal syntax or structure analogous to human language. It is an Apollo program for interspecies communication.

The timing could not be more crucial. Whales and dolphins face grave threats from human activity like pollution, shipping traffic and climate change. Last year, an unprecedented number of whale strandings occurred along America's Atlantic coast. Starvation from lack of food sources was cited as a factor. Tools like AI can uncover insights into their intelligence and culture.

Consider how researchers discovered distinct whale dialects. Marine biologist Shane Gero spent years recording sperm whales near Dominica. He found clans used different codas much like how human languages have regional dialects. Two young males, nicknamed Drop and Doublebend, were recorded "conversing" in clicks for over 40 minutes. AI-aided translation may finally let us comprehend whale conversations.

Machine learning has already shown promise decoding animal communication. In 2019, researchers used neural networks to translate ultrasonic squeaks between mice and rats. Another study classified fruit bat "arguments" over food versus shelter with 74% accuracy. These methods depend on large troves of data. For whales, scientists must record millions of codas to expose patterns. Hence the underwater mic arrays around Dominica which listen nonstop.

A New Kind of Rosetta Stone

Large language models (LLMs) have demonstrated an aptitude for analyzing patterns and generating coherent text. This capability has led some scientists to propose using these systems as a "Rosetta Stone" to translate the complex vocalizations of species like sperm whales and bottlenose dolphins.

The basic concept involves feeding recordings of whale songs into a pretrained LLM and having it generate corresponding texts in English. Researchers could then analyze these outputs to derive insights about the meaning and linguistic structure behind the cetaceans' acoustic signals. The system would effectively act as a statistical deciphering mechanism, looking for correlations between sounds and suggested translations.

Proponents argue that LLMs' ability to model relationships in massive datasets could allow them to find connections that humans might miss. Their knowledge is also not constrained by assumptions about what whale communication should be like. As one scientist noted, "We have yet to demonstrate any non-human language use." Rather than searching for recognizable grammar, an open-minded AI could uncover wholly alien linguistic phenomena.

Challenges Ahead

Of course, major challenges remain. We cannot assume whale communication maps neatly to human language. Their sense of identity, time and emotions are likely alien to us. But even recognizing fragments or patterns in their chatter could provide insight into whale culture. Experts compare it to unraveling conversations at a crowded party using scattered microphone snippets.

Researchers are understandably cautious in speculating about true whale-to-human translation. Even if AI cannot deliver a dictionary of whale-speak, advances in machine learning will further illuminate these extraordinary marine intellects.