CETI Looks Into The Complexities Of Whale Sounds With AI

What can we learn from the whales? It’s something that researchers at the CETI project (not to be confused with the SETI Institute) are working on in order to help drive awareness around language models that exist right here in our own world.

In a recent TED talk, CETI’s Pratyusha Sharma talks about the communication of sperm whales, and how humans can use that to learn more about other species and ourselves. Sharma is a graduate student at CSAIL and works with advisors like our own Daniela Rus to advance this kind of discovery.

The Way They Speak

As a starter, she gave the example of aliens speaking to humans verbally, or through a script – and again, distinguish CETI from what they’re doing in space research!

“Communication is a key characteristic of intelligence,” Sharma explained. “Being able to create an infinite set of messages by sequencing together finite sets of sounds is what has distinguished human beings from other species.”

Life on Other Planets

However, she said, CETI research indicates that we may not be alone on the earth in developing these kinds of systems.

In figuring this out, she suggested, we can get insights on other species, and understand our own language better as well.

Millions of life forms on earth, she said, share some form of language.

“They have their own physical and mental constraints, and are involved in their own unique ecosystems and societies,” she said. “However, we know very little about – their communications.”

So how do you decipher them?

In further explaining what goes on at CETI, she listed different stakeholders with credentials in areas like linguistics, biology, cryptography and AI. (Here’s some more background on the project).

Most of the research, she said, is taking place in the Dominican Republic, or in the Caribbean.

A Long Timeline

Explaining how the large brains of sperm whales have evolved over 16 million years, she described activity that shows advanced thinking:

“The members of the family coordinate their dives, engage in extended periods of socialization, and even take turns babysitting each other’s young ones,” she said. “While coordinating in complete darkness, they exchange long sequences of sounds with one another.”

The question, she noted, is this: what are they saying?

Researchers at CETI have identified 21 types of “codas” or call systems with a certain complexity.

“One of the key differentiators between human language and all animal communications is that beautiful property called duality of patterning,” Sharma said. “It’s how a base set of individually meaningless elements sequence together to give rise to words, that in turn are sequenced together to give rise to an infinite space with complex meaning.”

She outlined some of the principles through which CETI is building this species knowledge.

“Getting to the point of understanding the communications of sperm whales will require us to understand what features of their (vocalizations) they control,” she said.

Visualizing Language

Presenting a set of “coda visualizations,” Sharma noted that these simple communications correspond to complex behavior.

“(This) presented a fundamental mystery to researchers in the field,” she said.

She showed how the CETI work magnifies the structure of a coda:

“Even though the clicks might not have sounded like music initially, when we plot them like this, they start to look like music,” she said, presenting a combinatorial coda system. “They have different tempos and even different rhythm.”

This, she added, reveals a lot about the minds of these creatures.

“The resulting set of individual sounds (in the coda) can represent 10 times more meanings than what was previously believed, showing that sperm whales can be much more expressive than what was previously thought,” she said. “These systems are rare in nature, but not uniquely human. … these results open up the possibility that sperm whales’ communication might provide our first example of this phenomenon in another species. … this will allow us to use more powerful machine learning techniques to analyze the data, and perhaps get us closer to an understanding the meanings of their sounds – and maybe (we can) even communicate back.”

The research, she added, continues:

“Hopefully the algorithms and approaches we developed in the course of this project empower us to better understand the other species that we share this planet with,” she said.

This type of research has a lot of potential!! Let’s see what it turns up as we continue through the age of AI.