Whales May Be Key to Training AI for First Contact

The first whales appeared in a form different from what we know today almost 50 million years ago. In contrast, the massive developments we see around Artificial Intelligence today are perhaps the most modern feature of our times. The convergence of these two elements reflects the extensive scientific and technological progress we have made throughout our civilization. Given that the news might sound amazingly unbelievable – at first – we must start by unraveling it to its fullest extent possible. 

Click here to learn all about investing in AI.

Whales and Non-Human Intelligence Communication

A team of scientists from the SETI Institute, University of California Davis, and the Alaska Whale Foundation has been studying humpback whale communications. They intend to develop intelligence filters that can help search for extraterrestrial intelligence. 

The team was met with significant success when a whale named Twain responded to the ‘contact’ call played by the scientists into the sea via an underwater speaker. The contact call was modeled in a conversational style resembling a Whale’s ‘greeting signal.’ The exchange lasted for 20 minutes, and the Whale responded to each playback call, matching the interval variations between each signal. 

The Significance of the Study

According to Brenda McCowan, the lead author at UC Davis, this is believed to be the “first such communicative exchange between humans and humpback whales in the humpback language.” 

According to Dr. Fred Shape, the co-author of the study:

“Humpback whales are extremely intelligent, have complex social systems, make tools – nets out of bubbles to catch fish -and communicate extensively with both songs and social calls.” 

The study’s importance lies in the fact that this was the first time communication could be established between a human and a nonhuman entity that is intelligent and part of a complex social system. 

The research paper cited some characteristics of humpback whales’ communication style that make it all the more vital as a potential entry point to training Artificial Intelligence for First Contact. 

• The Humpback Whale’s brain hosts a large auditory cortex and clusters of spindle neurons.
• The research believes that impressive mental abilities likely underwrite the humpback’s “superlative acoustic prowess.”
• The humpback whale’s acoustic achievements are not confined to random sounds and calls. They involve lengthy, rhythmic, and constantly evolving songs. 
• Like humans, the humpback whale song includes the use of a similar frequency band composed of voiced and unvoiced elements, with silence punctuating the vocalizations of variable duration. 

These characteristics made the researchers believe that the diversity of signals a whale could produce was a key indication of how behaviourally flexible they were and – therefore – could provide an excellent opportunity to test technologies for interacting with a potential nonhuman intelligence. 

Establishing first contacts has always been a matter of great interest for researchers. National Geographic defines a first contact as the “initial encounter between cultures that were previously unaware of each other.” 

The world as we see it today is built of one culture, with its own language and social customs, migrating to a new place and mingling with people living there with another set of language and customs. For instance, in the Americas, the first contact refers to when indigenous people of the land came to interact with the Europeans in the 16th and 17th centuries. 

Establishing first contact has so far been time-consuming and cumbersome as it entails learning new signs – verbal, physical, and written. Learning these signals or communication patterns also involves documenting them, discovering the patterns within, reproducing them for the alien set of people to understand and interpret, and more. 

Scientists have always been interested in using technology to communicate with the unknown, and AI has been the latest addition to that effort.

The Use of AI to Talk to Nonhumans

According to Karen Bakker, a professor at the University of British Columbia and a Fellow at the Radcliffe Institute for Advanced Study:

“We need to understand nonhuman communication on its own terms.”

Efforts towards this objective have given birth to the field of ‘digital bioacoustics’ that studies how artificial intelligence and improved sensors can help observe and decode how a variety of species, including plants, share information in their own methods. 

An interview between Karen Bakker and the Scientific American provided details on how technology could help communicate with honeybees. 

Communicating With Honeybees

The relevant research was led by Tim Langdraf of Freie University Berlin. It successfully deployed computers and deep-learning algorithms to follow honeybees speaking to one another through their body movements as well as sounds. 

The positive results were an outcome of prudently combining computer vision with natural-language processing capabilities. The researchers have reportedly perfected algorithms to track individual bees and the impact the communication of an individual bee might have on another bee. 

Landgraf then encoded this information into a robot, naming it RoboBee. He designed seven or eight prototypes to come up eventually with a ‘bee’ that could enter the hive and emit commands that honeybees would obey. This led to a robotic inclusion going beyond the notion of establishing ‘First Contact’ with a bee issuing commands, telling a bee when to stop, and more.

Efforts like this, modeled around using AI to have nonhuman communication, have been taken up elsewhere in the world as well. For instance, the Earth Species Project is a non-profit that is dedicated to using AI to decode non-human communication.

1. Earth Species Project

The driving vision of the Earth Species Project was to leverage modern machine learning to build powerful semantic representations of language that we can use to unlock communication with other species. The Project has made several successful attempts to understand how nonhumans interact to date. 

The Crow Vocal Repertoire project, for instance, involved senior AI research scientists to work on mapping the vocal repertoires of two species of crow. The objective of the Project was to understand the role of acoustic communication in group coordination. The researchers believed that mapping the vocal repertoires of the Hawaiian crow and the Carrion crow could help unearth cultural and behavioral complexities crucial for planning effective nonhuman conversation strategies. 

Another group of senior AI research scientists involved in the Project of self-supervised ethnography discovery has been working on self-supervised methods for interpreting data collected from animal-borne tags, known as bio-loggers. The bio-loggers would help scientists record an animal’s motions and obtain audio-visual from the animal’s perspective. 

Bioacoustics is something that many large-scale organizations are also working with. Microsoft is one such example. 

2. Microsoft Research Working on BioAcoustics

Microsoft Research’s bioacoustics efforts are targeted toward leveraging AI to process audio recordings of diverse species. It studies and explores different aspects of sound production, dispersion, and reception in animals.

In one of its projects, the Microsoft Research lab is exploring call clusters for multiple animal species. It is developing self-supervised learning models that help explore clusters of animal calls that are either from different species or contain different stereotypes of calls from the same species. 

Another project by Microsoft Research is devising an automated bioacoustic analysis platform named Akoustos. The analysis platform intends to help conservation organizations and researchers with automated acoustic annotation and analysis with machine learning techniques. It will allow users to upload audio files and corresponding labels to the platform. The platform will also be capable of automatically processing audio data and building complex machine-learning models. 

In another project by Microsoft Research, titled ‘Using bioacoustics for multi-species classification,’ the concerned team evaluates deep convolutional neural networks for classifying the calls of 24 birds and amphibian species. 

In the fiscal year 2022, Microsoft reported US$198 billion in revenue and US$83 billion in operating income. It was the first time that Microsoft Cloud surpassed US$100 billion in annualized revenue. 

In 2022, Microsoft provided US$3.2 billion in donated and discounted technology to 302,000 nonprofits serving over 1.2 billion people globally. 

3. Google AI for Social Good

It has been close to four years now since Google AI for Social Good’s bioacoustics team created an ML model that helps the scientific community detect humpback whale sounds using acoustic recordings. Google developed this tool in collaboration with the National Oceanic and Atmospheric Association to help biologists study whale behaviors, patterns, populations, and potential human interactions. 

The Project has been developed in accordance with Google’s AI principles, which “recognize that advanced technologies can raise important challenges that must be addressed clearly, thoughtfully, and affirmatively.”

Google’s Project empowers users to leverage AI to explore thousands of hours of humpback whale songs and make their own discoveries with Pattern Radio. 

Alphabet, the parent organization of Google, registered a revenue of US$76.69 billion for the quarter ended on September 30, 2023. Its net income for the quarter stood at nearly US$20 billion, with a diluted EPS of US$1.55. 

4. Arbimon by Rainforest Connection

Conservation technology nonprofit Rainforest Connection has developed an online audio analysis platform by the name of Arbimon. The purpose is to make bioacoustics data analysis easier and more accessible. 

Arbimon is a free online platform. Scientists and ecologists can use it to upload their audio data and analyze it in the most advanced way possible. The platform equips users with tools for them to detect sounds from audio data and group them based on range, frequency, and similar other factors. 

Using Arbimon, it will be possible to process 10,000 files in a matter of a few seconds. And according to Burhan Yassin, CEO of Rainforest Connection:

“The platform was designed to specifically address the gap between research and conservation so that we can get to a point where we can drive conservation action on the ground.” 

As it keeps collecting more data, Arbimon plans to start mapping what biodiversity looks like on a holistic level. Every few days, Arbimon adds anywhere between 2 and 3 million recordings. It takes research institutes years to collect that much data. 

The platform now has more than 114 million recordings uploaded to it, with around 3,000 scientists having used it or using it. 

Headquartered in Katy, Texas, United States, Rainforest Corporation is a non-profit. It raised US$500,000 worth of grants from the Patrick J. McGovern Foundation on December 20th, 2023. 

5. Bioacoustic AI Consortium

This research-driven consortium spans eight countries and ten doctoral projects. It includes 23 experienced investigators, nine universities/research institutes, two museums, 5 SMEs, 3 NGOs, and one governmental organization. 

The vision or purpose of the consortium is to “make sense of animal sounds using AI.” At a functional level, the consortium’s research program develops new AI methods for sound directly within the applied context of acoustic wildlife monitoring. The Project believes that understanding animals better would help protect biodiversity, and AI can help monitor a variety of species. 

The consortium is on its way to developing best-in-class AI task formulations suitable for the special constraints of wildlife monitoring tasks, including the monitoring of hard-to-detect birds and mammals in European and tropical soundscapes. The AI techniques will help connect animal sounds with the social and behavioral structure of animal lives.

AI-Based First Contact for Good

The quest to uncover the unknown has consistently been the driving force in creating the extensive knowledge base of human civilization over time. AI is accelerating the rate at which we are advancing towards deciphering previously unknown signals and languages. Building relationships with non-human entities will enhance our understanding of them, leading to more effective protection. As biodiversity and sustainability become universal priorities, it’s nearly time to tap into the vast wealth of knowledge available in the non-human realm.