Understanding how dolphins communicate has been a long-standing challenge for scientists. These marine mammals produce complex sounds, ranging from clicks and whistles to squawks, that play a central role in their social lives. Despite decades of observation and recording, uncovering the structure and meaning behind these vocalisations has remained slow and difficult. However, the introduction of DolphinGemma, a new AI tool developed by Google in collaboration with Georgia Tech and the Wild Dolphin Project (WDP), marks a major development in this field. DolphinGemma applies advanced artificial intelligence to help researchers analyse dolphin sounds more efficiently, with the aim of identifying patterns and gaining a clearer picture of how dolphins interact through sound.
The Background: Dolphin Research and the Wild Dolphin Project
The Wild Dolphin Project, founded in 1985, is the longest-running underwater research programme dedicated to dolphins. Based in the Bahamas, this project focuses on a specific pod of Atlantic spotted dolphins. Over the past four decades, the team has collected a vast archive of underwater audio and video recordings, alongside detailed behavioural observations of individual dolphins. This database provides a rich foundation for AI models like DolphinGemma to process.
Through consistent fieldwork, researchers have identified specific types of dolphin vocalisations that correlate with different social behaviours:
- Signature Whistles: Individual dolphins produce unique whistles, much like names, particularly used between mothers and calves.
- Burst-Pulse Squawks: Typically heard during disputes or when dolphins display aggression.
- Click Buzzes: Linked to hunting and mating activities, such as chasing prey or engaging in courtship.
These vocal cues offer insights into the dolphins’ social organisation, but manual analysis of these sounds has always been slow, requiring years of painstaking work. This is where DolphinGemma offers a new approach.
How DolphinGemma Works: The AI Model at a Glance
DolphinGemma is designed to process dolphin sounds much like human language models handle text. Developed by Google, the model builds upon previous AI systems such as the Gemini and Gemma language models. It uses SoundStream tokenisers, which convert dolphin vocalisations into digital representations (tokens). These tokens are then fed into a machine learning system that can detect patterns and predict what sound might come next.
The model is trained to recognise sequences of dolphin sounds and identify recurring structures. With 400 million parameters, it is powerful enough to run detailed analysis while being light enough to operate on consumer devices like Google Pixel smartphones. This means researchers can carry DolphinGemma into the field and receive real-time analysis of dolphin vocalisations, rather than having to wait until they return to a lab.
Dr. Thad Starner, a leading figure in both DolphinGemma and a related project called CHAT, notes that by analysing these sound sequences, researchers can begin to uncover the rules that govern how dolphins communicate.
The Role of CHAT in Human-Dolphin Interaction
DolphinGemma is focused on understanding how dolphins talk to each other. The CHAT (Cetacean Hearing Augmentation Telemetry) system, also developed with Georgia Tech, takes this a step further. CHAT is designed to allow basic interaction between humans and dolphins, using synthetic whistles linked to familiar objects or concepts.
For example, a specific synthetic whistle might represent “seaweed” or “toy.” The idea is that dolphins might learn to mimic these sounds when they want those objects, creating a basic shared “vocabulary” between species.
CHAT devices include hydrophones and underwater speakers, with processing power provided by Pixel smartphones. This system can detect, classify, and respond to dolphin vocalisations in real-time, giving researchers a way to engage directly with the dolphins during dives.
Addressing Common Myths About Dolphin Communication
One of the most persistent myths is that dolphin sounds are simply random or basic signals without deeper meaning. In fact, researchers have long suspected that dolphins use a structured form of communication that may resemble aspects of human language. The challenge has been in proving this, given the difficulty of recording and analysing underwater audio over extended periods.
Another misconception is that dolphins can be trained to “talk” in human language. DolphinGemma and CHAT do not aim to make dolphins speak like humans, but rather to understand their own communication system on its terms. The goal is not to translate but to interpret the structure and function of dolphin sounds.
Practical Benefits of Using AI in the Field
DolphinGemma’s ability to run on portable devices like smartphones offers several benefits for marine researchers:
- Immediate Analysis: Researchers can analyse dolphin sounds in real-time, adjusting their methods based on immediate feedback.
- Energy Efficiency: Pixel devices use less power than traditional research equipment, making them suitable for long field sessions.
- Affordability: Using commercially available smartphones reduces the need for expensive, custom-built tools.
The next generation of CHAT systems, expected to use Pixel 9 devices, promises better integration of sound input and output, improved processing speed, and the ability to run more complex models while diving.
Open Access and Global Research Opportunities
Google has announced plans to release DolphinGemma as an open model by mid-2025. Initially trained on Atlantic spotted dolphins, the model can be adjusted for use with other species, such as bottlenose or spinner dolphins. By making DolphinGemma available to researchers worldwide, the hope is to accelerate discoveries and allow more scientists to study dolphin communication with fewer barriers.
This open model approach is expected to benefit:
- Marine biologists seeking to analyse local dolphin populations.
- Conservationists aiming to better understand dolphin behaviour in threatened habitats.
- Educators interested in using real-world data to teach students about animal communication and AI.
The Cultural and Ethical Questions Raised by DolphinGemma
If DolphinGemma helps prove that dolphins have a structured language, it could also suggest they have culture – shared behaviours and knowledge passed down through generations. This idea has major implications for how humans view animal intelligence.
Dr. Starner remarks: “If dolphins have language, they likely have culture. That means they have their own priorities, social rules, and traditions.”
Understanding these aspects of dolphin life could change public attitudes towards keeping dolphins in captivity or using them for entertainment. If we can understand what dolphins are “saying,” we may also need to listen more carefully to what they want or need.
Some experts and members of the public have already raised concerns about the ethical implications of communicating with dolphins. These include:
- How much interference is acceptable in wild dolphin communities?
- Should dolphins in captivity be given the chance to express their needs and desires more clearly?
- What responsibilities do humans have once we begin to understand dolphin thoughts and emotions?
The Future of Dolphin Communication Research
Looking ahead, DolphinGemma and CHAT represent just the beginning of what AI can achieve in this area. Potential future developments include:
- Expanding the model to work with other cetaceans like whales or porpoises.
- Using AI not just to predict sounds, but to link them with observed behaviours more accurately.
- Developing educational tools or public-facing applications that allow people to learn about dolphin communication interactively.
As more data becomes available and technology continues to advance, our understanding of dolphin societies could become much deeper, with new ways to protect and respect these intelligent creatures.
Conclusion: A New Approach to Understanding Dolphins
DolphinGemma brings together the fields of artificial intelligence and marine biology in a way that offers practical tools for researchers and opens new discussions about how humans interact with intelligent animals. By analysing dolphin vocalisations in real-time and allowing for possible basic interaction through CHAT, scientists are closer than ever to understanding what dolphins might be “saying.”
This progress raises important scientific, ethical, and cultural questions. As we continue to listen and learn, the tools we now have may help us build a more informed, respectful way of studying and living alongside one of the ocean’s most remarkable species.
About Search Engine Ascend
Search Engine Ascend delivers expert insights and practical knowledge in SEO and digital marketing. Our aim is to support businesses with accurate, thoughtful content that helps them succeed online. We provide analysis and advice grounded in current trends and technologies, helping our readers stay informed and ready for new challenges in the digital space.
