How MITs AI Is Painting a New Future for Conservation

In a world where artificial intelligence is often seen as a source of recycled or uninspired content, a new model from MIT is demonstrating that generative AI can have a soul. By blending science, art, and technology, this project is designed to foster a deeper connection with the natural world.

Developed by MIT Sea Grant, this innovative approach to visual storytelling is named LOBSTgER, which stands for Learning Oceanic Bioecological Systems Through Generative Representations. Its mission is to learn from natural processes to reveal the hidden beauty and ecological health of threatened marine ecosystems, starting with the Gulf of Maine.

The Fusion of Art Science and Technology

This groundbreaking project is a collaboration between two experts from different fields. It is co-led by underwater photographer Keith Ellenbogen, a visiting artist at MIT Sea Grant, and Andreas Mentzelopoulos, a PhD student in mechanical engineering at MIT.

The creation of LOBSTgER requires immense dedication both in the ocean and in the lab. Capturing meaningful marine photography is a challenging art form that demands multiple dives and patience with unpredictable conditions. On the technical side, training the AI model involves hundreds of hours of development and careful 'hyperparameter tuning' to ensure it generates scientifically accurate images—avoiding creations like a five-eyed shark with wings.

How Does Generative AI Create Images

Generative AI models like LOBSTgER, OpenAI's DALL-E-2, and Midjourney operate as what are known as diffusion models. This process involves training the AI on massive datasets of labeled images.

The model learns by taking an image and systematically adding 'noise' until it becomes an unrecognizable static-like pattern. It then learns to reverse this process, starting from static and removing the noise step-by-step to construct a brand-new image based on the patterns it learned, often guided by a text prompt.

Training an AI on the Richness of the Ocean

LOBSTgER's knowledge base is built from Keith Ellenbogen's stunning underwater photography, all captured within the Gulf of Maine, one of the world's most dynamic marine environments. This 36,000-square-mile 'sea within a sea' is home to over 3,000 species, from microscopic plankton to massive whales.

To ensure LOBSTgER is a valuable conservation tool, every image in its dataset is meticulously crafted with artistic intent, technical precision, and accurate species and location data. This rich, high-quality dataset allows the AI to generate incredibly realistic imagery. For instance, can you tell which of the sharks below is real and which was generated by LOBSTgER?

(Spoiler: The shark on the left is the AI-generated image, created after 30,000 training cycles.)

More Than Just Pictures AI for Deeper Insights

While creating beautiful images of sharks is impressive, the project's true power lies in its ability to help scientists analyze, categorize, and understand ecological changes. As conservationists collect more data than humans can possibly analyze, AI becomes an essential tool for drawing insights.

The creators liken LOBSTgER's potential impact to that of the camera's invention in the 19th century. Just as the camera gave us a new way to document the world, AI offers a new way to understand it. It can learn to recognize complex details like water clarity, subtle signs of disease on animals, coral bleaching, and the presence of contaminants.

LOBSTgER can also be used to enhance existing photographs, revealing details that might otherwise be missed. In the image below, an original photo by Ellenbogen is compared to an enhanced version created by the AI.

These details are critical for conservation. Is a whale covered in harmless barnacles or sores from an illness? Are waters becoming murky due to pollution? LOBSTgER is being trained to help answer these questions by showing aquatic ecosystems with unprecedented clarity.

AI as a Powerful Ally for Conservation

Projects like LOBSTgER prove that AI is more than just a content-creation engine; it's a revolutionary technology with immense potential for good. Its value depends entirely on how it's used. In ecology, AI is already becoming a vital tool for documentation and data analysis.

Though not involved with this specific project, Ali Swanson, Conservation International’s director of nature tech and innovation, has spoken about how AI can aid conservation, stating it will help scientists 'map and monitor changes and threats with far greater precision and speed.'

AI models like LOBSTgER can track intricate changes in wildlife health, population dynamics, and aquatic conditions. These same deep-learning principles are being applied to analyze camera trap footage, monitor deforestation from satellite images, and even create maps to guide plastic pollution cleanup efforts.

As ecosystems face increasing threats, we need better technological tools to analyze vast datasets and devise effective conservation strategies. Initiatives like LOBSTgER are paving the way for a future where technology and nature work in harmony.