MIT's new AI model is to blend science, art and technology to stimulate a deeper connection to nature, proving that generative AI doesn't have to be soulless.
Developed by MIT Sea Grant, this “new frontier of Visual Storytelling” is called Lobstger, short for learning marine ecological systems through generative representations.
And it does exactly that, learning from natural processes and better reveal the hidden beauty and ecological state of innately threatened marine ecosystems, like in Maine, where Robsgar's training data set was collected.
Coding and Building Frameworks
The project is co-led by underwater photographer Keith Ellenbogen, visiting artist at MIT Sea Grant, and student Andreas Mentzelopoulos, MIT Mechanical Engineering PhD.
Building a Lobstger requires a lot of work both in and out of the water. Marine photography is a challenging scientific art form that includes “multiple dives, missed opportunities, and unpredictable conditions” to capture meaningful content.
Similarly, training the diffusion model to generate the desired image requires hundreds of hours of development and laborious “hyperparameter tuning” or the Robsgar's learning process to avoid generating a 5-eyed purple shark with a winged 5th.
How Generation AI works
Generated AI entities such as Openai's Dall-E-2 and Midjourney are trained through a machine learning process that supplies large datasets of related labeled images.
They are also known as diffusion models because they “spread” a given image by adding more “noise” until it is spread out statically similar multicolored asexually in TV. The diffusion model then reverses the process and gradually removes noise, creating a new, desired image, etc., based on a text prompt.
Build a dataset in the Gulf of Maine
Lobstger is trained on a dataset consisting of underwater photographs of Ellenbogen taken in the Bay of Maine, one of the world's most dynamic ecosystems. The “sea in the ocean” of this 36,000 square mile bay is diverse in both geology and biology. It is shaped by deep basins, shallow banks and powerful tides that mix freshwater flowing through 60 rivers through the North Atlantic oceanic waters.
Maine Bay houses more than 3,000 species of seabirds and marine animals, from whales, sharks, seals, jellyfish and rocky plankton, forming the basis of aquatic food chains.
To make Lobstger a useful tool for maintenance, its dataset must be meaningful. Therefore, each image is “made with artistic intent, technical accuracy, accurate species identification, and clear geographical context.”
As an example of AI behavior, one of the following images is real, and the other is generated by Lobstger.
Spoiler Alert: The left shark is the product of the Robsgar's diffusion model after training 30,000 training “epochs” or passes through the data set entirely.
Use technology to help nature
Do creating artificial images of sharks directly encourage people to stop the scattering and choose plastic from the ocean? Perhaps not, but this initiative does something even more important. It enhances the AI's ability to analyze, classify, and reveal natural ecological changes.
Datasets are essential, but they are less useful if they do not have the ability to extract insights from the information they contain. That is becoming more and more humanly impossible given the amount of data collected by conservationists.
The creator compares 19th century Robsgar with the advent of the camera. As cameras have introduced an unprecedented ability to document and reveal the world, AI can help them do the same by understanding complex nuances such as water clarity, species-specific details, and aquatic conditions that constantly change under waves.
As an example, the following image of an American lobster has been enhanced from image to image from Lobstger.
These nuances that Lobstger is learning are very important – is that whale covered in barnacles or is it a pain caused by some kind of illness? Are these corals bleached? Do these waters appear dark as they are leaking into the ocean?
Therefore, Lobstger was not created in the only range that generates AI images. This is intended to enhance the impact of underwater photography by showing aquatic ecosystems in unprecedented details to reveal previously hidden ecological effects on various scales.
Prove that AI can be a permanent source of strength
Many people ride AI, and while it's all over the world with gloomy and recycled content, this is not a failure of AI itself.
It all depends on who is using it, how and why. In ecology, AI could be a unique tool for generating documents, data analysis, and actionable insights, and is now.
Ali Swanson, director of Nature Tech and Innovation at Conservation International, recently spoke about how AI can help conservation. Though he's not involved with Lobstger, Swanson said AI can help conservationists “map and monitor changes and threats with much more accuracy and speed.”
So, AI entities like Lobstger are to establish a futuristic type of integrity beyond image generation. This is to learn to analyze a larger dataset and to track complex changes in wildlife health, populations, and aquatic conditions.
And there is no need to limit such progress to the oceanic realm. The deep learning process developed here is used to monitor camera trap images to measure wildlife health, diversity and shift populations. From above, satellite data observes that the Earth's band becomes greener, bluer, or more barren, and selects an area for repair.
As a result of developing Lobstger and its AI ILK, these learning processes become smarter, becoming more accurately preemptive issues, such as deforestation, allowing you to create maps to collect plastic pollution from marine ecosystems.
Overall, initiatives like Lobstger are needed more than ever. Ecosystems are decreasing or disappearing, and better technical methods are needed to analyze huge datasets and plan the most effective strategies for conservation.
About the author: Ivan spends most of his time reading and writing about interesting things. A school scientist, Ivan frequently covers science, technology, history and culture, and sometimes writes internet comedies.
The only opinions expressed in this article are those of the author.
