How to build tomorrow’s AI for the Earth

Today’s AI is not sufficient for today’s environmental challenges. What led to today’s AI — and how can we build a new form of AI for tomorrow with Earth at its center?

As AI continues to advance and become more accessible, it is transforming the way we live, work, and communicate. But, for those of us engaged in the daily work of finding solutions for climate change and other environmental challenges, there is a gap between AI and climate impact. Simply put, you can’t solve climate change with AI-generated images, nor can you ask ChatGPT for the magic bullet for reducing greenhouse gas levels.

While there have been many laudable and exciting applications of AI towards solving environmental challenges, something still feels amiss. With the existential threat of climate change looming, why do most AI breakthroughs seem more preoccupied with chatbots, business innovations, or digital artistry than with our planet’s pressing needs?

Both AI and climate change have reached a new breakneck pace. So why is AI still not impacting environmental problems in the way we need it to?

I actually think the answer to this question is surprisingly simple. Simply put, today’s AI was not designed for this.

This post seeks to explore the factors that led to the AI of today, spotlighting the enabling factors that birthed platforms like ChatGPT and why today’s AI is not sufficient for tackling our climate challenges. This post also serves as a call to action for imagining the AI of tomorrow — an Artificial Intelligence designed with Earth’s environment as both its inspiration and its fundamental purpose.

Foundations of the Current AI Revolution

In the last decade, the AI landscape has undergone a seismic shift, evolving from a niche field of study to a cornerstone of modern technological innovation. This revolution didn’t happen overnight. It was the result of a confluence of factors, each playing a pivotal role in shaping the AI we know today. Importantly, these factors are baked into the very core of today’s AI, shaping both its structure and its applications.

Key factors enabling today’s AI

• Scale of Data: One way of thinking about GPT is that it has been trained by “reading the entire internet.” It is not hyperbole to say that without this vast digital trove of information, much of modern AI would be impossible. However, this means that the most useful AI applications in existence pertain to the kind of activity that most frequently occurs on the internet: discussion in human language and activity associated with commerce.
• Machine Learning Models: Over the years, researchers and scientists have developed sophisticated machine learning models capable of processing and understanding vast datasets. Transformers and large “foundation models” for text and images have been particularly influential. These models have showcased remarkable abilities, whether it’s understanding human language or recognizing intricate patterns in images.
• Computing Power: Parallel to the growth in data and models has been the advancement in computing power. Cloud computing providers have democratized access to high-performance computing resources. Moreover, the evolution of GPU hardware has accelerated the training of complex AI models, reducing the time from months to days or even hours.
• Technical Ecosystem: The AI community has always thrived on collaboration. Platforms like HuggingFace have emerged as invaluable tools, offering pre-trained models, datasets, and a platform for researchers to share and build upon each other’s work. Such ecosystems have expedited AI research and application, making state-of-the-art AI accessible to a broader audience.
• Capital Investment: Behind the scenes, the financial muscle of tech giants has played a crucial role. Companies like Google, Microsoft, and Amazon have poured billions into AI research, infrastructure, and talent acquisition. Their investments have not only propelled their own AI endeavors but have also nurtured the broader AI ecosystem, leading to innovations that benefit industries and consumers alike.

While these factors have collectively ushered in an era of AI-driven innovation, there’s a flip side. The AI we’ve come to know, shaped by commercial interests and trained predominantly on internet data, seems somewhat detached from pressing real-world challenges, especially those related to the environment.

Envisioning a future “AI for the Earth”

If today’s AI has been shaped by and for the factors outlined above, what is needed to move AI in a positive direction for the planet? Let’s explore how we need to reimagine the factors that drove today’s AI in order to build tomorrow’s “AI for the Earth”.

Mostly solved factors

• Computing Power: Actually, this one is already largely solved. We’re living in an era of unparalleled computational capabilities. The same cloud infrastructures and GPUs that powered the AI revolution of today can be harnessed for our AI for the Earth efforts of tomorrow.
• Scale of Data: Contrary to popular belief, we’re not starved for environmental data. In fact, we’re inundated with it. From satellite imagery capturing deforestation patterns to sensors monitoring ocean temperatures, there’s a treasure trove of data waiting to be deciphered. This data, however, presents its own challenges. Unlike text, which is universally understood, environmental data often requires specialized knowledge for interpretation. For instance, while an average person can read a blog post, interpreting metrics like NDVI or measuring groundwater storage demands expertise.

Challenges

• Machine Learning Models: While we have foundation models adept at processing text and images from the internet, we’re still in the nascent stages when it comes to models designed for global environmental data. These models need to accommodate the many complexities of geospatial data, namely that all points are associated with a location and time and temporal and spatial scales that range over many orders of magnitude.
• Technical ecosystem: One of the significant hurdles in harnessing AI for environmental purposes is the absence of a centralized “public planetary data infrastructure” (a phrase borrowed from my colleague Holly Buck). Such an infrastructure would not only organize and standardize vast amounts of environmental data, making it readily accessible for AI models, but also serve as a repository for the outputs generated by these models. Imagine a platform where researchers from around the globe can access satellite imagery, climate models, and biodiversity metrics, all standardized and ready for analysis. Furthermore, the insights and predictions generated by AI could be stored, shared, and acted upon through this same infrastructure. Establishing such a comprehensive system is a monumental task, requiring collaboration across nations, industries, and disciplines. However, its potential to streamline and amplify the impact of AI on environmental challenges is unparalleled.
• Capital Investment: The AI ecosystem has thrived, in part, due to the massive investments from tech conglomerates. However, the focus of these investments has largely been on commercial applications. There’s a pressing need to redirect capital towards harnessing AI for the planet. This isn’t just a philanthropic endeavor; the public sector and private sector need to be equally engaged.

In essence, while some of the building blocks for an “AI for the Earth” are in place, there’s a chasm between where we are and where we need to be. If we’re to realize the full potential of AI for the environment, we need to work together to bridge this gap.

Rethinking AI’s “gold standard”

There is a final, critical point I’d like to make pertaining the difference between AI today and “AI for the Earth”. In the discipline of artificial intelligence, we often find ourselves marveling at AI’s ability to reach parity with “human performance,” from understanding language nuances to recognizing intricate visual patterns. But as we stand on the precipice of environmental challenges, it’s time to ask: Is emulating human performance the ultimate articulation of AI’s potential, or is there a broader, more profound purpose we’ve yet to explore?

When it comes to managing our planet, “human performance” is, to put it mildly, less than stellar. Deforestation, pollution, and biodiversity loss are stark reminders of our shortcomings. So, if our goal with AI is to address environmental issues, benchmarking it against human performance seems counterintuitive. Instead, we should aspire for AI to achieve what humanity has struggled with: the ability to anticipate environmental changes, devise interventions, and manage ecosystems sustainably.

Transparency in building AI for the Earth: An “open” road

Okay, now that we have some idea of what AI for the Earth should look like, how should we start building it? In the rapid evolution of today’s AI, one aspect has become abundantly clear: the need for transparency, openness, and public discourse. As we venture into the realm of developing AI for the Earth, these principles become even more paramount. Some of the many important considerations include:

• Open Source Software: Making our tools and algorithms publicly accessible ensures they remain a shared resource. This openness invites global scrutiny and improvement, democratizing the AI development process.
• Shared Data and Public Science: Data drives AI. Sharing environmental data accelerates research and ensures collective action.
• Public Engagement: The power of AI, especially in environmental contexts, necessitates public discourse. We must discuss its governance, ethical use, and potential implications openly. Unlike the current AI “wild west,” our approach should prioritize collective wisdom and ethical considerations.

In sum, our vision for AI for the Earth is rooted in collaboration and transparency. The only path to AI for the Earth is along a truly “open” road.

Glimpses of tomorrow’s AI for the Earth

At Earth Genome, we have been hard at work inventing technologies to support this idea of what a future “AI for the Earth’’ could look like. Projects like the Global Plastic Watch and Amazon Mining Watch serve as beacons of this new paradigm. They exemplify the potential of AI to monitor, predict, and intervene in environmental issues, offering a glimpse of what’s possible when we rethink AI’s design and ultimate purpose.

Recently, we’ve been excitedly working on a new technology that represents one of the world’s most advanced and complete examples of what tomorrow’s AI for the Earth could look like. Code named Earth Index, our new platform allows for general environmental monitoring by bringing together modern foundation models for environmental data with a brand new technical ecosystem of supporting services, including a UI that allows for anyone to “search the Earth” just as someone using Google could search the Internet.

Searching for gold mining in the Amazon

We’ve already been incredibly impressed by the potential for Earth Index (expect additional blog posts about Earth Index very soon!). Datasets that would previously take months to generate are now possible to generate in an afternoon, unlocking the huge potential of environmental data for our expanding number of partners.

Charting a sustainable future with AI

The journey of reimagining artificial intelligence is both challenging and exhilarating. As we’ve explored throughout this post, the potential of AI to address environmental challenges is vast, but it requires a shift in perspective, design, and purpose.

At Earth Genome, our endeavors, from the Global Plastic Watch to the Amazon Mining Watch, have been a testament to this new vision of AI. Our latest innovation, Earth Index, is not just a technological marvel but a beacon of hope. It embodies the essence of what “AI for the Earth” can achieve, making environmental monitoring accessible, efficient, and impactful.

But this journey is not ours alone. It’s a collective endeavor, and we invite you to be a part of it.

• Stay Informed: Dive deeper into our work and vision by checking out our blog. Each post offers insights, updates, and glimpses into the future of environmental AI.
• Be the Change: Take these ideas, share them, discuss them, and implement them in your communities. The power of an idea grows exponentially when shared and acted upon.
• Collaborate with Us: We’re always on the lookout for partners, collaborators, and visionaries. If you’re as passionate about the environment and the potential of AI as we are, reach out to us. Together, we can shape the future of AI for the Earth and chart a sustainable path for our planet.

In the face of environmental challenges, the fusion of technology and purpose offers a ray of hope. Let’s harness the power of AI, not just for innovation but for preservation, ensuring a sustainable future for all.