Climate Modeling and AI
Former Google head honcho Eric Schmidt thinks AI could be put to good use in sustainability and climate predictions across the world
At a time when different parts of the earth are facing heat waves, wildfires and floods, former Google CEO Eric Schmidt believes that use of artificial intelligence (AI) could go a long way in predicting extreme weather. Experiments by semiconductor giant Nvidia at creating an AI-powered “digital twin” of the planet could be the first of many steps in this direction.
The concept of digital twins represents a scenario where an actual real-world product, process or system serves as a digital counterpart for simulation, testing, monitoring and integration. Though the concept itself isn’t new, the growing power of generative AI is creating excitement over the future possibilities.
Digital Twins approach to solve climate prediction
What Nvidia is attempting to do here is create a digital twin called Earth-2 that will use weather prediction from FourCastNet AI model that uses massive data from the earth’s system to predict over the next two weeks, in a much quicker and more accurate way. Schmidt thinks such systems can potentially generate thousands of predictions instead of just 50 now.
These accurate risk predictions of disasters could provide vulnerable populations valuable time to prepare and evacuate, Schmidt says while noting that climate modeling could just be an early use-case for AI, which, with sensible regulation and proper support for innovative uses, could address science’s most pressing issues.
AI can change the way science works
“We can build a future where AI-powered tools will both save us from mindless and time-consuming labor and also lead us to creative inventions and discoveries, encouraging breakthroughs that would otherwise take decades,” he says in an authored article published in the MIT Technology Review.
Schmidt, who co-founded Schmidt Futures that bets on exceptional minds making the world better, says the world is limiting AI with large language models but there are many different model architectures that could have larger impact. Scientists have used AI models to identify an antibiotic to combat a pathogen while Google DeepMind model can control plasma in a nuclear fusion reaction, bringing the world cleaner energy options.
AI isn’t just about large language models
“In the past decade, most progress in science has come through smaller, “classical” models focused on specific questions. These models have already brought about profound advances. More recently, larger deep-learning models that are beginning to incorporate cross-domain knowledge and generative AI have expanded what is possible,” he says.
Schmidt believes that while the core of scientific process will remain the same as taught in elementary school, AI could revolutionize how each component of scientific research, starting from background understanding, identifying a hypothesis, conducting experiments, analyzing data and drawing conclusions.
Scientific process itself will be simplified
Review of scientific literature is already simplified by AI tools such as PaperQA and Elicit. The next stage is to generate hypotheses across a wider spectrum and narrow the net down faster, resulting in stronger options. He notes the example of scientists at Caltech using AI fluid simulation models to design better catheters that prevent infections.
In the experimentation process, AI could conduct these faster, cheaper and at a greater scale, says Schmidt while suggesting that AI-powered machines with hundreds of micropipettes can run day and night to create samples at a rate no human process can match. Instead of running a handful of experiments, scientists can use AI tools to run a thousand, he adds.
When it comes to analysis and drawing conclusions, self-driving labs can go beyond automation and use large language models to interpret the results or recommend another experiment. In fact, AI lab assistants could even order supplies to replace those used in the past and set up and run the next set of recommended experiments overnight.
AI tools can lower the entry barrier for new scientists and open up opportunities to those who may have gotten left out from the process, he said while warning that recognition of the areas where human touch is still important is the best way to utilize AI more effectively.
