Argonne’s Rick Stevens on Energy, AI, and a New Kind of Science

By Kevin Jackson

June 17, 2024

The world is currently experiencing two of the largest societal upheavals since the beginning of the Industrial Revolution. One is the rapid improvement and implementation of artificial intelligence (AI) tools, while the other is the sprint toward clean energy  deployment in the face of the global climate crisis.

Rick Stevens, Associate Laboratory Director and Argonne Distinguished Fellow
Credit: Argonne National Laboratory

Both of these technological changes will completely alter humanity’s trajectory. What’s more, their fates are intertwined.

Rick Stevens – the Associate Laboratory Director for Computing, Environment, and Life Sciences at Argonne National Laboratory – is one of the smart people who’s thinking deeply about how these two revolutions will interact. In fact, he co-authored Argonne’s AI for Energy report which discusses the lab’s current work as well as future aspirations for deploying AI tools during the clean energy revolution.

I was lucky enough to sit down with Stevens and discuss the report, as well as his musings on how AI could and should be deployed in the energy sector. While we couldn’t cover the entirety of the 70-page report’s contents, Stevens outlines some specific potential use cases of AI within energy as well as the challenges we’ll need to overcome.

A General Acceleration of Innovation

The report outlined five major areas within energy that AI could influence: nuclear energy, power grid, carbon management, energy storage, and energy materials. As we began our discussion, Stevens made a note that AI in energy should result in a “general acceleration of innovation.”

He initially mentioned nuclear reactors as a place where AI could accelerate certain necessary processes. The report itself stated that one of the largest obstacles to advanced nuclear reactors in the U.S. is a “slow, expensive, and convoluted regulatory process.” This is a task that is perfectly suited for AI.

“On the nuclear reactor front, one of the biggest targets for that community right now is trying to streamline licensing and helping to build reactors on a timeline within the budget,” Stevens said in our interview. “This is, of course, a huge problem for these projects.”

Staying within a timeline and a budget for nuclear reactors is challenging, as obtaining a construction permit and operating license for a new reactor in the U.S. can drag on for more than five years and can sometimes take decades. The report mentioned that multi-modal LLMs could help accelerate this process.

By training on datasets of scientific literature, technical documents, and operational data, these LLMs can help to streamline and expedite the nuclear regulatory licensing and compliance process. In a sense, these LLMs could act as virtual subject matter experts to help guide humans through the complicated regulatory process. On top of nuclear reactors, Steven’s mentions that the same sort of foundation model could help with the licensing process for renewable energies like wind or solar.

This is an overarching strategy that will apply to all scientific endeavors, not just energy. Steven’s mentioned the Frontiers in Artificial Intelligence for Science, Security, and Technology (FASST) initiative from the Department of Energy. Among other goals, this federal mandate is pushing to build capable foundation models that are experts in specific domains.

“The strategy that we’ve been working on in the FASST initiative is to build a handful of very capable foundation models,” Stevens said. “Think of them like ChatGPT but they are experts in some specific domain. You might imagine an expert model in the grid that knows everything about how grids work. The grid dynamics, the regulatory issues, the structural issues, the technical issues, the geopolitical issues – everything that humanity knows about building power grids, you could imagine a model that has all of that knowledge.”

With such potential for acceleration from AI, it will also be important to consider why we want to accelerate certain scientific fields. For instance, Stevens mentions drug development and how the success of these projects is literally a matter of life and death.

“You have a real motivation for trying to go faster, but you also want to go better,” Stevens said. “I think we need to help people understand that when we talk about accelerating science, we’re not just trying to turn the crank faster. We’re trying to build a better crank.”

This discussion will be especially relevant as we address the energy infrastructure issues that lead us to the current climate crisis. The worst-case predictions for climate change will lead to mass migration, famine, and water shortages. While it’s not a silver bullet solution, using AI tools to assist in the clean energy transition is of the utmost importance.

New Ways to Do Energy Science

As AI tools are relatively new – or at least many of their current capabilities are – implementing these solutions will require innovative ways of thinking. Stevens mentions the Stormer project as one area of AI with versatile use cases. This is a weather-specific vision transformer that can predict the global atmosphere 14 days into the future and is as accurate or sometimes more accurate than current partial differential equation prediction methods.

“(Stormer is) orders of magnitude faster (than current solutions), which means you can get a 10-day forecast in a few minutes,” Stevens said. “If you think about the application of that in the context of energy – say you’re running a wind farm and you’re trying to do capacity planning or plan maintenance. You’ll know what you have to anticipate.”

AI-based weather prediction tools will be important as we transition to clean energy solutions like wind power.

Stevens continued: “So far, that’s my favorite application because a large part of energy production and market-based pricing and where power is coming from in the grid is a prediction problem that tries to link up supply and demand. If we can get better models that can allow us to predict the factors that are affecting supply and demand, that means we can run at a higher efficiency. We can reduce cost and we can also help the market price better.”

On top of applying these AI tools in innovative ways, scientists in the energy space will also have to rethink how we do science. Stevens mentioned how the application of AI systems may benefit by operating under what he calls an inverse design.

Stevens stated that science currently proceeds by scientists thinking about something and making a hypothesis. In essence, the scientist guesses as to what they think might be correct and then they do experiments to test that guess.

While that process works wonderfully for humans, the implementation of AI tools might take a different path.

“If an AI can learn an entire domain deeply and it can reason about a specific material, then you can turn the whole process upside down,” Stevens said. “You can say ‘look I want a material that behaves like this – I shine a light on it and it turns purple.’ Rather than having to work forward through thousands of candidates and trying to search for things that turn purple as opposed to green, the system would operate under an inverse design. It might say ‘here’s the thing that makes purple when you shine light on it. This idea that you’re directly going to a solution is this idea of inverse design.”

Stevens is using an easy-to-understand example here with the purple-green distinction, but it isn’t hard to see how such an inverse design would be radically advantageous for scientists working on discovering new energy materials.

Pro-Science, But Never Anti-Human

It’s impossible to discuss AI innovations without also addressing the common fear that these tools will replace people. In Stevens’ mind, nothing could be further from the truth when it comes to integrating AI into the clean energy transition. When asked how we can safely apply AI tools to domains that demand success like nuclear reactors, he had quite a pithy response:

“Well, humans also sometimes get things wrong, and that is really important,” Stevens said. “We need to understand how things currently fail. Not just how AI fails, but how do complex systems where people are already making decisions fail?”

Both humans and Ai systems are prone to making mistakes. The right idea isn’t to never make a mistake, its to plan for mistakes and mitigate them.

We already operate in a world of human imperfection. As such, we embed checks and balances within our many complicated systems to catch humans who may be incorrect, incompetent, or malicious. Steven’s stated that we’ll have to do much of the same for AI and he mentioned a clarifying metaphor.

“Imagine you have somebody who can hit a lot of home runs, but they also strike out a lot,” Stevens said. “The question is how do you minimize the strikeouts while maximizing the home runs? More specifically for AI, can we build AI systems that have more awareness of their own mistakes?”

Stevens mentioned that there’s a technical term for this within AI called uncertainty quantification. This is where users want the AI to output a result, but they also want it to estimate how likely it is that the result is correct.

In a perfect world, this would allow us to tell the AI to only relay information to us that is correct – but we don’t live in a perfect world. Stevens stated that solving the problem of determining the validity of what a model is outputting is a huge area of research.

To solve problems like this at a larger scale, the report mentions that “laboratories must establish a leadership computing ecosystem to train and host data and foundation models at ever-increasing scales.” To Stevens, a “leadership computing ecosystem” would have several components.

“One aspect is that they train a big foundation model,” Stevens said. “These take many months on exascale-class machines. We would need to have essentially dedicated multi-access scale class hardware at the heart of the ecosystem for training. That’s what FASST is building out with even larger machines, heading towards this 100,000 AI exaflop class devices.

On top of these centralized machines, Stevens also mentions that this leadership computing ecosystem would also need to focus on edge devices.

He mentions a scenario where someone is monitoring a real-time system like a generator, the grid, or some other complex energy system. They would need sensors flowing into the model for inference, and they might also have a parallel simulator digital twin running in parallel. In such a scenario, the big machine would be used for the heavy lifting concerning these foundation models, but there will also be coordinated sensors and other devices on the edge to collect data.

As scientists build these foundation models, the pipelines of clean data being fed in will require fine-tuning of models as well as alignment. Stevens states that one might think of this as a layered process with the integration of many different kinds of facilities. He calls this an “integrated research infrastructure.”

“The concept is to tie the facilities together with high-speed networking, common APIs, common data interfaces, and control interfaces, so AI can read data directly from these facilities,” Stevens said. “If you were in a scenario where it makes sense to control them with AI, you would have a control interface. And you would tie all of that together with these inference engines.”

On top of this, a leadership computing ecosystem wouldn’t just share resources – it would also create a structured foundation on which to build new knowledge. AI tools are capable of thinking in ways that humans cannot, and this can often lead to exciting discoveries.

During our interview, we mentioned a research project where a surrogate model was trained on basic quantum mechanical results. Eventually, the model began to form salt crystals that it was not directly told about. While this is interesting in its own right, Stevens thinks we can take it a step further.

“If I integrate what we know about some domain, the model can synthesize that and make reasonable predictions like with these salt crystals – but we already knew about salt crystals,” Stevens said. “The question is whether it can make predictions about phenomenon that we don’t know about.”

This is exactly why AI will be a vital tool in the clean energy revolution. We have been using fossil fuels and legacy energy systems for so long that shifting gears will require new ways of thinking. While humans will obviously play a role in this shift, AI is capable of bringing about the new and innovative ideas that will help us stave off the worst effects of the climate crisis.

The integration of AI into the energy sector represents a pivotal moment in human history, where technological advancement intersects with the urgent need for sustainable energy solutions. As we navigate this transformative journey, it will be important to remember that AI should complement human expertise and be guided by ethical considerations.

Subscribe to HPCwire's Weekly Update!

Be the most informed person in the room! Stay ahead of the tech trends with industry updates delivered to you every week!

ARM, Fujitsu Targeting Open-source Software for Power Efficiency in 2-nm Chip

July 19, 2024

Fujitsu and ARM are relying on open-source software to bring power efficiency to an air-cooled supercomputing chip that will ship in 2027. Monaka chip, which will be made using the 2-nanometer process, is based on the Read more…

SCALEing the CUDA Castle

July 18, 2024

In a previous article, HPCwire has reported on a way in which AMD can get across the CUDA moat that protects the Nvidia CUDA castle (at least for PyTorch AI projects.). Other tools have joined the CUDA castle siege. AMD Read more…

Quantum Watchers – Terrific Interview with Caltech’s John Preskill by CERN

July 17, 2024

In case you missed it, there's a fascinating interview with John Preskill, the prominent Caltech physicist and pioneering quantum computing researcher that was recently posted by CERN’s department of experimental physi Read more…

Aurora AI-Driven Atmosphere Model is 5,000x Faster Than Traditional Systems

July 16, 2024

While the onset of human-driven climate change brings with it many horrors, the increase in the frequency and strength of storms poses an enormous threat to communities across the globe. As climate change is warming ocea Read more…

Researchers Say Memory Bandwidth and NVLink Speeds in Hopper Not So Simple

July 15, 2024

Researchers measured the real-world bandwidth of Nvidia's Grace Hopper superchip, with the chip-to-chip interconnect results falling well short of theoretical claims. A paper published on July 10 by researchers in the U. Read more…

Belt-Tightening in Store for Most Federal FY25 Science Budets

July 15, 2024

If it’s summer, it’s federal budgeting time, not to mention an election year as well. There’s an excellent summary of the curent state of FY25 efforts reported in AIP’s policy FYI: Science Policy News. Belt-tight Read more…

SCALEing the CUDA Castle

July 18, 2024

In a previous article, HPCwire has reported on a way in which AMD can get across the CUDA moat that protects the Nvidia CUDA castle (at least for PyTorch AI pro Read more…

Aurora AI-Driven Atmosphere Model is 5,000x Faster Than Traditional Systems

July 16, 2024

While the onset of human-driven climate change brings with it many horrors, the increase in the frequency and strength of storms poses an enormous threat to com Read more…

Shutterstock 1886124835

Researchers Say Memory Bandwidth and NVLink Speeds in Hopper Not So Simple

July 15, 2024

Researchers measured the real-world bandwidth of Nvidia's Grace Hopper superchip, with the chip-to-chip interconnect results falling well short of theoretical c Read more…

Shutterstock 2203611339

NSF Issues Next Solicitation and More Detail on National Quantum Virtual Laboratory

July 10, 2024

After percolating for roughly a year, NSF has issued the next solicitation for the National Quantum Virtual Lab program — this one focused on design and imple Read more…

NCSA’s SEAS Team Keeps APACE of AlphaFold2

July 9, 2024

High-performance computing (HPC) can often be challenging for researchers to use because it requires expertise in working with large datasets, scaling the softw Read more…

Anders Jensen on Europe’s Plan for AI-optimized Supercomputers, Welcoming the UK, and More

July 8, 2024

The recent ISC24 conference in Hamburg showcased LUMI and other leadership-class supercomputers co-funded by the EuroHPC Joint Undertaking (JU), including three Read more…

Generative AI to Account for 1.5% of World’s Power Consumption by 2029

July 8, 2024

Generative AI will take on a larger chunk of the world's power consumption to keep up with the hefty hardware requirements to run applications. "AI chips repres Read more…

US Senators Propose $32 Billion in Annual AI Spending, but Critics Remain Unconvinced

July 5, 2024

Senate leader, Chuck Schumer, and three colleagues want the US government to spend at least $32 billion annually by 2026 for non-defense related AI systems.  T Read more…

Atos Outlines Plans to Get Acquired, and a Path Forward

May 21, 2024

Atos – via its subsidiary Eviden – is the second major supercomputer maker outside of HPE, while others have largely dropped out. The lack of integrators and Atos' financial turmoil have the HPC market worried. If Atos goes under, HPE will be the only major option for building large-scale systems. Read more…

Everyone Except Nvidia Forms Ultra Accelerator Link (UALink) Consortium

May 30, 2024

Consider the GPU. An island of SIMD greatness that makes light work of matrix math. Originally designed to rapidly paint dots on a computer monitor, it was then Read more…

Comparing NVIDIA A100 and NVIDIA L40S: Which GPU is Ideal for AI and Graphics-Intensive Workloads?

October 30, 2023

With long lead times for the NVIDIA H100 and A100 GPUs, many organizations are looking at the new NVIDIA L40S GPU, which it’s a new GPU optimized for AI and g Read more…

Shutterstock_1687123447

Nvidia Economics: Make $5-$7 for Every $1 Spent on GPUs

June 30, 2024

Nvidia is saying that companies could make $5 to $7 for every $1 invested in GPUs over a four-year period. Customers are investing billions in new Nvidia hardwa Read more…

Nvidia Shipped 3.76 Million Data-center GPUs in 2023, According to Study

June 10, 2024

Nvidia had an explosive 2023 in data-center GPU shipments, which totaled roughly 3.76 million units, according to a study conducted by semiconductor analyst fir Read more…

AMD Clears Up Messy GPU Roadmap, Upgrades Chips Annually

June 3, 2024

In the world of AI, there's a desperate search for an alternative to Nvidia's GPUs, and AMD is stepping up to the plate. AMD detailed its updated GPU roadmap, w Read more…

Some Reasons Why Aurora Didn’t Take First Place in the Top500 List

May 15, 2024

The makers of the Aurora supercomputer, which is housed at the Argonne National Laboratory, gave some reasons why the system didn't make the top spot on the Top Read more…

Intel’s Next-gen Falcon Shores Coming Out in Late 2025 

April 30, 2024

It's a long wait for customers hanging on for Intel's next-generation GPU, Falcon Shores, which will be released in late 2025.  "Then we have a rich, a very Read more…

Leading Solution Providers

Contributors

Google Announces Sixth-generation AI Chip, a TPU Called Trillium

May 17, 2024

On Tuesday May 14th, Google announced its sixth-generation TPU (tensor processing unit) called Trillium.  The chip, essentially a TPU v6, is the company's l Read more…

Nvidia H100: Are 550,000 GPUs Enough for This Year?

August 17, 2023

The GPU Squeeze continues to place a premium on Nvidia H100 GPUs. In a recent Financial Times article, Nvidia reports that it expects to ship 550,000 of its lat Read more…

IonQ Plots Path to Commercial (Quantum) Advantage

July 2, 2024

IonQ, the trapped ion quantum computing specialist, delivered a progress report last week firming up 2024/25 product goals and reviewing its technology roadmap. Read more…

Choosing the Right GPU for LLM Inference and Training

December 11, 2023

Accelerating the training and inference processes of deep learning models is crucial for unleashing their true potential and NVIDIA GPUs have emerged as a game- Read more…

The NASA Black Hole Plunge

May 7, 2024

We have all thought about it. No one has done it, but now, thanks to HPC, we see what it looks like. Hold on to your feet because NASA has released videos of wh Read more…

Nvidia’s New Blackwell GPU Can Train AI Models with Trillions of Parameters

March 18, 2024

Nvidia's latest and fastest GPU, codenamed Blackwell, is here and will underpin the company's AI plans this year. The chip offers performance improvements from Read more…

Q&A with Nvidia’s Chief of DGX Systems on the DGX-GB200 Rack-scale System

March 27, 2024

Pictures of Nvidia's new flagship mega-server, the DGX GB200, on the GTC show floor got favorable reactions on social media for the sheer amount of computing po Read more…

MLPerf Inference 4.0 Results Showcase GenAI; Nvidia Still Dominates

March 28, 2024

There were no startling surprises in the latest MLPerf Inference benchmark (4.0) results released yesterday. Two new workloads — Llama 2 and Stable Diffusion Read more…

  • arrow
  • Click Here for More Headlines
  • arrow
HPCwire