[Scott] Next on "Energy Switch," we'll hear how artificial intelligence could affect energy and climate.

- We're not just creating these super computers just because, we need to understand how these materials can help with the longevity of some of our critical infrastructure components.

- And you can take this further and ask, "How do I plan for the future?"

Energy demand is going to grow with all the extreme weather events that we are seeing.

So 50 years from now, how do we plan for this increased energy demand while also enabling net zero?

[Scott] Coming up on "Energy Switch," could AI change energy?

[Announcer] Funding for "Energy Switch" was provided in part by The University of Texas at Austin, leading research in energy and the environment for a better tomorrow.

What starts here changes the world.

And by EarthX, an international nonprofit working towards a more sustainable future.

See more at earthx.org.

- I'm Scott Tinker and I'm an energy scientist.

I work in the field, lead research, speak around the world, write articles and make films about energy.

This show brings together leading experts on vital topics in energy and climate.

They may have different perspectives, but my goal is to learn and illuminate and bring diverging views together towards solutions.

Welcome to the "Energy Switch."

Artificial intelligence is everywhere in the news today and increasingly in our lives.

In energy and climate, AI is already being used to better model our electricity demand, so we can better match it with on-demand generation.

To better model wind and solar resources, so we can better predict their electricity generation output and to better model projected climate impacts.

We'll talk about all this and more with my guests, Pamela Isom is the director of the Artificial Intelligence and Technology Office for the US Department of Energy.

Helping the government contribute and respond to the development of AI.

With her is Dr. Anima Anandkumar.

She's a professor of computing at the California Institute of Technology and senior director of Machine Learning Research at NVIDIA.

We'll hear from these two experts on this episode of "Energy Switch."

How Could Artificial Intelligence Change Energy?

There's a lot of conversations about artificial intelligence today.

- Yeah.

- We wanna talk about it as it relates to energy and just what is artificial intelligence?

- Yeah, first of all what is intelligence, right?

So intelligence is the ability to learn and adapt to changing environmental conditions.

And so when there is learning from data and there is now insights based on that, a machine is able to either make decisions or assist humans in making decisions.

- And that's the ultimate goal is to assist humans in making the decisions.

- Right, and how is that different, or is it from machine learning?

What is this?

Are there differences there?

What does that mean?

- Yeah, machine learning is a subfield of AI because intelligence has aspects of learning, but then you're also adapting based on what you've learned.

And so the focus of like, how do we learn given a data set?

And how do we design algorithms that clean the right insights and perform a given objective or a task based on that?

- Okay, you said algorithm.

What's that mean?

[Dr. Anima chuckles] - You can think of it as a program or a set of instructions to do as task.

And in case of machine learning, that's an algorithm that says how we should look at this data set.

How should we process this?

What are the set of procedures to turn data into insights?

- Oh, interesting.

- And the only thing that I'll add to that is with the algorithms, it's like a recipe, but it's bidirectional.

[Scott] Okay.

- So with the recipe you are putting the ingredients in the container.

- Right?

- And that's it.

You're just doing what you're told.

- Right?

- With the algorithm, the algorithm combined with the data creates the model, and it's the algorithm that's the recipe, the instructions, and then it's actionable.

- Interesting, it's a long way from when I was in graduate school and I had a stack of punch cards.

We just hope they ran well.

[Scott laughing] It was one direction, hardwired, and you get it or you don't.

- You know, look at how much our computing abilities have grown.

You know, even our iPhone is now much more powerful than the supercomputers of '80s or even '90s.

And so that exponential growth in computing has been a big driver in the growth of AI.

- Okay, so it's not just the codes, but it's the actual hardware that's able to bring that power to it as well.

- Mm-hmm.

- The way I frame it as AI being a trinity of data algorithms and computing infrastructure.

[Scott] Okay, okay.

- So all these advances came together.

- So is it better?

I mean, why is this better than humans?

- The amount of data that is in the world today, it's explosive.

We talked about big data, now we're past big data.

I was reading this morning that, I think we're up to 2.5 quintilian bytes of data that's generated per day.

- Jeez!

- And so that's the beauty of the AI, is that it can take the data, process the data, sort through information, determine, for instance, if there are similarities, if there are commonalities, if there are patterns that we may not even recognize and decisions are being made with this information by the AI in seconds.

[Scott] Right?

- Even if humans were to be able to process all the data, we may not be able to always derive insights.

For instance, if I gave you all the historical data of the weather.

You know, we've been recording it for decades now.

We have terabytes of data or even more.

Can you tell me what the weather will be over the next week?

It depends, in Texas maybe so.

[Dr. Anima] - Bloody hot.

- But whether a hurricane will make a landfall or not, that's very hard to predict, so- - Yeah, and where and- - Yeah, so what AI is able to do is also overcome these uncertainties and give you, like, accurate risk assessments, and that's very hard for humans to grapple with, uncertainty and risk.

- Yes it is.

- And that's really important when we get to things like energy management and making sure that the grid is gonna perform, and looking for malfunctions that the operator may not detect.

But with the AI capability in the mix, this really adds value.

- Let's get as specific as we can for our listeners.

You know, where's an area where either we are currently or really close to have the potential to make a big difference on the energy side with AI.

- Yeah.

- Can you get tangible with me?

- Yeah, I think AI has already showing promise in being able to model better electricity demand, plan for renewables by first creating a good digital twin of our weather forecasting systems.

And with that we can plan how well the renewables can be produced.

And also modeling how that wind gets converted to energy.

What is that transformation process?

AI is able to predict that.

And with that we can then ask, what about the demand side?

Think of heat waves, there's obviously going to be more demand, right?

So again, weather predictions can help us forecast for consumer demand in addition to also the historical data about consumer demand.

[Scott] Yeah.

- And with that can we ask, is the grid resilient?

Is there a risk of outage now?

How should I change my ratio of renewables versus fossil fuels.

[Scott] Right?

- How do I plan for this production effectively?

- Right, so you're better able to understand the intermittency of solar and wind, and how to back them up with batteries and load following or peakers, and that kind of thing.

Just a lot more intelligent grid essentially.

- Yes.

- On the production side and on the demand side.

- Exactly.

- Right, we're going to use digital twins, and the next evolution of digital twins as you were talking about, 'cause we're into the next evolution now, using simulated data, using real data, using scenarios.

How do we anticipate what could happen?

- Yeah.

- How do we get that resilience that we need?

And it's just ripe for AI.

- Yeah.

- And you can look for alternatives.

So the AI can say, "Okay, well maybe wind is not the answer.

Maybe solar is not the answer in this particular situation."

One of the biggest opportunities is to look at how to balance that supply and that demand, and that generation of the electricity, and ensure that it gets to the consumer.

AI can help make sure that that happens as quickly and as efficiently as possible.

And then the other thing that's cool about AI, is it's able to look at are there potential issues with, for instance, the transformers?

Are there potential issues with the solar panels?

- And you can take this further and ask, how do I plan for the future?

Right, so today we can construct a wind farm in a certain location because it's high wind, but what will happen, let's say two decades from now with the climate change?

- Right?

- Is that going to die down?

And if so, do I plan ahead?

- Right?

- You know, and the energy demand is going to grow with all the extreme weather events that we are seeing.

So 50 years from now, how do we plan for this increased energy demand while also enabling net zero, right?

And there's a lot of technologies that need to be discovered, but can we use climate simulations using AI, couple them with weather forecasts using AI, and then couple that with wind farm digital twin using AI, and predict energy demand sometime even in the future, not just today.

- Sure, how does this knowledge about weather and the electricity demand, and solar and wind intermittency in it.

How does that benefit the whole overall energy picture?

What are the upside there?

- It's gonna help with the production.

It's gonna help with the production, it's gonna help with the maintenance.

It's gonna help with the distribution of energy.

We also have to think about the underserved communities.

These are all gonna be valued generators for the underserved population.

The energy's gonna get distributed to where it's needed.

- That interesting.

- Yeah, and ultimately right, so in addition to energy production, we need to think about energy transition.

How do we head towards the net zero goal?

And for that, there are several paths like increasing the renewable production means, being able to have better accurate weather predictions, but also being able to better model those processes.

Increase the efficiency of wind, solar, and as well as storage.

You know, how do we have better batteries, right?

Whether hydrogen, how do we use that as a clean source for storage?

And also ultimately, almost most scientists agree that we would need to target for negative emissions.

How do we extract carbon dioxide?

And so one of the projects we've been working in collaboration with Stanford University, it's called CCSNet.

It's using AI to model how carbon dioxide interacts with water deep underground.

So when we extract this carbon dioxide and we pump it into reservoirs underground, can we model what is the rock structure there?

How much pressure can it tolerate?

And so we are using AI to model these complex multiphysics processes, and doing it much faster.

Several orders of magnitude faster than what recurrent physical modeling can do.

- Right, how about energy equipment?

Like, you know, we have system failures and maintenance.

Can it help in these areas as well?

Have you done anything along that line?

- Yeah, yeah.

A lot of modeling that we've done with AI assesses, for instance, fractures and impact on materials, right?

Like when you impinge on a material, how plastic is it?

How much is it gonna deform?

It's very hard to predict that using only course calculations, because you need to go to the fine scale and ask what's happening at the molecular level.

- Yeah.

- And so what we are doing AI is to speed up computations at every scale, and sometimes even going to quantum scale.

- So is this the kind of thing where you're actually seeing this ahead of time incrementally and able to process like a joint failure or... - Yeah, exactly.

You can model those failures much more accurately and maybe in the process even design better materials.

[Scott] Yeah.

- And that's why you have things like the super computers.

We're not just creating these super computers just because, we need to understand how these materials can help with the longevity of some of our critical infrastructure components.

Not only the pipeline, but I think about the grid transmission line.

- Yeah, sure.

- And like Pam said, having loud super computing facilities are so critical.

Right, and then democratization of that computing to everyone.

- Yeah, yeah.

- So how do we ensure that researchers can easily access that?

But on top of super computers, what we also need is more investment in Edge AI.

So how do we miniaturize the sensors, and how do we place them on different equipment- - Edge A.

- Edge AI.

- Edge.

- Okay.

- The Edge.

- It's on the Edge.

So there is now AI on the Edge rather than in the Cloud.

And so with that, we know they are unlimited battery power.

So how do we ensure they can process information effectively?

When should they send it to the Cloud?

How can they monitor these processes or plants effectively?

- Right?

- And so that interaction between the Edge and the Cloud is also very important.

- Yeah, that's exciting.

I mean, I'm sure to me and others it sounds almost amazing, [laughing] but what's holding some of this back today?

What are some of the challenges for moving forward at a faster pace?

- I would say these are early days, right?

AI has barely been a decade since we saw the promise of neural networks in understanding images.

Now we are understanding language well and we are getting to AI for science, AI for climate, AI for weather.

And so a lot of that I think is human like.

But like x AI scientists like me collaborating with domain scientists.

- So bringing what you do to many different kinds of disciplines.

- Yes.

- Right, interesting.

- And I would say that, you see more use of AI especially in the health space, in the financial services space, and of course in the government.

Energy is where the opportunity is.

- I mean, what are the limits?

How much more efficient can this become?

- I think we'll be always limited by computing.

For instance, if I want to try to predict the properties of materials at the quantum level, to do that accurately even with just a 100 electrons would take longer than the age of the universe on current supercomputers.

- Say that again?

[Dr. Anima laughing] - Age of the universe, which is... [Dr. Anima laughing] - Really, to do that accurately?

- Yeah, to do that accurately.

- Yeah, it would take longer than 15 trillion years or something.

- Whichever, whatever that may be, yes.

And so, we have enormously difficult problems to fight and no amount of computing is going to be enough.

- Right.

- So the question is, how do we bring in our intelligence and use AI that can pick out the nuances?

Maybe we don't have to do it so brute force.

Maybe AI finds a shortcut.

And so how do we enable it to find those shortcuts effectively?

- Is it limited by the input algorithms that we provided, or actually are we to a point now where it can adapt its own algorithms from what it's learning?

Where are we there?

- I would say we are at the cusp of it but not it fully there.

Our algorithms are able to derive insights, but there's still shortcomings.

For instance, we've looked at all the historical weather data, we can then predict weather pretty well using AI.

But if I were to ask about weather a 100 years from now, - Right.

- Under climate change conditions, such data doesn't exist.

- Sure.

- So how do we then predict scenarios where data isn't available, but we do know laws of physics.

We can find these extrapolations, but can AI do that well?

- Right.

- And that's a challenge.

- Which goes to the historical dependency that AI leans on today, we wanna move AI to the place where rather than the velocity and the amount of data, representative data.

- Interesting.

- And the hope is if AI can learn features that can overcome these constraints of computation.

If we can speed up and still be accurate in predicting what the climate is going to be with much more confidence than what we have today.

- Yeah.

- That would have enormous implications.

- Right?

- Yeah.

- What are some of the other things we're facing?

- There's always that concern about the trust of the AI and making sure that the AI itself is high quality and never taking that for granted.

So that when we make recommendations, the uncertainty level is really, really low.

'Cause if you lose confidence in the AI, if people lose confidence, they won't use it.

And then the last thing I'll say is, the impacts to the underserved communities.

So oftentimes the communities are left out.

The data sets just aren't considered.

For instance, if you don't have GPS and you're building a model, and it's based on traffic patterns, and maybe those traffic patterns are based on GPS data-- [Scott] Right?

- You're kind of left out.

[Scott laughing] And so that's the reputation that we need to overcome.

- None of this comes without an energy cost in terms of use.

I mean how much energy are we talking about here as we ramp up the compute power and data acquisition and all this?

How much energy?

- We know that the data centers and the compute power, there's been designs and deliberate activities to reduce the burn and the load.

It's somewhere between three and four percent at this point.

- I did not know that.

- Some of the thermostats and things that we're putting in the smarter buildings, now that is generating more of a benefit.

So that's more along the lines of around 40%.

- Right, I would think- - And then supercomputers are not using as much energy as some of the newer ones, as the ones that were built before today.

- For the same computer power?

- For more.

- Yeah, yeah, interesting.

- For more compute power.

- So the rebound effect has slowed down-- - And we can use AI to further make this efficient, right?

Like think of scheduling.

We have many users trying to get onto the Cloud or the supercomputer.

How do we effectively manage the workloads.

- Let's talk just about your visions a little bit for the future.

Where do you see it making it the biggest difference in the next few years?

- I think that definitely is gonna make a difference with climate.

There's a wonderful opportunity to do more around predictions, around forecasting, around simulating situations like ice sheets, for instance.

So it's one of our labs is working on that.

And I think in the EV space.

- Yeah.

- To provide power generation for homes when they're not using the vehicles.

[Scott] Right?

- So take advantage of getting the price points down by utilizing the power that's generated for those vehicles.

- Yeah.

- I really agree with Pam and what I hope is AI can help us build much more accurate digital twins of lots of physical processes, right?

For instance, even the process of rainfall.

When rainfall occurs is still hard to predict.

- Right?

- Pam mentioned ice sheets earlier.

You know, people may be surprised but it's still, we don't have a good model for how they develop and how much do they impact climate change?

So there are so many processes on this planet if you think about, right?

Like how much are the plants really contributing to right negative emissions?

How much carbon dioxide are they absorbing?

Can we predict that from satellites?

So coming up with those kind of new techniques and using AI to speed up all these processes, right?

And can we do much better simulations and digital twins?

That's my hope.

- Yeah, interesting.

Every form of energy has impacts on the environment.

Some atmospherically, some mining, some dumping, et cetera.

Can AI play a role here?

Can AI say, "Hey, in a life cycle analysis of trade offs between all these things."

can we really start to optimize better?

- Yes.

- The whole's energy system.

- Absolutely.

- And that's where the digital twins help, right?

For instance like predicting the life cycle of a battery or all these materials, and what is also the impact of finally the end of cycle.

You know, how do we discard them and what is the impact that has on the environment?

- Recycling is really important.

- Yeah, yeah.

But it's not done because it's cheaper to make new.

- AI can't help with that.

So AI can then predict or help to predict the impacts, first of all the impacts of an investment.

The longevity and the sustainability of the investment.

And then help us to look at what are the best ways to use the materials once the materials reach the point of degradation.

But you need AI just because the manual process, it's too much.

So that's where a capability like AI will come along.

- And the hope is once we can have faster simulations, there's much less of a barrier to bring this into planning.

[Pamela] Yeah.

- Yeah, yeah.

- And the other aspect is also mining for essential minerals and these rare earths that are... And so that's where AI could also help.

First in analyzing the surfaces and also in terms of autonomous mining.

So can you make it environmentally less impactful that way.

- And human rights potentially too which is such a huge issue.

- Right, it is.

- How do you see a government role in all this?

What are the things we can agree on as almost across parties.

- I think that the government's role in this is to continue with the research to help us to get where we need to go.

Continue to showcase some of the things that are happening.

Because some of that can be so complicated to the consumer.

I mean, you want them to really get why this is important.

- Final thoughts, you have been wonderful visit.

I've really enjoyed it.

Final thoughts you want to share?

- You know, I am looking forward to tomorrow.

I'm so looking forward to the advancements in the grid because of the things that we're doing.

- Yeah.

- And I'm so proud of the attitude that we have as a society to make the world cleaner.

- Yeah.

- I'm so proud to be a part of the Department of Energy because we are really pushing that message as well.

- Your optimism is contagious, so I appreciate it very much.

- To me indeed I think what I wanna emphasize is the democratization, right?

So the models we are building.

Whether it's for weather, whether it's for wind farms, we are open sourcing them at NVIDIA.

We are giving this out to the community and what we really want is people to be able to use these models and do other downstream tasks or make them better.

And that's where in collaboration with national labs, with universities, with public private-partnerships, how do we solve this very challenging problem?

And I think AI is, in a way a great unifier, because people from so many different disciplines can come and talk to one another.

And me as an AI scientist, I'm always like a kid in a candy store because I'm always learning about so many cool new things and asking how AI can be part of that.

- Really, I've enjoyed our visit.

Appreciate you for being here.

Scott Tinker, "Energy Switch."

We heard that artificial intelligence is being used increasingly to better predict weather, which can allow us to better understand how different forms of electricity generation will produce energy, and how to better balance them within the system to keep power flowing to users.

How to better model the wear on infrastructure to remedy problems before they happen.

How to model and understand projected climate impacts, and how to search for and optimize efficiencies across the system.

Those energy savings should more than offset the electricity that AI systems consume.

Today, AI is limited by computing power, and its historical data sets may not represent historically underserved communities potentially leaving them out of its models.

But as computers become more powerful and more efficient, and data sets and the algorithms that analyze them continue to improve, AI should become more and more valuable in helping human users make better decisions.

[dramatic music] ♪ ♪ ♪ ♪ [Announcer] Funding for "Energy Switch" was provided in part by The University of Texas at Austin, leading research in energy and the environment for a better tomorrow.

What starts here changes the world.

And by EarthX, an international nonprofit working towards a more sustainable future.

See more at earthx.org.