Pods of Science

How to Do Science Faster with Artificial Intelligence

Episode Summary

Artificial intelligence is revolutionizing our world. You can see this everywhere from your mobile phone to the development of self-driving cars. AI is also revolutionizing the way we do science and the way we tackle problems important to our nation. That's why the Department of Energy has developed a new co-design center, known as the Center for ARtificial Intelligence-focused ARchitectures and Algorithms, or ARIAA to help solve some of the most challenging problems by employing novel artificial intelligence and machine learning techniques.

Episode Notes

Pods of Science | Episode 3 | How to Do Science Faster with Artificial Intelligence


Welcome. I’m your host, Jess Wisse. On today’s episode we’ll talking about a new research center created by the U.S. Department of Energy. Stay tuned to learn more.


JW: Pacific Northwest National Laboratory, Sandia National Laboratories. Georgia Institute of Technology. What do these have in common? They are three powerhouses in the realm of artificial intelligence, and now they are working together.

Want to know who the man is at the helm of this new collaboration? Meet, Roberto.

RG: My name is Roberto Gioiosa. I'm a senior computer scientist at PNNL in the high-performance computing group. My background is in hardware and software core design and mostly focus on the design of operating systems, runtimes, and programming model in particularly looking at emerging a future architecture both for processing, memory, and networking.

I came to PNNL in 2012 after years of other experiences in both academia and industry. Ever since I joined PNNL, I've been trying to lead efforts on software and novel hardware for computational scientists to speed up the solution to their problem and therefore solve scientific challenges.

JW: Artificial intelligence and machine learning seems to be cropping up everywhere these days. From self driving cars to your new smart phone, its everywhere. Even Alexa, Amazon’s voice assistant, is getting smarter with each passing day. Soon, she will be able to guess what you might be thinking with a new feature called Alexa Hunches.

Originally called thinking machines in the 1950s, artificial intelligence is a sub-field of computer science where machines develop the ability to think and learn on their own. Artificial intelligence, also known as AI, allows computers to perform tasks that historically could only be done by humans; think of things like visual perception, speech recognition, language translation. And that’s just the beginning.

RG: Artificial intelligence and machine learning is something that is helping us revolutionize the way we do research. Rather than starting from the top and using first principle to solve a problem, we are trying to see what the data tells us about the problem. This is something that you see every day— look at natural phenomenon and you're trying to find a correlation between what you observe and what are the reasons for that the causal relationships that are in there. In some cases, you know this naturally is complicated and it's not easy to go and have a complete understanding of what is happening without knowing anything about the entire process. What AI is doing for us is helping us do reverse engineering of natural phenomenon.

You have probably seen tons of movies about AI and how that can help, but the fundamental thing is we are looking at the data and we are trying to infer the structure of the phenomenon from the data.

JW: Roberto is the director of a new co-design center, known as the Center for Artificial Intelligence-focused Architectures and Algorithms, or (ARIAA). ARIAA is taking AI & machine learning to the next level.

RG: ARIAA is essentially a tool, a means, in which we are trying to understand what are the
requirements from our application domains. In this case, are power grid, cybersecurity, graph analytics, and chemistry, and how artificial intelligence and machine learning can support these domains to allow novel discoveries.

JW: AARIA will explore how AI and machine learning can support four areas that touch virtually every American’s life. Whether we’re aware of it or not we encounter power grid, cybersecurity, graph analytics, and computational chemistry almost every day. These are the disciplines where new medicines are created, where the fate of our online identity lies, it’s how masses of information is analyzed, and where our lights magically turn on with a flip of the switch.

RG: AI is revolutionizing our world. You see that from your mobile phone to self-driving cars and all of this. Under the umbrella of AI there is a lot— a lot of activities, a lot of different kind of AIs. I think the U.S. government recognized the importance of a coordinated strategy to solve these problems.

JW: Earlier this year the Department of Energy made a commitment to accelerate AI. Programs like AARIA are in line with President Trump’s call for a national strategy to assure AI technologies are developed to positively impact the lives of the American public.

RG: Artificial intelligence is also revolutionizing the way we do science and the way we tackle important problems being at the national security, chemistry level, and new materials. DOE and Secretary Perry have recognized the importance of this moment and they are putting together strategy. The DOE Artificial Intelligence Technology Office is the first step, but there are also other activities including ARIAA that will support DOE strategies. We are really looking into what would science look like in the future where we have infrastructure and tools that can use AI to help us.

JW: ARIAA is centered around a concept known as “co-design.” This concept, of co-design, takes into account both the capabilities of computing hardware and software. Roberto and his team must determine what types of applications will run best on a given hardware set-up, while also considering the type of hardware that will be needed when new software is created. It’s a balancing act that is never resolved and requires Roberto to constantly imagine the future of computing.

RG: If you go from the bottom up, at the hard level we are looking to understand what kind of hardware accelerator will be necessary in the future to support AI workloads, but also scientific workloads that leverage AI.

At the software level, we are trying to understand without the obstruction that needs to be put in place so that the way scientists can use our software and hardware without having to be hardware experts.

At the application level, we are trying to identify the opportunities for using AI models to either replace first principle computation or to support other computation that we have in place like the economic in a simulation producing some data in a AI framework next to it to try and understand what's happening and then having a closed loop where you can actually modify the next round of simulation.

The crosscut research is how to put all of this together to make sure that the hardware we're building, the software we are building, and the application algorithm that we are developing all make sense and they are impactful to the DOE mission, PNNL, and the society at large.

JW: Not only do all of these components have to work seamlessly together they also have to be portable, so that anybody in the world can use the design.

RG: And so what this will allow us, is to solve current problems in a much faster way. But especially it will allow us to tackle problems that today cannot be solved because they are too complex.

To do that is not just one part of the hardware/software stuff that you need to address – you need to go from the top down. What ARIAA will look at is identify these places in our application domains that may require AI machine learning support and develop the proper hardware that needs to be put in place so that this model can be accelerated and speed up. We would do that in a way that can be portable. And of course we plan to release our products, both software, hardware design, and application, as open source to the community, so any other person in the world can go and download our models, our software stack, or our architecture designs.

JW: How is Roberto going to pull off such a huge task? He’s not going to do it alone. He’s confident in his team.

RG: I think one of the strengths of this project is the people, the team we have together. Our partners in ARIAA are Sandia National Laboratories and Georgia Tech. I firmly believe that one of the strengths of this project and one of the reasons why it was selected is because of the team we have together.
JW: And it’s not just the fact that Roberto has some of the brightest minds in computing working at ARIAA, he credits the ability to do this to those who came before him.

RG: This is a very large problem that we are trying to solve, and it wouldn't be possible if we didn't have experience in doing similar form of activities, and in artificial intelligence in particular. A lot of the credit for having this project started that goes to the fact that PNNL, Sandia, and Georgia Tech and have invested in the past on developing infrastructures, people, and knowledge.

JW: Thanks to Roberto’s tireless efforts and the newly developed ARIAA scientists across disciplines will have the ability to do carry out their research even faster. Roberto likes to call it the science of the future.


JW: Thanks for listening to Pods of Science. Want to learn more? Follow us on social media at PNNLab. We're on Twitter, Instagram, Facebook, and LinkedIn. You can also visit our website at pnnl.gov. Thanks for listening.