 Hello everyone and welcome back to Dallas where we're live from supercomputing. My name is Savannah Peterson, joined with my co-host David and we have a rocket of a show for you this afternoon. The doctors are in the house and we are joined by NASA ladies and gentlemen. So excited, please welcome Dr. Dan Duffy and Dr. Bill Putnam. Thank you so much for being here guys. I know this is kind of last minute. How's it to be on the show floor? What's it like being NASA here? Oh, it's exciting. We haven't been here for three years so this is actually really exciting to come back and see everybody to see the show on the floor, see the innovations that have happened over the last three years. It's pretty exciting. Yeah, it's great. And so because your jobs are so cool and I don't want to even remotely give even too little of the picture or not do it justice, could you give the audience a little bit of background on what you do as I think you have one of the coolest jobs ever? You too, Bill. I appreciate that. I run a high performance computing center at NASA Goddard for science. It's high performance information technology. So we do everything from networking to security to high performance computing to data sciences, artificial intelligence and machine learning is huge for us now. Large amounts of data, big data sets, but we also do scientific visualizations and then cloud and commercial cloud computing as well as on-premises cloud computing. And quite frankly we support a lot of what Bill and his team does. Bill, why don't you tell us what your team does? So I'm an earth scientist. I work as the associate chief at the global modeling and assimilation office. And our job is to really maximize the use of all the observations that NASA takes from space and build that into a coherent, consistent physical system of the earth, right? And we're focused on utilizing the HPC that Dan and the folks at the NCCS provide to us to the best of our abilities to integrate those observations. You know, on time scales from hours, days to seasonal, to monthly time scales. That's the essence of our focus at the GMEO. Casual, modeling all of NASA's earth data. That in itself as a sentence is pretty wild. I imagine you're dealing with a ton of data. Oh, massive amounts of data, yes. I mean, as much as one probably could now that I'm thinking about it. I mean, and especially with how far things have to travel. Bill, sticking with you just to open us up. What technology here excites you the most about the future? And that will make your job easier. Let's put it that way. To me, it's the accelerator technologies, right? So the limiting factor for us as scientists is how fast we can get an answer. And if we can get our answer faster through accelerated technologies, with the support of the NCCS and the computing centers, but also the software engineers enabling that for us, then we can do more, right? And push the questions even further. You know, so once we've gotten fast enough to do what we want to do, there's always something next that we want to look for, so. I mean, at NASA, you have to exercise such patience. Whether that be data coming back, images from a rover doesn't matter what it is. Sometimes there's a lot of time, days, hours, years, depending on the situation. I really admire that. What about you, Dan? What's got you really excited about the future here? So Bill talked about the accelerator technology, which is absolutely true and is needed to get us not only to the point where we have the computer resources to do the simulations that Bill wants to do and also do it in an energy efficient way, but it's really the software frameworks that go around that and the software frameworks, the technology, the dealing with how to use those in an energy efficient and most efficient way is extremely important. And that's some of the, you know, that's what I'm really here to try to understand better about is how can I support these scientists with not just the hardware, but the software frameworks by which they can be successful? Yeah, we've had a lot of kind of philosophical discussion about this, the difference between the quantitative increases in power in computing that we're seeing versus the question of whether or not we need truly qualitative changes moving forward. Where do you see the limits of, you know, if you're looking at the ability to gather more data and process more data more quickly, what you can do with that data changes when you're getting updates every second versus every month seems pretty obvious. Is there a, is there, but is there, is there a near term target that you have specifically where once you reach that target, if you weren't thinking ahead of that target, you'd kind of be going, okay, well, we solved that problem. We're getting the data in so fast that you can ask me, what is the temperature in this area? And you can go, oh, well, an hour ago, the data said this. Beyond that, do you need a qualitative change in our ability to process information and tease insight out of chaos? Or do you just need more quantity to be able to get to the point where you can do things like predict weather six months in advance? What are your thoughts on that? Yeah, it's an interesting question, right? And you ended it with predicting weather six months in advance and actually I was thinking the other way, right? I was thinking going to finer and finer scales and shorter time scales when you talk about having data more frequently, right? So one of the things that I'm excited about as a modeler is going to higher resolution and representing smaller scale processes. At NASA, we're interested in observations that are global, so our models are global. And we'd like to push those to as finer resolution as possible to do things like severe storm predictions and so forth. So the faster we can get the data, the more data we can have in that area would improve our ability to do that as well. And your background is in meteorology, right? Yes, I'm a meteorologist here. Excellent, okay, yeah, yeah. So I have to ask a question and I'm sure all the audience cares about this. And I went through this when I was talking about the GOES satellites as well. What is it about weather that makes it so hard to predict? Oh, it's the classic chaos problem, the butterfly effect problem. And it's just true. You always hear the story of a butterfly in Africa flaps its wings and the weather changes in New York City. And computers are an excellent example of that, right? So we have a model of the Earth, we can run it two times in a row and get the exact same answer, but if we flip a bit somewhere, then the answer changes 10 days later, significantly. So it's a really interesting problem. So do you have any issue with the fact that your colleague believes that butterflies are responsible for weather? No, I don't. Is it responsible for climate change? Does that bother you at all? No, it doesn't. As a matter of fact, they actually run those butterfly-like experiments within the systems where they do actually flip some bits and see what the uncertainties are that happen out seven, eight, nine days out in advance to understand exactly what he's saying, to understand the uncertainties, but also the sensitivity with respect to the observations that they're taking. Yeah, it's fascinating. It is. That is fascinating. I'm sticking with you for a second, Dan. So you're at the Center for Climate Simulation. Is that the center that's going to help us navigate what happens over the next decade? Okay, so no one center is going to help us navigate what's going to happen over the next decade or the next 50 or 100 years, right? It's going to be everybody together. And I think NASA's role in that is really to pioneer the models that Bill and others are doing to understand what's going to happen in not just the seasonal, the sub-seasonal, but we also work with GISS, which is the Goddard Institute for Space Studies, which does the decadal and the century-long studies. Our job is to really help that research and understand what's happening with the client, but then feed that back into what observations we need to make next in order to better understand and better quantify the risks that we have, to better quantify the mitigations that we can make to understand how and effect how the climate is going to go for the future. So that's really what we're trying to do. We're trying to do that research to understand the climate, understand what mitigations we can have, but also feedback into what observations we can make for the future. Yeah, and what's the partnership ecosystem around that? You mentioned that it's going to take all of us. I assume you work with a lot of partners, probably both of you. Yeah, I mean, obviously the federal agencies work huge amounts together. NASA and NOAA are huge partnerships, USGS, huge partnerships, DOE. We've talked to DOE several times this week already. So there's huge partnerships that go across the federal agency. We work also with the Europeans as much as we can, given the sort of the barriers of the countries and the financials, but we do collaborate as much as we can with it. And the nice thing about NASA, I would say, is all the observations that we take are public. They're paid for by the public. They're public. Everybody can tell them. Anybody can tell them. They're around the world. So that's also, and they're global measurements, as Bill said. They're not just regional. Do you have, do you have specific, when you think about improving your ability to gain insights from data that's being gathered? Do you set out specific milestones that you're looking for? Like, I hope by June of next year, we will have achieved a place where we are able to accomplish X. Do you put, so what milestones do we have here? So, yeah, I mean, are you sort of kept track of that way? Do you think of things like that, like very specific things, or is it just so fluid that, as long as you're making progress towards the future, you feel okay? No, I would say we absolutely have milestones that we like to keep in track, especially from the modeling side of things, right? So, whether it's observations that exist now that we want to use in our system, milestones to getting those observations integrated in, but also thinking even further ahead to the observations that we don't have yet. So we can use the models that we have today to simulate those kind of observations that we might want in the future that can help us do things that we can do right now. So those missions are aided by the work that we do at the GMEO and the NCCS. Okay, so if we extrapolate really to the what if future is really trying to understand the entire Earth system as best as we can. So all the observations coming in, like you said, in near real time, feeding that into an Earth system model and to be able to predict short-term, mid-term, or even long-term predictions with some degree of certainty. And that may be things like climate change or it may be even more important, shorter-term effects of severe weather, which is very important. And so we are trying to work towards that high-resolution, immediate-impact model that we can really share with the world and share those results as best we can. Yeah, I have a quick follow-up on that. I bet we both did. So if you think about AI and ML, artificial intelligence and machine learning, something that people talk about a lot, there's the concept of teaching a machine to go look for things, call it machine learning. A lot of it's machine teaching. We're saying, you know, hit the rack on this side with a stick or the other side with a stick to get it to kind of go back and forth. Do you think that humans will be able to guide these systems moving forward enough to tease out the insights that we want? Or do you think we're going to have to rely on what people think of as artificial intelligence to be able to go in with this massive amount of information with an almost infinite amount of variables and have the AI figure out that, you know what? It was the butterfly. It really was the butterfly. We all did models with it, but you understand the nuance that I'm saying? It's like, we think we know what all the variables are and that it's chaotic because there are so many variables and there's so much data, but maybe there's something we're not taking into account. I'm sure that's absolutely the case. And I'll start and let Bill jump in here. There's a lot of nuances with AIML. And so the real approach to get to where we want to be with this earth system model approach is a combination of both AIML train models as best as we can and as unbiased way as we can. And there's a big conversation we have around that, but also with a physics or physical based model as well. Those two combined with the humans or the experts in the loop, we're not just going to ask the artificial intelligence to predict anything and everything. The experts need to be in the loop to guide the training and as best as we can, as we can in an unbiased, equitable way, but also interpret the results and not just give over to the AI, but that's the combination of that earth system model that we really want to see the futures, combination of AIML with physics based. But there's an obvious place for AI and ML in the modeling world. And that is in the parametrizations of the estimations that we have to do in our systems, right? So when we think about the earth system and modeling the earth system, there are many things like the equations of motions and thermodynamics that have fixed equations that we know how to solve on a computer, but there's a lot of things that happen physically in the atmosphere that we don't have equations for and we have to estimate them and machine learning through the use of high resolution models or observations and training the models to understand and represent that. That's the place where it's really useful for us. There's so many factors. But we have to make sure that we have the physics in that machine learning in those training sets. So physics-informed training isn't very important. So we're not just going to go and let a model go off and do whatever it wants. It has to be constrained within physical constraints that the experts know. Yeah, and with the wild amount of variables that affect our earth, quite frankly. Yeah, which is, geez, which is insane. My God, so what technology or what advancement needs to happen for your jobs to get easier faster for our ability to predict to be even more successful than it is currently? You know, I think for me, the vision that I have for the future is that at some point, you know, all data is centrally located, it's centrally shared. We have our applications are then services that sit around all that data. I don't have to sit as a user and worry about, oh, is this all this data in place before I run my application? It's already there, it's already ready for me. My service is prepared, and I just launch it out on that service. But that coupled with the performance that I need to get the result that I want in time. And I don't know when that's going to happen. At some point it might. You know, I don't know. It's hurting for you. But that's... So there are a lot of technologies that we can talk about. What I'd like to mention is open science. So NASA is really trying to make a push and transformation towards open science. 2023 is going to be the year of open science for NASA. And what does that mean? That means a lot of what Bill just said is that we have equity and fairness and accessibility. And you can find the data. It's findability, it's fair data, you know, FAR, findability, accessibility, reproducibility, and I forget what the I stands for. But these are tools and things that we need to, as a computing centers, and including all the HBC centers here, as well as the scientists need to support to be as transparent as possible with the data sets and the research that we're doing. And that's where I think is going to be the best thing is if we can get this data out there that anybody can use in an equitable way and as transparent as possible, that's going to eliminate, in my opinion, the bias over time because mistakes will be found and mistakes will be corrected over time. I love that. Yeah, the open source science ended this. Now it's great and the more people that have access, people I find in the academic world, especially people don't know what's going on in the private sector in vice versa and so I love that you just brought that up. Closing question for you because I suspect there might be some members of our audience who maybe have fantasized about working at NASA. You've both been working there for over a decade. Is it as cool as we all think it is on the outside? I mean, it's definitely pretty cool. You don't have to be modest about it. I mean, just being at Goddard and being at the center where they build the James Webb telescope and you can go to that clean room and see it. It's just fascinating. So it's really an amazing opportunity. So NASA Goddard as a center has information technologists, it has engineers, it has scientists, it has support team members. We have built more things, more instruments that have flown in the space than any other place in the world. The James Webb, we were part of that, part of a huge group of people that worked on James Webb and James Webb came through and was assembled in our clean room. It's one of the biggest clean rooms in the world. And we all took opportunities to go over and take selfies with this. As they put those mirrors on them, it was awesome. It was amazing. And to see what the James Webb has done in such a short amount of time, the successes that they've gone through is just incredible. Now, I'm not a part of the James Webb team, but to be at the same center, to listen to scientists like Bill talk about their work, to listen to scientists that talk about James Webb, that's what's inspiring and we get that all the time. And to have the opportunity to work with the astronauts that service the Hubble telescope. These things are just- You're literally giving me goosebumps right now. I'm sitting over here just- Just an amazing opportunity. Well, Dan, Bill, thank you both so much for being on the show. I know it was a bit last minute, but I can guarantee we all got a lot out of it. David and I both, I know I speak for us and the whole CUBE audience, so thank you. We'll have you anytime you want to come talk science on the CUBE. Thank you all for tuning in to our super computing footage here live in Dallas. My name is Savannah Peterson. I feel cooler having sat next to these two gentlemen for the last 15 minutes and I hope you did too. We'll see you again soon.