PRES. HARROZ: Hi, I'm Joe Harroz, President of the University of Oklahoma. I want to welcome you to our Conversations with the President. This platform gives me the chance to talk to some of the great people who make OU so special. Make sure you are subscribed to Conversations with the President, and you'll be the first to know when new episodes are released. Let's get started.
All right first thing first. I want to thank Dr. Karlos Hill and Marilyn Luper Hildreth for joining us on the last episode. If you haven't watched it yet, please do. They are absolutely stunning. So please, please listen to that. Welcome back to Conversations with the President.
The University of Oklahoma changes lives in so many ways and we talk a lot about change in the lives of students and we talk a lot about the impact that they have. One of the things we don't talk nearly enough about is the impact that our faculty and their research has not just on Oklahoma, but on the entire nation and the world, those that are foremost thought leaders, and that brings us really to our guest today who is one of those foremost thought leaders, a great intellectual and someone having an impact in the world.
Let me go over a couple of the qualifications that you have. We will talk about these more in‑depth, but this is a really exciting show. Dr. Amy McGovern, Dr. McGovern came here in 2005. I still got you by a few years here at OU and by the looks of it, about 20 years of age. The year full – the Lloyd G and Joyce Austin presidential professor, and you have dual appointments in both the School of Meteorology and the School of Computer Science. You are also the director of the National Science Foundation AI, Artificial Intelligence Institute for Research and Trustworthiness in trustworthy AI in weather climate and coastal oceanography. There is an acronym for that to save us time in the future, which is AI2ES. We are absolutely thrilled that you're here. Your research focuses on developing and applying machine learning and data mining methods for real‑world application and with a special interest in high‑impact weather. So it's great to have you here. Thanks for joining us.
DR. AMY MCGOVERN: Thank you. I'm excited to be here.
PRES. HARROZ: So this is -- there is no topic. I guess 2023 hit last year and all anyone in the popular press was talking about was artificial intelligence. And I am no Sam Altman, which will be revealed very clearly in the line of questioning that I have, but it is exciting, and so often when we talk about AI, it seems like we are talking about one of the extremes. Either it is the worst thing that ever happened, right, generative artificial intelligence, machine learning. Either it’s the worst thing in the world and will end humanity or it's the greatest thing in the world that will change everything. Before we get into the specifics of your area of work, can you sort of give us your background on AI and how you think about it?
DR. AMY MCGOVERN: Background meaning where I started in AI?
PRES. HARROZ: Yeah, where you started in AI, and then, you know, in the national and global conversation around artificial, you know Davos, two weeks ago that is all they spoke about. Kind of give us your thoughts, give me your thoughts on whether AI is a good thing or a bad thing for humanity.
DR. AMY MCGOVERN: That is an excellent question. I have been doing AI since before AI was cool.
PRES. HARROZ: So since before 2023?
DR. AMY MCGOVERN: Exactly. Sometimes people are surprised to discover that AI was actually in existence like 50 years ago. I mean not that long, but you know, I think that it is really you are right it has burst into the national conversation entirely it’s going to save us all or it is really bad because these are all the bad things that are happening with that, and I don't think either of those views is the right answer. We need to be really careful about our applications with it, but I think AI really is going to transform our ability to do a lot of things in the world. So ‑‑ and we can talk about those specifics about what it will do, and we can talk about them for weather. One of the questions I get asked all the time is are you trying to replace our jobs? I think that is what you see in the news is a lot of people saying is AI here to replace our jobs and my answer really is no. We are trying to help augment what you can do. We are trying to give you new capabilities. We are trying to help focus on the parts. Give you pieces that can help you focus on the parts humans are good at and we are giving you new tools that can help you with the parts that are perhaps mundane or just better forecast that you can then put together in a new way.
PRES. HARROZ: Absolutely. But there are hazards that could come with AI, are there not?
DR. AMY MCGOVERN: There definitely are.
PRES. HARROZ: What are the biggest hazards that you see?
DR. AMY MCGOVERN: With AI in general, I think those get covered in some of the national conversations. A lot of the questions of letting the AI do your work for you and unnecessarily checking, like ChatGPT, it can write for you, and it can write somewhat convincingly, but it hallucinates convincingly, it makes up things. And if you are not understanding that it is making things up, I think the biggest danger is exactly that that you are believing things you shouldn’t believe. So false pictures, false video, false writing, false news, all of that fake stuff I think is really one of the biggest dangers. It could really change the way that we are viewing the world. If we are getting ‑‑ if our news that’s coming at us is full fake news, and it is also filtered by the lens of what the AI thinks we want to look at, we are just going to end up in this thinking about one thing without really understanding and looking at the broader context and that is a fear.
PRES. HARROZ: Yeah, no it’s like you said, as with everything, it is neither all good nor is it all bad. So why don't you, since you are a pioneer in AI and since you have been doing it much longer than many folks at any, certainly a national level consciousness about artificial intelligence except for sci‑fi movies, tell us about your area of research, and about the institute.
DR. AMY MCGOVERN: So I've been working since I got to OU, I've been working on AI for weather, trying to improve our understanding and our prediction of weather events, and I’ve particularly been looking at severe weather because that is what we have in Oklahoma, right. I did not grow up here and when I moved here, no one told me that we would be on roof number four in 19 years.
PRES. HARROZ: You generally have four and a half years per roof.
DR. AMY MCGOVERN: Right. Exactly.
PRES. HARROZ: The way the math works out.
DR. AMY MCGOVERN: We can't stop that severe weather, but if we can improve our predictions of it and our understanding of the foundation of why it is happening, then we can give people more warning, for example, on a lot of these events and that is something I've been working on since I got here. Our AI Institute is focused particularly on a variety of high impact weather events, it’s not just the severe weather that we get here in Oklahoma. We are looking at winter weather, icing events, big snowstorms, you know, tropical cyclones, and then also trying to look further out to sub seasonal, and then we also work on coastal ethnography application, saving the sea turtles, for example. I’d be glad to talk about those too. Just looking at a variety of high‑impact weather events and with those same goals, can we create AI that is trustworthy, which we want to define what trustworthy means, but can we create AI that is trustworthy that people are actually going to use to help improve their decision‑making and actually really impact to save lives, save property.
PRES. HARROZ: Yeah. Yeah, with the number one meteorology school in the country, we have even ‑‑ those of us that are not experts in this area, understand that we have the national severe storms lab and that we see high‑impact weather on a regular basis. Certainly, in your 19 years here in this position, we have seen the increase and frequency of these events, not just around Oklahoma, but nationally and globally. So the importance of this work seems to be increasing not decreasing.
DR. AMY MCGOVERN: I think as the climate is changing, we are getting a shift in those severe weather distributions, but we are getting more extreme weather that is coming in and definitely causing a greater need for what we are trying to do.
PRES. HARROZ: Yes, and so much economically. I mean very business has an impact because of the weather. Obviously, human lives are those that are also in the balance. You started off, I assume, with meteorology has always relied upon computer scientists and you are at that intersection to bring in large data sets to try and predict the weather. And now it is taking this next step, and I think this is the machine learning, the artificial intelligence side, I take it, where you get these large data sets, but then you actually have that AI take it a step further and actually do its own “thinking” to try and predict with greater accuracy what will come out of those data sets. Is that a fair statement.
DR. AMY MCGOVERN: Yeah. It's a fair statement, there are sort of multiple ways that AI is being applied to the meteorology field and one of the things that has also changed in the last 12 to 18 months, there has been a real revolution in that. I will give you both answers so you can see there are lots of ways we've been using it. When I started here, a lot of what we did was take the numerical weather models that are already the models that are based on physics and do post processing. Try to use AI to improve the postprocessing because there is always biases in the models and try to improve our predictions. That involved taking a lot of the data you’re talking about. In the last 12 to 18 months, getting us actually back to ChatGPT, those are foundational models and generative models. They’ve started applying those to weather in general and trying to do global scale weather modeling and that is something I think if you'd asked people five years ago they would have said no, that is still 10 or 15 years in the future and all of a sudden there are companies out there and it’s primarily being done to private industry that are making global scale weather models straight up from the data. They are skipping the numeric weather prediction models.
PRES. HARROZ: Really?
DR. AMY MCGOVERN: Yes.
PRES. HARROZ: What accounted for this huge jump in the development of generative AI for weather modeling?
DR. AMY MCGOVERN: The availability of the CPU and the data. GPU is really what it is. It’s those graphical processing units. You can get them, and you can get so many of them now and then the data is becoming more available worldwide.
PRES. HARROZ: What have the advances been and at what rate are we ‑‑ what is the rate of acceleration on our modeling for severe weather? I know that for a long time you could not make a reliable prediction that was beyond two or three days. Can you give us an order of magnitude of the impact of the accuracy through these methods?
DR. AMY MCGOVERN: That is an excellent question. The postprocessing and the methods, which I didn’t mention in your other question which are hybrid methods where they take the numeric weather prediction models, but they are taking pieces of them and replacing them with AI because they can make them faster and more accurate. Those are definitely improving the accuracy out much further. The global scale AI weather models aren't quite ready -- there seems like there is a new paper every week. Probably by the time the podcast gets processed and released I will feel like I shouldn’t have said that, but right now they're not quite ready to say that they are revolutionizing it, but they are on the cusp of doing that. I think it depends on what kind of weather you are really trying to look at whether or not we can predict it enough days in advance. I think that things like looking at large‑scale, like we are likely to have tornadoes and hail and big wind events. That kind of thing, we can do better with AI now out to about eight days. Whereas as you said, a couple years ago, we couldn’t do more than two days out. The precision of like it’s going to hit my house in the next hour is still really, really difficult. There's so much chaos in the system that that is actually still really hard. That is something two students who just graduated last year were working on trying to improve the prediction specifically for hail and convective initiation, just thunderstorms in general out to about an hour. To be able to do that without -- before they have started and to be able to say there is going to be hail at your street in an hour and that's still a really challenging problem. AI is going to help us.
PRES. HARROZ: That is stunning, is it not that the development forever, you go back to the Farmer's almanac trying to predict what the weather might be like. We know the implications, and we were talking your son is a senior right now in hours in aviation. Just in the industry of aviation, being able to know where severe weather events are likely to occur, can save –
DR. AMY MCGOVERN: Billions of dollars.
PRES. HARROZ: Billions of dollars.
DR. AMY MCGOVERN: Those hail ones we were talking about, imagine if you could say very, very accurately you're likely to have hail in the Oklahoma City airport tomorrow. Major three‑inch hail, move all of the planes out of the way.
PRES. HARROZ: They are all gone. Right. You move them to where it is likely to be safe.
DR. AMY MCGOVERN: Right and you cannot do that if you only have 15 minutes notice, which is the current average for a storm that is formed.
PRES. HARROZ: The scheduling of flights where they go and the entanglements. We have seen debacles over the last year and a half in the aviation industry where airlines are locked up for four, five, six, seven, eight days because they have not predicted severe weather events. Then once that starts, it triggers all sorts of negative events.
DR. AMY MCGOVERN: I think some of that is modernization of their system too, but they are working on that fortunately. I have to ask, I get asked about how some of the stuff will affect people in the cockpit too and I think eventually it will, but right now the cockpits aren’t running AI, you know, live to tell you what you are going to see 10 or 15 minutes from now. But you can imagine now all of that it’s a broader scale of what you are talking about. It’s definitely – it will affect all of the industries. It will affect farmers too. If I can tell you we are about to have hail or tornadoes or drought, we can – a major heat wave, then we can do something to prepare, especially on the drought scale. If we can do something in the sub seasonal scale – so you were asking about days in advance.
There is a scale, a weather scale, of what you're trying to predict, sort of there’s the now scale and do I care about what will happen in the next hour and there’s also the next seven days scale. But then there’s looking at 14 days, 21 days. That gets much, much harder. As you get farther and farther out in the future, it gets more toward the climatology, which gets you to the Farmer’s Almanac, which you are talking about. It is the averages. That is the long‑term averages, and that scale is something that still hasn't really been ‑‑ we are starting to tackle it, but it's really, really hard, but it's really impactful. If I can tell you about these major cold freezes and these major heat events that are coming through at a scale of about 14 days notice, then governments can take action. Farmers can take action to protect their crops, you know. There is a lot we can do.
PRES. HARROZ: Absolutely. And you are right, seasonal prediction is incredibly exciting. I mean, you could know what crops to plant or not plant, if you can model that out.
DR. AMY MCGOVERN: Right and subseasonal is kind of in the in between the weather and the seasonal prediction scale and the subseasonal is even harder. It’s just there is so much chaos in there, it’s just hard to predict.
PRES. HARROZ: Yeah. Do you see this rate of development and change? You have been surprised over the last 5 to 10 years at the pace of change. As you sit here today, do you think that level of iterative change is likely to continue?
DR. AMY MCGOVERN: I think we are sitting on the cusp of a huge change. And I think it’s largely the private industry, investment that hopefully is going to keep working with government to do a revolution in our change. I use that word sometimes when I try to – you know when you are writing grant proposals you talk about revolutionizing things, because you really are, but I really think that right now we are sitting on that because these generative AI models are going to make a tremendous change. It is akin to when they introduced the numerical weather prediction models. And actually, just today we were reading in my research group a paper, from the 1970s, where they were complaining about the introduction of the numerical weather prediction models and how it was changing the field of meteorology and I think we are just sitting ‑‑ if you replace all of the statements about numerical weather predictions with the words AI in there, it is the same complaints you are hearing now. Why do we need to learn the physics when the AI -- what are we teaching the fundamentals for if the AI is going to take over? It's not.
PRES. HARROZ: I have had more than a few students ask those sorts of base-level questions.
DR. AMY MCGOVERN: Right. If the AI is going to do it for me, why do I need to learn this?
PRES. HARROZ: Absolutely. It gets back to the calculator.
DR. AMY MCGOVERN: Right. Those exact same questions came out with the calculator too. Why do I need to learn math if the calculator does it for me?
PRES. HARROZ: Division, when the calculator is going to do it for me. We think about your institute and the impact that you are having. If you continue on your path of success and with the development and the acceleration of the application of machine learning of generative AI, what would success look like for you in terms of the impact of your work on people and property. What would success, in your research, look like over the next five to 10 years?
DR. AMY MCGOVERN: That is an excellent question. NSF would define it by papers, but I would add other definitions, which I think NSF would probably agree with is that I would like to see as actually having our products in use, in daily use by the forecasters, by the emergency managers, by the general public so that they are saving life and saving property. And that gets us back to that trustworthy part. We are doing – that trustworthy is in our name because it was part of the theme that we applied to and that our foundational core is that we are understanding what it means to make trustworthy AI for really high impact decisions. That matters across a wide variety of -- we're talking about weather, but medicine too. You want your AI to be trustworthy for that, right. So success to me is that we have understood what it means to be trustworthy for these high impact decisions and that we’ve created something that really is saving lives and property and it doesn’t have to just be weather, it could be that we fundamentally understand something that works across a variety of high‑impact decisions.
PRES. HARROZ: Yeah. So when you think – when we talk about trustworthiness, earlier, you spoke about biases that can be introduced. Can you kind of walk us through that, what kind of biases could be injected into this space that can be deleterious? And how do you ensure they are trustworthy, and those biases are removed?
DR. AMY MCGOVERN: That is an excellent question. When you hear about the biases in the data in the news, mostly what you are seeing is that it is largely racial and gender bias, that is largely what you see. You train it all on white faces and it doesn’t understand black faces or you train it on Asian faces and it doesn’t understand white faces and things like that. People think about that in the beginning and are looking at the weather saying this does not apply, but there are biases in the data and those biases -- we just had a paper that came out last week that talks about categorizing the biases and we had four main categories and a bunch of underlying categories, and the four main categories structural and systemic biases of which some of the racial data falls into, but doesn’t fit for weather necessarily because we are not looking at faces, but it can affect the data that you record, the data that’s available.
For example, there is a lot of data that’s only available in the more hiring, the more higher socioeconomic bracket because that is who is affording the instruments that they are providing crowd sourced data or that is where they are trying to recover the data. That can lead to a skew in the data. Another category that is interesting is data, skews in the data, but something unique to our science and weather is the skew from the laws of physics. So you can have data -- like I will use radar because that is a nice easy example. The radar, if the earth is round and the radar points out in a straight line, I know that does not show up on the podcast audio, but at least it will show up on the video. When tried to look at the tornadoes, you cannot necessarily see the tornadoes if you get too far away from the radar because physics does not work that way because the earth is falling away. If you are trying to build a model, for example, that was depending on that really low-level data, you would not be covering the whole United States.
Then if you didn’t realize that you didn’t have that physics‑based bias to your data, you might release a model that you thought worked great that is in fact not covering a bunch of the United States and that could be a real problem. We actually have a picture in a paper that we wrote a couple of years ago that shows that in the southeast United States, and this is totally not on purpose, I want to emphasize that. But when NOAH was putting their multimillion dollar radars out, they were covering the bigger cities, but they missed a lot of the rural areas and the rural areas that they missed were primarily black. And so if you were to put a model out that only used that closed coverage, you would miss a lot of the primarily black areas and that would get you back to that racial bias.
PRES. HARROZ: Absolutely.
DR. AMY MCGOVERN: And that ethic, it is really important that you counteract for that bias that you do the best you can to get all of the data that you have. Another example would be like the global south. We just don't have a lot of instruments in the global south. The United States is really, really rich in instruments.
PRES. HARROZ: Yeah. Yeah, it’s fascinating, especially that physics bias that exists that you don't account for the Earth's curvature, and you have linear projection, and you just don't pick it up.
DR. AMY MCGOVERN: Right, and I only gave you one example of a physics bias, but there are other ones out there. There are biases that can happen in the weather models themselves because they don't understand the laws of physics, so they can learn something that matches the data, but might be a complete violation of the laws of physics and you have to account for those too.
PRES. HARROZ: I just hope all of my friends heard me say linear projection and that you agree with that because I think I felt a lot smarter then. I had two classes in physics, and I think that is all I learned. So I appreciate that. Tell me, there was an announcement today that came across my desk that we were able to release that involves the National Institute for Science and Technology. Is that right?
DR. AMY MCGOVERN: Standards.
PRES. HARROZ: Standards and technology. Tell us about this.
DR. AMY MCGOVERN: So NIST just released -- a couple months ago they put out a call for people who were interested in working on something that came from a call from President Biden. He is trying to focus on AI in general and AI safety and NIST is going to be creating this thing called AI Safety Institute – AI Safety Consortium. There is another letter in there, but anyway, it is about fundamentally understanding what it means to make trustworthy AI and how we can sort of guarantee that AI is going to be trustworthy and that is something that we are trying to do inside AI2ES and something that a wide variety of people, when we talked about the medical application, we really want to make sure that AI is, it is really snowballing down the hill. We want to make sure that what we're doing is trustworthy and I think the cool thing about NIST getting into this is that NIST is the one who makes the standards for a lot of what we have out there. I think people don't realize that, but they put the standards out. People have to follow the standards and then you know they are safe.
PRES. HARROZ: Yeah, so to me this is really exciting. We just sort of take a step back for the listener if they are not familiar with NIST, I don't know how long it has been around, but their job is to set standards. So if there is one type of plug that you use, I mean, if they are different private manufacturer of a certain kind of electrical outlet that they can create a standardized set that everyone can conform to so that you don't have to buy four different kinds of plugs to get electricity out of the wall.
DR. AMY MCGOVERN: But also, there are standards in what they have to be able to handle, so you know none of those plugs are going to catch fire.
PRES. HARROZ: That is it. So NIST does this and it is fascinating because now this National Institute for Standards and Technology is making sure that the artificial intelligence has these standards to ensure trustworthiness, right, in what they do. It fits perfectly in your space, and you are a part of this, are you not.
DR. AMY MCGOVERN: Yes, so there are actually two people – two groups at OU that have joined in with it and we are really excited to be part of it. It just, as you said, it just kicked off so I don't have any results to produce yet, the press release just came out.
PRES. HARROZ: Yeah, but it’s incredibly exciting that our researchers, you and others, and DISC is part of this?
DR. AMY MCGOVERN: Yes, DISC is the other entity.
PRES. HARROZ: And what is DISC?
DR. AMY MCGOVERN: Data Institute for Societal Good.
PRES. HARROZ: Yeah, DISC is Data Institute for Societal Change, which is positive and positive change and Dr. Ebert --.
DR. AMY MCGOVERN: Dr. Ebert is going to be like why did you not have your name off of the top of your head?
PRES. HARROZ: No, that wasn’t a fair question. I didn’t get NIST right, so I think we're even. To me, this is really exciting. I mean this idea of engaging in standards to ensure that what comes out of it is trustworthy is essential. And to me, it links back to this first question we had together, this question about is AI, right, is generative AI going to end humanity or allow humanity to flourish and for us to actually be able to answer that in a way that is refined and to make sure it leans more towards the good than the bad. One of those things is you have to make sure that there are standards and that they are fair and reasonable, right.
DR. AMY MCGOVERN: Right. If we don't have those standards, I do worry about where we are going to go because I think it is so open that somebody will create something that is not good.
PRES. HARROZ: Yeah, and it applies. We have spoken, as you said -- you have brought it back a couple of times. I think it is important for the listeners. We have been talking about it around severe weather events, but this applies to everything from medicine to pick your industry, right.
DR. AMY MCGOVERN: I don't want my plane to fly itself unless it’s meeting certain standards. I don't want my doctor to diagnose things unless it’s meeting certain standards. Anything that is high impact, you want it to have standards. If it is making a picture of me, I certainly – we can talk about the copyrights of all of that. I agree with that artists need their copyrights, but it is not going to kill me if it makes a bad picture of me. Whereas if it is a robot during surgery and it does the wrong thing, it could kill me, and I want it to meet a standard.
PRES. HARROZ: Yeah, absolutely. Although, I could use some image enhancement personally. Like I would want my generative AI to give me a flowing head of hair that is somewhat godlike. That is my goal.
DR. AMY MCGOVERN: After the show, I can give you a URL that we have been playing with that actually you can put in a picture, and you can make whatever you want. We had some fun with it recently.
PRES. HARROZ: We will do that on the video.
DR. AMY MCGOVERN: You can give yourself a glowing head of hair.
PRES. HARROZ: It will be like Fabio sort in 1980 I think is what I am looking for.
DR. AMY MCGOVERN: I put myself in an astronaut costume, and then I had me skiing down a hill. For some reason, it had me with flowing hair down the hill.
PRES. HARROZ: Can you give me an example of other industries that have had rapid change on society like generative AI where we have had to put in place standards that have helped ensure that it is done ethically and in a way that is under control, end quotes. Are there other examples like that?
DR. AMY MCGOVERN: That is an excellent question. I would say medicine is probably one of the better ones because there is so much happening in medicine. We take our IRB training, and I don't know if you have to do that anymore, but you’ve done it I'm sure you have done it. They teach you all of the bad things that people have done to people, and you learn all of that and you think what in the world were they thinking and now there are ethical rules in place so that we cannot just experiment and say well we are going to withhold the drug from you.
PRES. HARROZ: Yeah, and to me, I look at those other industries and I think what will allow us to avoid the panic and to avoid the worst outcomes with this remarkable technology. We know in the future we are looking at quantum technologies that will present maybe even more dizzying iterative steps, right. We are able to handle responsibly through a set of standards, and you are doing that. That is what excites me a lot about the work you are doing is that you and Dave Ebert, through DISC and through the work that your institute are helping not just our students learn, but helping the nation and the world as we adapt to generative artificial intelligence and machine learning.
DR. AMY MCGOVERN: I'm actually teaching a class on AI ethics this semester as well, so responsible and ethical use of AI focused on earth science, but just in general also just responsible and ethical use of AI. I feel like it should be something we teach. The way AI is becoming so common, I think this needs to be something almost every student needs to understand, right, because they just need to understand the implications of the AI, and what they are doing when they are using it.
PRES. HARROZ: Yes. I think that our students that have hopped on ChatGPT of late could probably make sure they have a good course in the ethics of using AI. Dr. McGovern, thank you so much for taking the time. I can see offset they are looking at their watches, which is to say Joe, you are enjoying yourself, but we are close to the end of 25 minutes. Thank you for what you are doing at the University of Oklahoma. You are one of those colleagues that we have here that I'm just beyond proud to be associated with, and it is clear from this podcast the energy that you bring, and it is clear from all of the, you know, all of the national entities that want you to be part of it that you are making a massive impact. We talk about changing lives, and you do it every day. Thank you so much.
To the listeners, thank you for joining today's show and make sure you are subscribed to catch all future issues. This is a good example of how exciting they are. Thank you for being here, thank you for joining Conversations with the President.
