Latest / Atmosphere with Emily Gracey / AI & Weather: Where Do We Draw the Line?

Transcript

AI & Weather: Where Do We Draw the Line? · Atmosphere with Emily Gracey

• Emily Gracey: Meteorologists have always used tools to look into the future. Satellites, radar, computer models. And now we have one more. Artificial intelligence. I don't think that anything I put in there is going to replace the meteorologist. I think it will empower them. AI is already analyzing weather data, writing forecasts, creating content. So avoiding it probably isn't the answer. But knowing where to draw the line. might be. The AI doesn't understand context and community. It also can't communicate empathy. Today on Atmosphere, we're diving into AI and weather, where it's already being used, what it can do well, and what we probably shouldn't hand over to it. Because whether you're a meteorologist using AI or just scrolling through weather information on your phone, this affects you. Where's the red line here? What are we not willing to cross? What are we willing to not let it do? I'm Meteorologist Emily Gracie, and this is Atmosphere, a podcast where science becomes conversation. Each week I sit down with the experts behind the forecasts, the research, and the headlines to explore not just the science, but the human stories behind it. Through natural curiosity, we're making science more understandable, more relatable, and hopefully bringing a little more comfort in how we experience the world around us. Welcome to Atmosphere. I'm Meteorologist Emily Gracie. Before we get started, a quick thank you to our sponsor, Weather 100. Their support helps me keep having conversations like this one, digging a little deeper into the science, technology, and the people changing the way we understand our world. I'll share a little bit more about Weather 100 later in the show. And a quick bit of housekeeping, if you like Atmosphere, if you like this episode, make sure you're following or subscribed. On YouTube, on Apple Podcasts, and on Spotify. It costs nothing, but it really does help more people find the show, which in turn will keep the content flowing week to week. Okay, Tim Heller. Tim and I have known each other for years, and we've had versions of this conversation many times, usually over drinks and without microphones. Tim spent 35 years as an on-air meteorologist, including decades in a major market. He's an AMS certified broadcast meteorologist. His work has earned him four Emmys, and he was one of the early broadcast meteorologists experimenting with augmented reality in weathercasts. So innovation is not a foreign concept to Tim. In 2019, he stepped away from daily television and started Heller Weather, which is where he works as a weather communications coach, helping meteorologists figure out how to work more efficiently and communicate more effectively. And lately, a big part of that work has been artificial intelligence. Tim has developed his own AI weather tool specifically for meteorologists. It can help analyze weather data, identify important parts of the forecast, and create first drafts of everything from weather stories to social media posts and push alerts. Which makes Tim an interesting person to have this conversation with because he's not approaching AI as someone standing on the sidelines wondering. what it might do someday. He's building with it, he's using it, and he's also thinking pretty seriously about where the boundaries should be. We get into what AI is already doing in meteorology and whether it could eventually cost meteorologists their jobs. AI generated weather graphics and avatars, we'll talk about the environmental cost of all of this technology. And what happens when we start outsourcing too much of our own expertise. Because for all the things AI can do, There are still some things Tim says it can't replicate. And maybe those are the things that have become even more important as technology continues to get better. Here's my conversation with Tim Heller. Well, Tim Heller, I'm so happy to have you on Atmosphere today. We've had so many kind of like bar conversations about this over the years. So I'm glad I finally get to record some of the stuff you're saying because a lot of stuff you say is just so interesting and fascinating. You're doing very innovative work in this space. So I can't wait to pick your brain today. Welcome to Atmosphere. Thank you very much for having me. Yes, of course, former broadcast meteorologist as well. That's you know, you're going way back here. But you've had a little transition in your career in in recent years, right? Yeah, 2019. I decided to step away from the on-air side of the business, step out of the studio and started my own coaching company. And mainly because I felt like there was nobody out there helping the meteorologist communicate more effectively and work more efficiently. And as you know, from working in the business, a lot of times state TV stations will hire consultants who come in and they tell us how to do our job and they have no idea. How we do our job. And some of the advice they give us is downright silly sometimes. You know, I can remember one time being told you need to say the word neighborhood more often because it focus groups really like that word. I'm like, really? I just need to say the word neighborhood. Another one was you have to use the word forecast at least three times in your presentation because that that engages the viewer. And I'm like, okay, if I was doing my job, I would naturally say that word, but I don't need to count the number of times I do it. So I decided I I need to the business needs a coach. And so I decided I'm gonna do it. And and so that's what I've been doing since 2019. I've gotten a handful of T V stations I'm coaching around the country and from the northwest corner to the southeast corner and everywhere in between. And it's been great. I love what I'm doing. Yeah, you're doing great work too. It's like concrete information that's actually useful. Like you said, consultants sometimes come in and they just say things based off of focus groups, but You you really connect with meteorologists and help them tell the story better because you're such a good communicator. And I the what I love about you is when you present at conferences, you always bring up stuff that like people haven't thought of. One of the first things I saw you present was about radar and how when we're looking at radar, we're looking at it as most of us looking at it as people who are not colorblind. But what about the colorblind community? And you brought up this whole area that nobody even thought about before and how small changes to how we present the weather can be more inclusive to so many different groups. Yeah, you know, and just talking about that, I mean, meteorologists love the rainbow color table. And there's no one rainbow. And and I in one of the presentations I gave at a conference, I showed all the different legends that are available for all the different types of data we have in one of our graphic systems. And There's no one rainbow. And the color yellow, which is the most luminous outside of white, can be anywhere in that range. And for people who are colorblind, we many, most of them have issues with the shades of red and the shades of green. And when you think about it, we're showing the good weather, oftentimes the shades of green and the bad weather and shades of red. But if they look the same, some people could confuse the bad weather for good weather. And so removing Shades of green out of that and switching it to blue, to yellow, to orange, to red clears it up for all viewers. But I know that's a big ask because all of our apps use the the rainbow color table for radar. I don't think we'll ever get away from it, but I do think we need to be aware of that as meteorologists in the at least when we can avoid that rainbow color table. So many good little tidbits like that. But today you're here to talk about. AI and weather. And this is something that in recent years you've really embraced in so many ways. So take me back because this it feels like this is accelerated pretty quickly, at least in my world. But when did you start using AI in what you do? So AI was originally a release to the public in the fall of 2022. That's when OpenAI released Chat GPT. And so people started playing with it right away. And so I started to dabble in it. And I will admit my initial response to AI was the same as it was when the internet was first introduced. And that was, I see a lot of slop. It's kind of fun, but might be something useful here. And nowadays, of course, we can't live without the internet. We're on it right now. and so I started experimenting with it in the summer of twenty twenty things. And by the fall, I had kind of realized like within six months, like this is the tool for broadcast meteorologists. It's available at the right time. For us, and it's the right tool because my experience of working on the air, and as you know, one who's worked on the air, we're being tasked to do more and more and more. And TV stations are not hiring and in many cases reducing their staff. So it's not just the on-air broadcast. In fact, I preferred the term multimedia meteorologist because we're on air, but we're also on multiple social media platforms. We're doing streaming updates, some Meteorologists are now doing weather shows once a week or even daily weather shows. And they're doing newsletters. Well, so the quantity has increased. And as a result, because we have limited amounts of time, I think the quality has been reduced. We're not thinking through things like the color of the radar and stuff like that. And and and some of the some of the things that we're showing on the air. So I kind of felt like AI could be a solution here. And out of frustration, really, I wrote an article that's on my website still today called the Weather Workstation of the Future. Frustration because I felt like the big companies that create our graphics or that we use for our graphics and our data for the on-air broadcast, I felt like they had the resources that they could create an AI component to help us out, to sort the data, to help produce content. And so I created this concept called the weather workstation of the future, focusing on the forecasting the weather, using AI to forecast the weather, to create weather content and then distribute it. And out of frustration where nothing was happening in that field, I decided in 2025 that I'm gonna make my own application. nobody else is doing it. And so I started vibe coding an application. It was really simple. First, it just took the latest AFD from the Weather Service. And for those who don't know what AFD is, can you explain? Several times a day, the National Weather Service issues an area forecast discussion. And it's basically a very technical discussion. Of how the weather is and is changing over the next seven to 10 days. And a lot of it's very detailed, very technical. They are very long, sometimes a thousand, two thousand, or three thousand words. Nobody has time to read all that if you're a broadcast meteorologist and you've got to be on the air in 20 minutes and you got a webcast and you got a tease that you got to do. I mean, there's so much that you're you're juggling. And every broadcast meteorologist at some point tries to read that because it's another set of eyes looking at the data. So I thought, well, let's take that very technical discussion and simplify it in less than 250, 300 words. And then let's take and make a conversational version of that that you could use then for your weather story on your website, kind of reducing the workload. I call this a first draft. It creates the first draft of the story. So I created a little application and in the process of creating that, also had it in the back end run some test to make sure that it was not leaving out important information. that it wasn't hallucinating, that it wasn't glossing over stuff that should be more important. In other words, it was weighted toward more extreme weather and unique weather conditions. And after testing that for a couple of months, and I literally have the files, if anybody ever wants to look at them, over a hundred files where the AI has where I ran the AI through these tests and tried to verify and and verify the content, I released it to my clients. And they loved it and immediately started requesting things. Can you do headlines? Can you do social media posts? Can I all right, I think I'm on to something here. So long story short, I in the early part of 2026 took a step back, rebuilt the whole application as a dashboard, add a ton of new features to it, hired an independent contractor who helps me with some of the coding to make sure that it is stable. I contracted a data provider so that the data we are pulling into the system is has been verified, is legit. It does not hallucinate. It does not go out and access outside information. We've got a whole database that we've built and storage locker that we're using on the back end to ensure that the content is reliable. And now I've got dozens of meteorologists that are using it every day to create, not only to help them forecast, but also create content for social media, for their website, for newsletters, whatever you need, this application does it. I'm curious if any of your meteorologists that you work with, if this has in in response to this, created more work for them in the way that their station is now saying, like, wow, they're just like churning this stuff out. So you know how like stations want more and more, more, more, more. So is there a paradox here where we start to use this stuff so much that it just creates even more work for us? I th well, what's happening, the feedback I'm getting from meteorologists is this in total is saving them about three to five hours a week for their personal shift. And so what it comes down to about 45 minutes to an hour per day that this is saving them. Well, there what is the term? Something about, you know, nature doesn't like a vacuum. we're filling that time. It's not like we are sitting there and like, okay, well, I got the forecast. I'm putting my feet up. I mean, there's always something that needs to be done, at least the my experience so far is that this is helping them work faster. So it's it's it's increasing the efficiency of what they can do, which then leaves them time to do more creative things, more creative storytelling, more in-depth analysis. Cause one of the things my application does when they first sign on is that they get the latest forecast from the weather service, which is kind of like a good starting point. But not only that, but it identifies days that are disruptive or possibly severe based on criteria that they pick. I'll give you an example. I've got a power company who's using my application. They're very concerned about wind. So they've there's some thresholds that are built in that they can control and they have wind criteria that they're monitoring. And so anytime that the wind forecast is above a certain threshold, it identifies that day as a potentially disruptive day. So what it does as a meteorologist, it says, okay, I need to focus on Thursday because that's going to be a very windy day. And so I don't need to be spending all my time on days when it's partly cloudy and you know, I have 90 degrees. I can come over here and focus on the days that are disruptive. And the same thing for broadcast meteorologists that are worried about the next big weather event. We've got a whole bunch of thresholds that are built into the application that they can control and they say basically monitor for these conditions. So when those trigger, then they can isolate their forecasting time to a particular few days. So they forecast faster. They can they get first draft content that already has written to include that. So they can come in and localize and personalize it. And it speeds up their workflow so that way they can spend more time doing things like in depth analysis and creative storytelling. So so far, nobody's putting their feet up on the desk. They're they're filling the void. Gotcha. Yeah. Maybe getting a little bit more comfortable with their forecast. Do remember always kind of being panicked, like I'm going and I still have nightmares about this. I haven't worked in TV in years. And I still have nightmares that I'm going on air and I don't know the forecast. yeah. I do too. I've been off air for almost 10 years and I've still panicked every once in a while. That and I'm late to work and I forgot my makeup. Those are the other two. I have this. So the last station I worked at was the ABC owned station in Houston. And there literally is this hallway. That leads from the newsroom into the studio that gets narrower as you go down it. And so in my dream, I am running down this hallway and it keeps getting narrower and narrower. I can never get to the end of the hallway. I can never get to the studio. Maybe that hallway was on purpose to play mind games with the employees. okay. Well, let's you've you're a veteran in this business. So I kind of want your opinion on some stuff too. For people who maybe are thinking about using AI and listening, or even like viewers and listeners who are taking in weather information in like a digital platform and they're looking at what potentially could be AI generated, what could be a person, what could be a combination of both. So let's start with like the meteorologists. They're putting stuff out there. Responsible uses versus irresponsible uses. What have you seen that you're Super impressed by, and what have you seen that you're like that's cringy? I can start the cringy part first. so there is a network right now that actually has an AI meteorologist doing part of their presentation on TV and on digital streaming. And they actually font it, and I can't remember it's something avatar or AI Avatar. So they're being transparent to let you know this. She has like 14 fingers. No. But Her hand gestures tend to repeat themselves over and over and over. And the delivery is very robotic. And I know that this is a first generation and this will get much better with time. And we've seen that in our own social media feeds of some stuff that you're like, okay, is this real or not? which is why apparently they are now, you know, I guess tagging AI content so that you're aware that this person may not be real. By the way, I'm here, I'm real. I'm assuming you are too. So I think some of the cringy stuff are these AI avatars. They when when you're actually taking real information and feeding it into the avatar and having it read it back, it comes across robotic. there's no emotion in there. and I know that we'll get better. I think that we've got to be responsible, hopefully, as people who create that content, of at least identifying it so that the people understand that this is a robot. This is not real. I I still think that there is a use case there that could be valuable. And research has shown that people respond to other people, like they are more apt to respond to a face than they are a map or a graphic. If I'm an on-air broadcast meteorologist and I'm on TV tracking severe weather and the weather service issues a new tornado warning, but I want to get it out of my digital platform, my social media, could I use an AI avatar of me? Delivering that warning, fonted and identified correctly, but me saying, or at least a version of me saying there's a tornado warning in effect for these communities, take shelter, and then tune into my streaming platform where we're doing live updates. Could that be effective use? I don't know. Nobody's doing that yet, but something tells you, isn't Amy Freeze doing that? I think I've seen she is. She started to. Yeah. She's doing it more of the weather forecasting, right? The I think like a daily forecast. I don't know that is she if she's doing that on severe weather or not. I think she was like testing it, but I'm not sure how far she got with that. So I'm just I'm like wondering if there is a use case for TV stations and who have you know, they've got streaming, but they've got broadcast, they've got apps, you know, they got all of these different platforms. Is this a way for us to communicate quick little snippets of information? To get people to take shelter, to get people to respond, that might be more effective than just a radar map, or just a graphic that says tornado warning. I mean, things that you're going to scroll past, you may not scroll past somebody's actually talking to you. I don't know. That to me, I think that that's worth exploring. I do think that there needs to be kind of some testing and certainly transparency, to see how it could be used effectively. but right now, that's from the avatars seem a little cringy only because a lot of it sounds robotic, the gestures look, you know, repeat themselves over and over. but it'll get better. Do you worry? I mean, I've seen we've seen cases where from the meteorologist perspective, say they're working for a station and and that content is out there. And then it's out there also for the use of criminals. And we've seen that. happen to colleagues in our field where that has been used for the wrong reasons. So you know, that's that's like the concern I think of a lot of meteorologists, like, you know, I don't want my likeness being stolen. Right. Well your likeness is already out there on social media. You know, you're already you're already doing videos on social media, you're doing videos on YouTube. So that that is out there. I could see, you know it's a risk all of us have. Doesn't matter if you're an honor broadcast meteorologist or general public and you're posting pictures on your Facebook page. That is available to people that are out there. that's part of the reason why I, when I worked on the air, never identified my children and any s social media post. It's why you see some people on social media today putting little emojis above the faces of their kids because they don't want them used. you know, I I think that's we have to worry about. All of us have to worry about that, right? So it's not just a broadcast meteorologist. That's something I can't control. I'm trying to find an effective, appropriate use of AI for my meteorologist. That's one of the things that I refer to, the content that they is created by the application as a first draft. I use that all the time. In fact, there's a what they call a loading text that comes up while the AI is functioning and it literally says creating your first draft to remind them that this is just. A first draft that you need to edit it. You need to personalize and localize it. And in the training and the onboarding, when I'm teaching people how to use the application, I remind them that the AI doesn't understand context and community. It doesn't understand how your community responded to the last big event. It doesn't understand how that event affected people in your area. For example, I live in Houston, where after Hurricane Harvey, we had to be very sensitive to how we approached rainfall and flooding. And to this day, the station I work for still does not talk about flooding in very general terms. They talk about flooding in terms of street flooding, bayou flooding, river flooding, and area flooding. I mean, we break it down because everybody had different experiences in that. And we know that, but the AI wouldn't necessarily know that. I know that because I live here, that I experienced it, and I know my people. So the first draft that the AI creates for you is just that. It doesn't understand context or community. It doesn't know that. Friday night's a big football game. At Saturday, there's a farmer's market downtown. All of those types of things you can then add to the general weather story. But let the AI do the general stuff. It doesn't take a creative genius to write tomorrow's going to be mostly sunny and a high of 90. Let the AI do that first draft. And then you come in and add some your own special flavor, your own special swagger, and put out the social media post, put out the the weather story. But to me, that's how as people who use AI, That's how we can differentiate our content. We let the AI do the first draft and then we come in and finish it. It's really interesting. and I guess that's from the perspective of somebody who's doing like daily weathercasts, because I don't do daily weathercasts, but I I kind of flip it. So I usually start with my own words and like a mumble jumble writing of here's what I'm thinking, almost like journaling. And then I throw out an AI and I'm like, help me make sense of this, help me put it. Chronologically correctly, let's smooth things over and make the grammar a little better. Well, and I think a lot of people do. And so there's a part of the application that we're building right now called Content Creator. And one of the things that you can do, and first of all, it will I surface five to seven different stories that you could tell on any day on every day. And it will look at your forecast, it'll look at the thresholds that you've set up to monitor. It'll look at severe weather, tropical weather, fire weather, drought. It looks at all of the data and says, here are five stories that you could tell, either on air or online or whatever format you want. And you pick the topics, you l and it basically looks like like an outline. And it already surfaces for you some facts that are based in the data. You can then rearrange the facts, you can add your own, and then you tell the AI, you know, I need to write a web story. Of 200 words, generate. And in 12 seconds, it creates the story that you have decided that this is what I want to talk about. So you have a little more control over it. And you can even do some very open prompting. you can say, you know, we got a backdoor cold front coming in, explain this. You know what it is, but you don't want to have to sit there and write the whole thing about a backdoor cold front, but you hit generate and the AI knows. What a backdoor cold front is and creates that content for you. Do you worry that any station management is going to look at this as a replacement for meteorologists and think, well, why don't we just put the reporter on weather now? Because we've got this doing the job that the meteorologists that we needed the weather expert for before. Yeah. I as a broadcast meteorologist, I'm concerned about that. I don't think that anything I put in there is going to replace the meteorologist. I think it will empower them. and that's you know, that's something I'm thinking about, obviously, because that's my former job as an on-air meteorologist. And that's really how I approach this. Like, how can I help, if I was, if I was still on the air, what would I need to help me do my job better? And then how can I use AI to do that? It's interesting, the people that are using my application, because I get analytics to see who's using it, what how often they're using it. I don't get the specifics, you know, like what they're actually using it. are you know the sp specific details is very limited, but I can see how often people are using it in the last time they use it, for example. And I know these people because I coach them. And what age group do you think is most likely to use this application? Older or younger? Younger. I thought so too. It's actually the older group of meteorologists that are using this most often. The younger meteorologists I've got several younger ones that will not use it. And that's fine. They don't have to. It's their their station as other members of the team are, but they just won't use it. And I thought it would be the younger ones because it's new technology. But here's my theory on this. And it's totally a theory. The younger ones haven't realized yet that they can't do it all. The older ones who have been in the business a while have realized I can't do it all. I need help. And here's a tool that will help me do it. This doesn't surprise me at all, actually, because I mean first instinct was, yeah, that's surprising. But like I I feel like the younger meteorologists feel like they have something to prove too. Where it's like, I just learned all this stuff in college and I'm gonna I've gotta prove myself. And the older crowd, we're like, We're tired. Like we we don't have anything to prove. We've put in the work. I wanna go have dinner. What is normal weather? Between dramatic headlines, record this, extreme that, and hundreds of weather apps all showing you basically the same information, it can be surprisingly hard to tell what's actually unusual. That's where Weather 100 comes in. Weather 100 is a minimalist desktop weather display that compares your local weather with the last 100 years of history, showing you how unusual it really is. From today's maximum temperature to last month's rainfall to this week's highest wind speed. Weather 100 updates every day with the latest rankings, giving you context you just don't get from a normal weather app. Every Weather 100 also includes access to the free iOS and Android app, letting you explore weather rankings for locations all around the world. Starting at $49.99 with free shipping worldwide, Weather 100 has five different models and 10 colors to choose from. And listeners of Atmosphere will get 20% off by using the code EMILY20 at checkout. That's EMI20 for 20% off your entire purchase. Weather 100, your weather in context. okay. Well, going back to AI and like the visuals, I because I want to ask you about graphics, because I'm seeing and I've heard very mixed opinions on this too. Especially I understand the perspective of, you know, digital meteorologists who work for themselves and and not having the finances, I guess, to put into like big graphic systems and creating stuff using, say, Chat GPT. But you look at them and it's a very specific look and it's very obvious that this was created by AI. And so I hear other meteorologists say, like, that's AI slop, like, I don't want to see that. What is your thought on that? I hate it. I hate the the the slop that's out there. It's immediately recognizable in terms of graphics. Sometimes the pictures are tough. but you know, when you see text and it's all using the same kind of, I don't know, airbrush font or whatever it is. I mean it's it's It and it's very detailed. It's got lots of color and it's like overly produced. Random graphics. Yeah. It looks like those t shirts you buy at like six flags in the nineties. Exactly. Exactly. Like the name of the roller coaster in this weird font. yes. But it's funny because that's one of the the requests. Cause I talk to my clients are using the AI application all the time. And they're all like, I really wish it could make some graphics for me. I'm like, That is a I I don't, you know, I've had conversations with my my the developer I'm working with, like how can we create like a tool here? And so one of the things we're toying with is maybe connecting through an API call, which is an application interface call to your Canva account. So if you have a Canva account, you set up this integration with my tool and Canva so that you have already the template. The station colors, the station fonts, or even if you're a digital meteorologist operating on your own, you put your logo or whatever, and you say, okay, here's the template I want to use. And then my tool, and when you need it, my tool would go out to the app, to Canva, create, plug in the right data, and then come back and give you the graphic that you need. And that sounds complicated. It's actually pretty easy. a lot of the stuff that you're using online right now uses API calls to generate. And so that's I think where we may end up going, maybe sooner than later, because that seems to be an overwhelming request I'm getting from people is I I really want this to help me create graphics for my social media. And we and the content creator that is we're releasing now, it there's a conversion tool in there. So it takes your sixteen by nine graphics and converts them to nine by sixteen and does it really well. I've spent a lot of time prompting it so it does it correctly, but At that's to me is like the the the band-aid or at least the next step is like, okay, I I can't create the graphic for you, but I can help you convert the graphics you have. There I mean, there's a need because those are everywhere. And so obviously something needs to be done about that because they also need to be differentiat differentiated between legitimate meteorologists and yeah, like bots or just random people who are putting out information by just prompting. so there's a lot of confusion and social media with some of the weather information going out. Well, and then I it's it's funny because I like I said, I work with a power company who's using my application, which is not what I in intended it to be used for. I got a radio station that's using it. So their disc jockeys have some legitimate weather tidbits that they can toss, you know, put in, which was not what I intended. So at this point I'm kind of taking a step back and and there's a certain point when the applic when I The application has fulfilled what I think are the broadcast needs. And I need to look at the whole application, say, okay, how can I expand the use of this to a hotel chain, to, you know, an independent person who's got a blog and they want to include some weather information, but they need help creating that. and they need a graphic to go along with it. So I think as, you know, a business person, I've got to think a little bit broader than my broadcast meteorology community, because I think there's a There's a need there outside of broadcast meteorology. Sounds like a prompt for Chat GPT. Yeah. Who is my target audience? Yeah. I can't tell you how much conversation I have with AI. Like we Is it addictive? I feel like it's a little addictive. my God. My wife will come in, she'll in my office, she'll go like, okay, at what time are we going to bed tonight? I'm like, I'm in the middle of something. Any advice for people who are who take in the weather information, not the meteorologists, but the viewers, the listeners, the people absorbing this information, anything that they should know or be on the lookout for when it comes to this. I think we have to assume that AI was involved in everything we read, everything we see, everything we hear, because it is every application on my desktop right now has some kind of AI component built into it. And many of them I didn't even ask for, they just showed up. It's to the point where AI is going to be involved in almost every task that we do. And that means that the content that you consume, either on air or online or on social media, may have an AI component. Even before AI was a thing, we should have been second-guessing everything we read online because people are making stuff up to generate clicks. And I'd like to think that my broadcast. Friends are not doing that. And I don't think they are. But there are people out there that are making stuff up just to get a click. So you gotta, even before AI came around, we should have been questioning all of that. And now we really have to, because AI is generating some of the stuff that we're reading online. And we gotta question is this real or not? I think skepticism can be healthy. all right. Well, I'm gonna be a skeptic here. And ask you about the environmental cost of AI as a scientist, as a meteorologist, how do you factor that into the ethical use of AI? And how does that play a role in your use at all? Absolutely. Absolutely. As a developer, I'm really and as a developer of a weather application, I understand that the people using my application are going to be very sensitive to the environmental effects of using AI. and so one of the things that we have done, and I've worked with a developer in making sure that this that we're doing this as best we can, is that we're trying to limit the use of AI in the application. So for example, one of the things that we're doing is we're caching the data. So if one person from Houston looks at the data from Houston, that is cached and put into storage so that if somebody else looks at Houston, it just pulls it back from storage. We don't have to go and get that again. But even that process has been reduced from an AI ask to an API ask. So there's no AI involved in that. So we're trying to limit where we're using AI. And we're making it optional for the user. Like, for example, my tool will translate from English into Spanish. Initially, we thought about having the whole application translate and everything automatically translate, but that uses a lot of AI prompts. And so what we decided to do is make it optional. Like you pick which part of this. If you want it to translate, you just hit the translate button. And incidentally, it doesn't translate, it rewrites in Spanish and in what is called conversational American Spanish. And so that's that's a choice that the user makes, says, okay, I want, I want this rewritten. That does use AI, but we're very cognizant of how much we use AI. And one of the things that we've learned is that rather than doing multiple prompts. We actually are using AI more. I guess environmentally friendly way by giving it one huge prompt. So there's one big prompt. When we do, like, for example, the weather story and the social media post and the push alerts, that's one very large prompt, which uses less resources than saying, give me a weather story, give me another weather story, give me another social media post, give me another. You do pretty well. Let me how about the push alert? Give another one. So rather than doing 15 different prompts, we do one big one. And that supposedly is a little bit better in terms of the amount of AI usage. Yeah. And it's kind of like you know, the balance of like what we're communicating because like this video call itself is right an environmental cost. This is using water. So, you know, my thought is always like, but what's the story I'm telling? Will it in the long run help educate a greater audience about this? And does that create a balance? I don't know. It's a question we're still trying to figure out, I guess. so we're just Well, let me ask you, do you think because this is becoming a bigger and bigger concern, especially as communities are dealing with data data centers being built in their backyard, that we'll get to the point where we may see on screen every time we do something what the AI cost is. I mean, maybe you go to Chipotle and you're seeing how many calories you're taking in. You, you know, you order a book online and it's telling you what the carbon emissions are of when it's delivered to you. So maybe. That's possible. or at least some recognition of like this is gonna cost you like some of my it's it's funny because I'll ask the AI. I'm like, okay, if we implement this change, what is going to be the what is the cost? Because as a bit as I'm paying for the AI usage on my side. And then of course I've got a subscription fee for the people that use my application. I gotta make sure that my costs are being covered by the subscription, right? So I ask, what is this gonna cost me per call per every time we use AI? And it will be like point zero zero zero zero four. Like, okay, I don't even know what that means. Yeah. It's like a fraction of a fraction of a fraction of a cent. I think we're okay. Yeah, but then the the irony of like, okay, then we see something on screen of what the cost is, but what was the cost to create that little box to put that little thing on screen? You know, it could keep going down a rabbit hole here, I guess. Well, and I think I think there's some misconception that these data centers that are coming in that they're all AI, like. A large part of that is, you know, it's your Netflix uses AI to help figure out the next thing. And that's inconvenient. Nobody wants to hear that. Right now. Like you want to shut down the data centers, you got to shut down your computer, you know? okay. Well, is there anything that you could use AI for day to day? It doesn't have to be necessarily in weather that you've chosen, like made a decision not to. It's almost like an a question I ask every day. Like, My Gmail now, and I don't even remember asking it to do this, but my Gmail will recommend a response to an email I get. And I'll look at it, I'm like, yeah, no. And I'll write my own. And I do that quite a bit. I just don't like the way it writes, but it's uncanny how much like I have a very particular signature. I do a dash capital T, capital I, capital I don't know why. I just do that. And it's figured that out. Like this is how you sign your emails. and it's so it's it's learning. And I've discovered that like I write very conversationally. I'm not very formal in the way I write. And I discovered that the the the replies that it recommends initially my response was no, I'm never gonna use this, never gonna use this. Now I'm like, okay, that's pretty close. Good enough. It's getting to know you. Right. It's watching always. Is there anything No, but I think that's an important question. Like, what are we? And that's one of the things when I'm coaching my meteorologist, like, you've got to have this decision as a team. What are we willing to allow AI to do? What are we willing to not allow AI to do? And it's really that second question is the most important. Like we can use AI to identify potential alert days or impact days or whatever TV stations are calling them now. but ultimately we're gonna make the decision as to whether or not that happens. I mean, the Hurricane Center is using Google's Deep Mind, an AI model, to track tropical systems. It did extremely well during Melissa in 2025. And they even acknowledged that it was the best model out of all of them. Are they letting the model do the forecast? No, it's one of the tools. So I I, you know, I think that, yeah, we've got to, we've got to use this tool, but we've also got to be willing to say, okay, what are where's the red line here? What are we not willing to cross? What are we willing to not let it do? And look, for me it was I'm not going to let it write my email responses. But sometimes, in fact, I think I sent you an email recently that actually was written by AI. Ooh, I'm going to go back and look now. Figure out which one. But it's going to be the one with the dash in it. I I always take those out. But for example, the the AI models right now are very good at. Forecasting, but they can't forecast unprecedented events because they're based on historical data. So we know that we can't, you know, we can't just turn over all the forecasting to AI. We can it's one of the tools, but not the only tool. As I mentioned to you before, that it it it can't understand community or complexity. It also can't communicate empathy. So we got to be careful that we're not just turning all our communication over to AI, that we review it. And sometimes we just reject it and do our own. And that's okay. But the it's kind of nice though, I think that in a world where we predict for a living, that there's the unpredictable still, that there's still things that we don't have control over. You know, so there's a lot of discussion right now on the weather service of this worn on forecast, the tornado of worn on forecasting. And, you know, where Ultimately, I mean they're using I think there's an AI component there of figuring out like where the storm's gonna go and and some of that's machine learning, like knowing how storms operate in the future, but it's almost like they're forecasting tornadic development. I think what, like a half hour before it happens, it's like the lead time is exceptional. and not just for tornadoes, but other severe weather events. So do we let that operate by itself? You know, you've got a major outbreak, or is there always somebody there? And I think that's also part of this standard operating procedure. You decide, what am I going to allow AI to do? What part? And it doesn't have to be the whole task. It can be a part of your task, sorting through model data, figuring out impact levels. But what are we not allowing AI to do? And then what safeguards do we have in place to make sure that we're not allowing the AI to function without us? Sounds wild to think about like tornadoes that haven't existed yet and saying that they are going to exist. But there was a point with hurricanes where it was the same. And now it's kind of we just don't even think about it, like this is going to develop even though it doesn't exist yet. We know it's going to. So it's just the science is evolving. Well, the science is and I think that's a good point, Emily, is that there's so much data out there that human meteorologists, it I can tell you from even just a local forecaster that. I struggled some days because there's so much data to look at. Like what do I look at here? I got all of these models for some of them are the next few hours, some but some of them for the next 15 days. Which model's the best model? So one of the things that, for example, that I try to do in my tool is that it does a model analysis for you. So it says to you that here, you know, you can compare the models. You pick a couple of models to look at, and it tells you how those models are, what's the difference in them. How they're performing. I even have in the application a bias built into it. So it automatically calculates the my the bias of the model so that when you're looking at the data, you can click apply the bias correction and it corrects the data based on the bias of that model. And by that we mean for those non-meteorologists, sometimes the models tend to be too warm, too cool, too wet, too dry. And so the application can compare the recent forecast with the observed data. And calculate a bias down to the hundredth of a degree or hundredth of an inch. And it says, okay, this model has been 3.62, 3.62 degrees cool. So let's apply a correction to the data to account for that bias. And it brings the data up, for example. I could do that on my own. It would take me forever. And this is done in one click. So that's an I that's an example of using AI to do one part of. the forecast or I do it to do the whole it doesn't do the whole forecast for me, but it does part of the analysis for me. So that way I can just look at the data and quickly make some decisions. Mm-hmm. Yep. Efficiency. It's been the number one thing for me that's helped for sure. And I think efficiency, it if you work more efficiently, I always use the phrase work efficiently, communicate effectively, because I think it goes hand in hand. If I'm working more efficiently, I'm able to communicate more effectively. I mean able to communicate better. Because I'm allowing AI or whatever to help me work a little bit faster. I mean, that's why we use spell check, right? Yeah, we could, I could go to the shelf here and get a dictionary and look up every word, or I could just let the spell check correct it for me. Remember when that first came out, we were like, we're gonna get too spoiled by this. well, Tim, that's all I have. Is there anything else you want to share before I let you go? It's just that I think that we're in a really interesting time right now that, you know, we're really at the infancy. Of AI, of what it can do. I do think that we're kind of in the wild west of AI, if you will, that there's a lot of gold out there. There's a lot of explorers and there's a lot of pioneers. And right now there's no sheriff. And so things are kind of crazy. And I do think that eventually, you know, the good stuff will come to the top, rise to the top. I do think that we have to be skeptical of how we use it. We have to be careful how we use it. But I think if we have if we all approach it smartly, I think it will benefit us individually and maybe even a society as well. But it requires us to be, all of us, the human in the loop. Thank you to Tim Heller for joining me on this episode of Atmosphere to talk about AI. If today's episode sparked any questions, or if you just want to talk about weather, you can connect with me on Facebook, Meteorologist Emily Gracy, or follow along on Instagram at Atmosphere Podcast. For those who like to go beyond the audio, I'm also publishing deeper dives, episode recaps, and extra content every week on Substack. If that sounds like your thing, I'd love for you to subscribe and join that community there. If you enjoyed this conversation, taking a second to follow Atmosphere on YouTube, Spotify, or Apple Podcasts ensures that you won't miss the next one. Leave a rating or review that also makes a big difference in helping new listeners find the show. Atmosphere is a production of Mike Burst, produced, edited, and hosted by me, Emily Gracie. And if you're looking for a quick roundup of the week's biggest weather talkers, check out Weekly Weather Brief. It's a 10-minute video podcast where I'm joined by fellow meteorologist Karen Jerriman. We give you a rapid fire rundown of what you may have missed that week. It comes out every Friday morning, so subscribe to Weekly Weather Brief on YouTube, Spotify, or Apple to stay up to date. I'm meteorologist Emily Gracie. Thanks for taking the time to slow down, stay curious, and explore the atmosphere.