A TV Meteorologist Builds An AI Avatar That Creates Its Own Forecasts

Most TV meteorologists have well-honed routines logging on for the day: Check the latest data from the National Weather Service and National Oceanic and Atmospheric Administration, start to prep the day’s forecasts and ready the graphics, maybe bang out an early post on social.

For Amy Freeze, it’s a little different: Pour a coffee, play with the dogs, open up the laptop and do a quick quality control check on the fresh round of forecasts her cadre of AI avatars have assembled before hitting “publish” to her digital platforms.

Meteorologists are often TV stations’ most tech-friendly staffers, but Freeze has notched up her embrace by a step-change level. A veteran TV forecaster with more than 20 years’ experience, her most recent stint was at Fox Weather and before that, nearly a dozen years at New York’s WABC. Now Freeze has hung out her own direct-to-consumer shingle, which would itself be a unique-enough move among TV meteorologists, even while a growing number of former national anchors are turning to platforms like Substack to develop DTC relationships with viewers.

But Freeze is pushing past other enterprising journalist solo acts. In just a few months, she has built and deployed avatars of herself that can analyze and present the weather. Currently, these virtual Amy Freezes are pumping out 27 forecasts a day across nine different local markets with plans to expand to 16 cities soon.

Beyond the forecasts, Freeze has also trained an avatar skinned with her body and voice to interact one-on-one with users for more personalized weather queries on her website, a feature currently available only to paid subscribers (non-subscribers can have their questions answered by a more traditional text chatbot; Freeze cites the AI’s API call costs behind the upsell).

“The real crux of an avatar is being able to see it develop a personality and have information so it can create a conversation with you,” she says.

Like any good robot army, these avatars have been mobilized in service to Freeze’s larger ambitions. “I’d like to create an ecosystem,” she says, “and this ecosystem allows me to do as much as I can as a meteorologist. It allows me to maximize my output as a forecaster.”

That ambition, and the steps Freeze has already taken to realize it, already puts her at the far end of the AI news adoption spectrum in the U.S. In an industry where most newsrooms are still kicking generative AI’s tires, she has jumped behind the wheel and put the pedal to the floor.

‘What If I Was A Robot?’

None of these goals were on Freeze’s agenda when she parted ways with Fox Weather last spring and faced a crucible moment in her career. At Fox Weather, a digital- and streaming-centric outlet, she would often spend six or seven hours live at a time anchoring through extreme weather situations.

Freeze saw that after more than two decades in TV, she didn’t understand nearly enough about the business side of it beyond the job she turned up to do each day. That had left her ill equipped for the industry’s speed of change and the trajectory of her own career within it. “There’s a need for forecasters to not let the business happen to them but to be part of the business,” she says.

“I felt like a robot. And I thought: What if I was a robot?”

That question led her down an autodidact’s path through AI technology and avatars. Ultimately, it brought her to collaborate with a developer, Sachin Satarkar, working on a proprietary automation code. That code uses layers of AI to take avatars beyond just presentational functionality, where anchor-avatars are already in use in newsrooms in countries like China and South Korea, for instance. Critically, Freeze and Satarkar’s avatars were taking a step further toward actual data analysis, reading weather maps and producing forecasts.

Training The Avatars

Satarkar’s work starts with a base layer of familiar AI technologies like ChatGPT and then codes on top of it. That coding essentially involves giving the avatars meteorological training.

For Freeze, who has a B.A. in communications from Brigham Young, a B.S. from Mississippi State University and an M.S. in environmental sciences from the University of Pennsylvania (she also is certified in broadcast meteorology from the American Meteorological Society), that meant inputting exactly the kind of material she consumed as a student. “You have to put in a lot of what you want the knowledge to be,” she says. “If I was training you to be a doctor, you would have to read a lot of books, go to a lot of classes, have a lot of experiences. The avatar needs the same things.”

In this case, that includes her own body of work — from past on-air forecasts she’s delivered to blog posts she’s written and technical papers she’s compiled on certain weather subsets like rapid-intensification hurricanes — specifically for the AI to ingest.

In order for the avatar to resemble her, Freeze also needed to spend time in the studio recording different words and gestures to give it her body and voice. So far, she estimates having spent about half a dozen hour-long sessions to do so, wearing different outfits and modeling a series of expressions, gestures and phrases for the AI to adopt. She continues to log such sessions as needed to hone the AI’s presentational nuances.

Three times a day, Freeze now reviews the batches of local forecasts the avatars produce, making sure they pass professional muster and don’t feel too off-kilter from her real persona.

“The avatar goes through a rendition to read and interpret what the data is, and then I can see what that looks like,” she says. “I can go, ‘Yes, that looks good,’ or, ‘No, I would never say that.’ As you do that, you are teaching and training the avatar so that those inaccuracies become fewer and fewer.”

She says she also makes clear to viewers when what they’re watching is an avatar versus the real her in each video produced.

Freeze likens the result to a “digital twin — someone who could think and act like you and hold the information in their brain and be readily available and accessible.”

As noted, that accessibility manifests for her site users as a window to ask direct questions, and it has its guardrails, she hastens to add. Non-weather topics and personal questions are brushed aside. Virtual Amy Freeze sticks to business.

If all this sounds a little too Black Mirror for comfort, part of Freeze’s imperative is to channel fear into fascination for both fellow TV meteorologists and the networks and stations that employ them.

“I really hope that my colleagues will see this as an empowering moment for us, not as a limitation or something that might take away a job,” she says.

The industry’s AI adoption is inevitable, she says, but meteorologists like herself will be indispensable in getting the most out of machine learning. Freeze says TV groups would be mistaken to turn to more subject-inexperienced staffers to train AI. “If you try to use inexperienced people with that, they might know the tech, but they won’t have the institutional knowledge of the meteorology and science required to teach this,” she says. “It’s critical.”

AI’s Business Upshot

Freeze says her AI avatars let her scale her local and regional forecast output, where she’s targeting the biggest markets initially. New York, Los Angeles and Boston are among her current coverage areas, but she plans to keep adding more locations as her quality control allows. The ultimate goal is for users to set their own location for the avatar-delivered forecasts.

This frees her to produce daily content focusing on severe weather hotspots along with more evergreen videos on subjects like hurricane preparation.

Underneath the hood, Freeze is drawing upon a range of other technologies to produce her content, from The Weather Company and Tomorrow.io for weather tech to LiveU and Streamlabs for video streaming. She stresses that most of her weather data and video she uses is accessible by anyone for free, whether from NWS and NOAA or hazard cams from various state departments of transportation.

In addition to her website and mobile app, Freeze also uploads her content daily to a YouTube channel, a Roku app is currently in beta with an imminent launch and she produces a regular newsletter for LinkedIn. She plans to sell advertising into all of those platforms as she draws upon her social media follower base — which reaches back through earlier stints in Portland, Ore.; Chicago; Philadelphia; and New York plus her national Fox Weather following — to build an audience for her new brand.

A major component of her business model is to offer consultative services for broadcasters, networks or individual meteorologists who want to make their own forays into AI use, along with white labelling the technology she’s developed to avatarize herself. She says she’s offering that technology at different levels from the strictly presentational to the analytical, forecast-producing models depending on a prospective client’s preferences.

In early May, Freeze will also launch the second season of Thin Ice with Amy Freeze, a video/audio podcast. The new season largely finds her talking shop with veteran colleagues like Dallas Raines, James Spann, Larry Sprinkle and Audrey Puente.

Getting Back Agency

Freeze says the biggest takeaway of experience so far has been the empowerment she’s drawn from learning the technology and taking the business reins into her own hands.

“I did this because I didn’t want any limitations,” she says. “We all have our patterns that we stay in, but we have to realize innovation is happening exponentially.”

She says that so far, her investment has been more in time and training than in capital outlay, a lesson she thinks broadcasters should heed as they make their own forays into AI newsroom use.

“The technology is available in very affordable forms,” Freeze says. “Money is not the barrier to television stations developing new technology. The barrier is the desire to do it, then putting in the work.”

And for meteorologists like her, it’s about having faith that putting in that work isn’t laying the groundwork for their own obsolescence.

“A meteorologist is not replaced by an avatar,” Freeze says. “A meteorologist enhances the avatar.”