How can streaming pros deal with all of the copyright and fact-checking pitfalls of using AI systems trained with public datasets as error-ridden and inappropriately expropriated as the internet itself? Ben Ratner, Director of News Technology, Boston 25 News, discusses how to navigate this minefield with Steve Vonder Haar, Senior Analyst, Intelligent Video & Enterprise, IntelliVid Research, Jeremy Toeman, Founder, AugX Labs, and Corey Behnke, Producer & Co-Founder, LiveX in this clip from Streaming Media Connect 2023.
Ratner begins by addressing a question from webinar viewer Beth Rosen: How do you deal with AI hallucinating copyright issues and fact-checking? “So for those of you who don’t know,” Ratner says, “AI is largely built on top of lots of information scraped from across the internet, oftentimes without permission. So how do we deal with this particular issue in media and streaming in general?”
Vonder Haar says, “For me, it’s a garbage in, garbage out type of situation. If you’re going to take information from the Web, God bless you, because you’re not going to really have a trusted source of information from which to draw. The real future for AI will be, at least in the business sense, in the development of limited data sets that are used to inform decision-making within a specific corporate network [and] a specific realm of individuals.” He then asks Jeremy Toeman what AugX Labs is seeing for their customer base regarding current issues with AI.
Toeman briefly explains what AugX Labs does. “In a nutshell, what we do is we take a narrative, we match it with stock content or user-owned media, and we put it together with a little light editing product, so it’s very easy for people to use.” He notes that awareness of digital rights management within the context of AI-generated content varies based on the type of user. “For example, we have a partnership with Getty,” he says. “And so when our customers who are marketers make videos, they are using Getty content, [and] they understand this is rights owned media. When we find random individuals using it, they’re using things like GIPHY, and they’re using Google image search and things like that. So, as we can see, there’s sort of like the known universe and this whole new world. I take a very copyright-friendly approach to the world. So, at all times, our generative components are fully isolated. For example, if you don’t want to use our script writing capabilities, write your own, no problem. We have talked to marketers who say, ‘I’m not allowed to use a script that AI wrote.’ So by us offering sort of a ‘whatever’ level, we get to address users as they need.”
Regarding AI hallucinations, Toeman says, “It’s a real problem. The way we’ve approached it is [as that it is a] developer responsibility. It’s our responsibility that, if you want to make a video about tacos, in the middle of it, there’s not all of a sudden a discussion on hamburgers. That’s on us. And so I think candidly, that’s up to every product team out there, whatever the tools they’re making, whether generating images, video, text scripts, whatever.”
“A taco was just a hamburger you didn’t create well enough, and it kind of just fell apart on the grill,” Ratner jokes.
“That sounds like an SNL skit, but I’m not sure!” Toeman says.
Ratner asks the group, “Are you dealing with putting your own information into these models? Information on video scripts, that kind of thing? Are you just using ChatGPT’s existing library of knowledge?”
“I can speak to that,” Behnke of LiveX says. “We have a product called switcher.ai, and we do not use the dataset that’s public. I am a technical director. I have been over the course of my life, and I know other technical directors. One of the things that I did for the language model was if I’m going to have an AI replace the technical director, I need to have the inputs be from something I know. And a lot of live streaming and live broadcasts are very formulaic. Let’s take a house of worship: I’m going to have the pastor, I’m going to have the choir, then I’m going to go back to the pastor, then I’m going to go back to the choir. It’s very formulaic, take sports, the NFL.”
Behnke emphasizes that massive outside datasets are unnecessary if you’re working within the predictable parameters of specific types of content. “So if we take the model and then we give [each] model a main formula, we don’t have to use outside dataset,” he says. “I think right now, in this world we are living in, that will differentiate everybody else. If you’re just using one whole public dataset, I think that you don’t know where everything’s coming from. So it’s really important to own your data set in a way you know exactly what’s inside of it.”
Learn more about a wide range of streaming industry topics at the next Streaming Media Connect in November 2023.
Generative AI is a game-changer for all sorts of businesses, and how to leverage it is a key strategic and technical concern for a range of streaming organizations looking to monetize their content and operations. Darcy Lorincz of Barrett-Jackson Auction Company, C.J. Leonard of Mad Leo Consulting, and Reality Software’s Nadine Krefetz explore generative AI’s current and imminent impact in this clip from their panel at Streaming Media Connect 2023.
29 Aug 2023
Ai-Media’s Matthew Mello talks with Tim Siglin about the evolution of AI captioning in this exclusive interview from Streaming Media East 2023.
21 Jun 2023
Randy Bapst of AiBUY sits down with Tim Siglin to discuss AI-based OTT overlay transitions in this interview from Streaming Media East 2023.
12 Jun 2023
Tim Siglin of Help Me Stream Research Foundation sits down with Ben Ratner, the Director of News Technology at Boston 25 News, for a chat about AI, social media, and the future of broadcast news.
01 Jun 2023