HITS 2023: ‘The Good, the Bad and the Ugly’ of Generative AI

After many false starts, true artificial intelligence (AI) is finally here, with generative AI, which is probably disruptive.

But, beyond the hype, many questions remain. As a few examples: What is it really (in)capable of? Where is it going? What and who will it disrupt in the industry? How do we prepare for an AI-augmented future? What ethical rules does the industry need?

Those issues and more were addressed by Yves Bergquist, co-founder and CEO of CortoAI and director of the Project on AI and Blockchain in Media at the Entertainment Technology Center (ETC) at the University of Southern California, on May 23, during the Hollywood Innovation and Transformation Summit (HITS) at The Culver Theater.

Speaking during the session “Generative AI: The Good, the Bad, the Ugly,” Bergquist pointed out ETC is a think tank and research center at the USC School of Cinematic Arts.

“We’re funded by all the Hollywood studios and large tech companies,” he said. “So we have a pretty intimate view of what artificial intelligence means in the studio system because we play a role in supporting the development [and] the technology in the community. And we’re all familiar with what’s happening in artificial intelligence, especially generative AI and what’s been happening in the past six months. I think it’s really important, as we read things and we listen to commentary, to understand and parse out people who are talking about generative AI who are accountable to their audience and growing their audience. And that generally includes a lot of hyperbole.”

On the other hand, he explained: “It’s really, really important to pay close attention to who is talking about AI. And you’ll find something really interesting is people who are accountable to building AI applications that are actually deployed in the real world tend to be a little more conservative in their outlook [for] the capability itself than people who are just accountable to commenting on it. And so I highly encourage you to exert critical judgment when listening to commentary about generative AI.”

Nevertheless, he said: “What is happening in generative AI right now is obviously really, really interesting. Lots of really, really great capabilities. Generally not as powerful as it’s cracked up to be. Generally not as disruptive as it’s cracked up to be. But very powerful. Very, very state of the art.”

Meanwhile, he said: “The methods behind the models are interesting because they’re getting more and more sophisticated. You have models within models within models, multi-mobile architectures with a lot of human[s] in the loop, and all of that is very, very well integrated. And, for the right amount of money, because it’s extremely expensive to develop, creating some very serious capabilities.”

There are, he conceded, “lots of questions and limitations [but] one of the things that no one really talks about is the fact that running a ChatGPT prompt” or a comparable prompt is “extraordinarily expensive and that what we’re all paying to run these prompts right now is not sustainable over the long term.”

Also, he said: “All these companies are extremely financially precarious and they’re just burning through cash because they’re charging us to run these prompts. It’s nowhere near what they need to charge us to be able to be economically viable. So the unique economics of generative AI are still very fuzzy [while] the capabilities are hit and miss.”
Additionally, he pointed out: “The regulatory framework is also very, very uncertain. So there’s just a lot of questions around it. So I just want to make sure that we’re all sort of level on the fact that a lot of the hype of the past few months is sort of dying down in the face of” issues including regulation and economics. Generative AI is also “probably not as good as we’ve been told it is,” he added.

“However, it is very interesting because the philosophy behind all of the generative AI models is fairly new and fairly disruptive,” he explained. “For a very long time in the history of artificial intelligence, you had this tension between what we call the symbolists, who are people who were really putting an emphasis on building reasoning systems, systems that could understand human symbols, and the machine learning crowd … who was thinking that this top-down approach needed to be complemented with a bottom-up approach that was basically the data will learn the features itself. You don’t need to input any kind of feature or any kind of map of the world into it.”

He added: “All of the generative AI models that we see now are what we call hybrid models. They are machine learning models integrated with hand-coded maps of the world called knowledge graphs. And that’s what makes it really, really powerful and really relevant to the world that we live in. Because the world that we live in is not a world that can be understood by just pure machine learning models.”

Generative AI is creating a huge opportunity for AI-enabled content curation, he told attendees, predicting: “As more content becomes available, the need for curating that content, for understanding what resonates with what audience segments at a DNA level is really, really important.”

Matching the right content to the right audience at that DNA level has always been where the main opportunity has been for AI in media and it still is, he said.
The Hollywood Innovation and Transformation Summit event was produced by MESA in association with the Hollywood IT Society and presented by Amazon Studios Technology, with sponsorship by Fortinet, Genpact, Prime Focus Technologies, Signiant, Softtek, Convergent, Gracenote, Altman Solon, AppTek, Ascendion, Coresite, EPAM, MicroStrategy, Veritone, CDSA, EIDR and PDG Consulting.