Since the November 2022 launch of ChatGPT, an AI tide has risen around the media business.
Existential uncertainty has seeped into storied entertainment professions, from writer to actor, executive to special effects designer. But the flood waters are also fertile with potential creative renewal and (for the C-suite) radical breakthroughs in cost-efficiency.
2025 will be the year the levee breaks. Whether the effects are good or bad depends on where you are within the media industry.
Partly, this change will be via the media’s own, belated engagement. Disintermediated by the vastly bigger tech industry (chip maker NVIDIA alone is almost a 1,000 times bigger than ITV, the UK’s premier listed TV company) film, TV, music and games executives I talk to have transitioned from a position of ‘Not for us’ to: ‘Are we behind?’ Production and distribution companies are now testing AI at scale.
Nerves are on edge. Copyright lawsuits, licensing deals, startups, crawler blockers and more all abound to address the problem around who owns the models created with copyright-owned training data. Contrary to wishful thinking in media board rooms, the question of whether machine reading content to train a mathematical model constitutes legitimate engagement - or just theft - is not resolved, even at a philosophical level, let alone in any courts.
Competing uncertainties such as these will reshape every sector of the media this year. So here, I am offering specific predictions, plus a final, reassuring observation.
Text-to-video goes mainstream
The first prediction is that TV and film production studios, having been exceptionally slow and cautious about embracing generative text-to-video AI tools thus far, will pivot to a nuanced, AI-integrated approach.
Whilst talent unions blocked the road, briefly, in the Hollywood strikes of 2023, the C-suite mood has shifted from conciliation and fudges, to proactive investigation of new implementation routes, or even production territories.
This is based on the existential risk of disintermediation of their high-priced product by AI-driven market entrants which are unwilling to respect the de facto 20th century oligopoly of content creation studios. It's not nice to read, it will change some industry jobs, but this is undeniably true.
Cris Valenzuala, CEO of New York-based, high-end AI video engine Runway said: ‘Runway is not an AI company. Runway is a media and entertainment company.’
That credible threat, along with fabulously well-funded Chinese tools Hailuo/Minimax and Google Veo (all excellent), Pika Labs and OpenAI’s Sora (strong but falling behind somewhat), Adobe’s rights-cleared Firefly and Meta’s meme-friendly image generators (both currently weak but scalable) is a real danger to media companies if they don't embrace the opportunities for low-cost video creation. So they partially will do so, well aware that Chinese AI companies will not slow down just because US talent unions want them to.
Prediction: Expect to see all the major content studios integrate AI tools substantively into pre-and post-production, world creation and other visual dimensions of their work. This generative AI may include creating background artists (‘extras’). But do not expect focus on AI actors in substantive roles.
Machine learning (outside generative AI) expands
A point often not understood in the media is that generative AI is a very small subset of the wider field of machine learning (ML) in general. So the second change is that the statistical power of machine learning tools at large, not just the more controversial generative AI, will start to be embraced not just as creative engines for making audio and video. They will be a powerful suite of new analytical sources that could power a completely different dimension of programme and project research for an industry that arguably has creatively flatlined, from history to lifestyle programming.
For instance, taking recent academic research using machine learning to decode communication in the natural world - elephants, bats, fish, whales, even trees - could underpin an extraordinary reinvention of natural history programming, the highly scalable global model of high end Life on Earth type of TV series, and one of the few international broadcast TV genres still firing on all cylinders. The same is true for history programming, where machine learning has already decoded charred scrolls and discovered entire ancient cities in the Yucatan jungle, plus where direct reconstruction is now possible on low budgets.
Prediction: Expect to see transformational new TV series on human/animal communication, and mid-evening TV shows using ML tools to recreate a 1920s railway journey across Britain as it might have been seen from a steam train, or life in Ancient Rome.
Ask the audience (replica)
The third change is that synthetic audience models now being used by advertisers at major agency groups like WPP and Omnicom to test and learn different new product ideas, pricing changes and marketing campaigns, will find their way into entertainment too. This is where AI replications of the audience is used for market research, rather than the more expensive and slower options of real-world market research, or A/B testing in the wild. Just as Harvard researchers accurately checked the demand impact of a price change by asking ChatGPT, so this approach is starting to be used in the media and entertainment industries, whereby specific content ideas can be tested and iterated in a similar fashion.
We’re seeing film studios use tools to assess the potential future performance of scripts against legacy box office performance of analogous material.
Prediction: We will see both TV broadcasters and production companies start to embrace this notion of the sophisticated audience model for benchmarking ideas against their potential response. This will elide the creation process and mean that producers will arrive at the moment of pitch far better informed and with much more sales information to back up their proposition.
Fear not the paper tiger
None of that means that anyone in entertainment believes that AI will replace the human voice. That risk is an AI-generated paper tiger.
In 2024 I spoke at over 50 events with top level media organisations around the world. No one is talking about using AI actors for major roles. And not once did any executive or major league creator say to me that they think there is a likelihood that they will pivot their output to AI as the storyteller, rather than the tool deployed by the storyteller.
All of which could mean that AI in entertainment is going to move much faster in 2025 than it has done for the past three years – but it’s not going to replace the media industry itself.
Dr Alex Connock is Senior Fellow here at Said Business School, University of Oxford. He teaches on our MBA and Executive MBA programmes and wrote the 2022 book Media Management and Artificial Intelligence (Routledge).
