How AI media startups can help solve existential issues for the industry and artists
By Ishan Sinha
I encounter a surprising number of AI media startups that offer tools whose use I cannot intuit; whose features make me wonder, “Who would actually want to do this?” And whose models seem to have been built or trained without considering intellectual property (IP) law.
Perhaps there’s an advantage to entering an industry with no prior knowledge. But is it a good idea? In media? Take Napster. Napster disrupted music. But the music labels sued it and forced it to shut down. I believe history shows the music and entertainment industry is unkind to startups that ignore copyright: The value chain is complicated and full of litigious players and established rules.
This is relevant because we have entered an era where AI is good enough to produce believable images and art—where you no longer have to be trained as an artist to create remarkable drawings or songs—and I expect there will soon be a lot of this content. Especially since this next wave of startups can build upon already strong foundational models that are now two years old.
In this article, I explore some of our observations and the enduring media dynamics I believe some generative AI startups overlook. It’s based on our thematic research plus conversations with nearly 80 people building in this industry, investing in it, or directly impacted by it.
Don’t build a legally dubious business
Naivete may free you to innovate, but it may also expose you to age-old risks. Where major labels attacked Napster, they eventually welcomed Spotify, which played by the rules and paid them royalties. Spotify respected copyright. If you’re a founder working in this industry, take the time to learn it. There are layers of financiers and administrators, music labels and movie studios, producers and distributors, each with their own economic interests and legal teams. After our work researching this space, my conclusion is that these players are willing to work with startups, but on their own terms.
Don’t make the mistake of ignoring them and training on datasets you do not have permission to use. IP owners are watching and many are taking legal action. For example:
- Suno and Udio are being sued by major labels for copyright infringement.
- Pocketpair Inc. is being sued by Nintendo for patent infringement.
- OpenAI is being sued by the New York Times for infringement.
- Stability AI is being sued by Getty over watermarked images.
- Anthropic is being sued by Universal Music Group over song lyrics.
- Perplexity is being sued by News Corp for copyright violations.
Those lawsuits are compelling AI companies to pay publishers for access to their content. To me, these lawsuits suggest that the days of unregulated training data are coming to an end. And while we’ve spoken with IP attorneys who don’t think there will be significant legal action to shut down companies like OpenAI, we do believe startups should put aside some of their capital to fight legal cases as they arise.
But consider this. What if instead of running these risks, your startup embraced IP law? Some are building technology to do that. For example, NEX, Raive, and Bria AI are working to build foundation models that allow application developers and IP holders to know precisely what was in the training set. That could greatly reduce the risk and give IP holders the opportunity and incentive to participate, like the Council of Fashion Designers of America, which has partnered with Bria AI.
All to say, we believe it’s best to build a business where the model respects the existing ecosystem.
Try to solve a real problem
I recently came across an AI video generation app where you can type text to generate a several-seconds-long video. The tech is computationally expensive so it can’t actually create more than that. And the output is typically not very watchable. In which case, what is the point? This is one place I believe some AI startups go wrong—they don’t address an existing problem.
There are plenty of existing media problems to solve. As just one example, AI could help IP owners instead of exploiting them. As an artist myself, I take offense to infringement given the effort and dedication that goes into putting your work out there, just to have others use it without permission. Today many artists struggle to prevent others from infringing upon their copyright, and the companies that represent those artists sometimes use software to monitor the internet and social media to, say, block YouTubers from using song clips. But they are finding it harder to protect their work when more creators are remixing and speeding up their songs.
Could your AI tech startup safely empower remixers to respect artists’ rights? To allow for widespread, IP-safe sharing and reuse? Some music labels may be coming around to the idea. “Rights holders understand that this process is inevitable, and it’s one of the best ways to bring new life to tracks,” Meng Ru Kuok, CEO of music technology company BandLab, told Billboard.
Or consider how expensive intellectual property is for artists to create, and what a short life it can have. Artists or labels make money when their music is streamed, but there’s often a drop-off after launch. What if that wasn’t the case? What if an AI platform allowed fans to remix songs with the artist’s permission to spark a new generation of fandom, as has happened for the band Fleetwood Mac on TikTok? (The AI startup Hook — one of our portfolio companies — is focused on creating this platform.)
Or consider smaller artists without the legal resources necessary to license their music. Could startups help them license to local radio stations or TV channels? It’d be similar to how publishers like The Atlantic strike deals with OpenAI, but for everyone else. Two companies, Created by Humans and Human Native AI, are working on such licensing platforms.
Or what about using IP to help with generative AI’s engagement problem? ChatGPT famously reached 100 million users, making it the fastest-growing consumer application in history, but then usage fell. We read this initial spike as evidence consumers are excited, but now these companies must give them a reason to stay. Perhaps media crossovers could sustain this demand. Coca-Cola, for one, opened its archives to allow creators to remix images with DALL-E. The so-called “creator economy” is expected to reach $500 billion by 2027. How can you help those creators monetize while working with existing IP holders?
I believe founders need to think critically about the problem they’re solving. Look at big consumer businesses. They tend to be big because people use them. How are you using technology to allow people to do things they previously couldn’t, and would pay to?
Find a way to differentiate
Let’s say you work at a media-related startup that has built a legally upstanding business that addresses a real problem. If it’s a real problem, there are likely competitors. How do you differentiate? I believe it helps to deeply understand what the media and entertainment industry is facing today to solve those problems in a noteworthy way.
To me, the great challenge in media today is not on the supply side. Artists and media companies want to grow, but they are constrained by the ever-finite sum of consumer attention. Box office visits are down marginally this year. The streaming wars are tapering off. There’s too much competition for too few eyeballs.
Yet I see many AI companies still trying to help creators create more. We can now create far more content than anyone can possibly watch. Streaming consumers aren’t looking for infinite titles—their attention is finite and they’re looking for shows and movies they enjoy, which they say is nearly as important to them as cost. What makes Netflix and Max (formerly HBO Max) great, in my opinion, is the quality of their content. So how can AI startups solve that problem? How can they help build audiences and bring more viewers for more time?
The demand/distribution side is also where I feel the money is made—the value is captured at the connection to customers. For example, Spotify doesn’t make music, and Netflix only makes some of its content. How might your startup open new channels to allow IP owners to distribute their works? I think there is still room for new players. For instance, Webtoon, which offers smartphone-native comics, just went public at $2 billion.
Source: Spotify (as of 12/09/24), Riverside FM, Netflix (as of 12/09/24), A24, YouTube (as of 12/09/24), ProTools, SnapChat (as of 12/09/24), Shutterstock (as of 12/09/24)
Finally, don’t discount the power of fandom. Many artists and creators are worried about AI’s power to create, but I personally believe people are invested in these artists because they are people. Would someone listen to AI Drake and get just as much value? Perhaps. But what many people seem to really care about is Drake’s beef with Kendrick Lamar. Artificial music doesn’t replace real drama.
So how can AI help those artists do more things that are truer to them? Perhaps AI can help them with writer’s block, remaster old footage, or finish an incomplete song as Paul McCartney did with John Lennon’s work. Platforms are flooded with content. If fandom matters, what are the bottlenecks to those fans enjoying the artist even more?
Media companies want to work with AI startups—on their own terms
I believe there is great potential for generative AI startups in media and entertainment. Especially since the foundation models are so good, creators can build on top of them. (Including new, IP-protective ones.) The question is, will they build in a way that plays by the industry’s rules and uses those forces to their own advantage? Or will they fight like Napster and potentially go down in a blaze of legal action?
I’m optimistic and excited to see what emerges. If you’re a founder building in this space, or a researcher or investor, I’d love to hear from you.