Í viðtalinu rekur hann hvernig gervigreind er notuð þessa dagana í kvikmyndabransanum og hvernig notkunin gæti orðið.
Hér er brot úr viðtalinu:
What are the most important aspects of AI for filmmakers?
It’s the creative potential. The recent advantages AI has made in the field of creative capabilities is a game changer. When we were talking about AI three to four years ago, it was all about optimising repetitive processes like mathematical and organisational tasks. It was about deep learning, speech recognition, and so on. We thought this was something which could potentially disrupt the blue-collar jobs, but not the white-collar jobs. Not the creative ones. As long as we had creative jobs, nothing could happen to us in terms of digital disruption. But in the autumn of 2022, some major milestones occurred simultaneously: The release of ChatGPT, the advancements in generative AI for the creation of pictures and moving images. All of a sudden, it was pretty obvious that this was something that would not only affect the organisational, but also the creative tasks, which makes it especially relevant for creative businesses. This technology is a general-purpose technology. It’s not an innovation. It’s much more groundbreaking. Like the invention of the smartphone or the Internet, or like the invention of electricity. There is a lot of things you can do with this, and we are still in the very, very early initial phase.
What are the most powerful AI tools for filmmakers right now?
The large language models are already a super-efficient tool. Our first rollout wave within our company, which we are currently working on, is making these models available to all our employees. An accompanying task is to roll out governance: What is the framework under which we can actually use these models? What can we use them for? What do we have to keep in mind while using them? There are has several aspects to this: Rights issues, ethical issues, and so on. You have to find a technical solution in order to make it available to the whole team. You have to invest in training in order to make everybody familiar with how to use the tools. This “capability building” is something we are very much focused on right now.
There is valid research data indicating that if you roll out these tools to all of your employees, you will see a leap in productivity and creativity. There’s a very good research study from Harvard University from last summer in which they carried out tests of certain tasks. With one sample group working with AI and another sample group working without AI, fulfilling the same tasks, and a human sample group that had to evaluate the results. It was obvious that the sample group working with AI was much faster, better and more productive than the sample group working without AI. So, this is something which carries a lot of unreleased potential. It’s something you want to roll out at a large scale as fast as possible.
What are you using AI for now?
We are currently working on using AI to summarise scripts, and this works pretty well. But we are experimenting with all aspects: It could be research, project development, assessment, estimation, economic projections, legal aspects… Really a lot of things. These models still tend to make mistakes, ignore certain parts and hallucinate other parts. But it’s getting better and better and better. With every month, practically. But a qualified person still has to oversee and judge the results.
What are the wildest perspectives?
Maybe my wildest perspective for the next five years is that general artificial intelligence is going to be discovered: That you will have human-like capabilities in AI. Once you have achieved this, it won’t take long until you have superhuman capabilities. Because the first thing you would ask this AI to do, is to create another version of itself, but a better one, and ask this one to create an even better one and then a better one, and this adds up to a scenario which is widely described as the “singularity”. If we are getting there, my wildest prediction is that we will either end up in Utopia or we will end up in Dystopia, like we have seen in numerous films.
How do you think AI is changing the way films are going to be produced in the future?
The way I see it, there are two paths: the transformative path and the disruptive path. As for the first one, we are right now talking about optimising scripts and pre-visualising ideas, enhancing the visual effects, and so on. This is more or less the way we’ve always been producing films. It’s a kind of evolution, some parts are getting better, easier and cheaper. But the bottom line is, we shoot films the same way we did before: Actors in front of a camera and the whole crew around it. However, when you think about more disruptive AI infused ways to shoot, it currently looks like there are two approaches. The one is complete virtual production, which basically means that the shooting is not much more than just motion capturing of the actors. So, for instance, you have a scene and a room, and a couple is having an argument. You have some trackers in the room that capture data points from their facial expressions, from their body movements. Then you export the whole data set into a game engine like Unreal or Unity, and you create the whole film in the game engine. You need a director, two actors, the motion capture technicians, and a very focused post-production team. The actors don’t necessarily have to look like the characters that will appear in your film, and they don’t need to have the same voice. This way of production enables 100% creative control of the content and total creative freedom for the director. So this is one vision. The other vision is to completely AI-generate the films by describing what you want to see. For example: I want to see a shot in a medieval village with a handheld camera following my protagonist, and I get one. So far, the outcome is as predictable as a slot machine. You put in the prompt, you get something, but not necessarily what you had in mind. You don’t know exactly what this will look like, what the actor will look like, what the acting will be like, and so on. Even if you give specifics, you still get a very wild result that’s not really controllable yet. And the consistency between the different shots is also a topic. However, this technology might evolve to a point where you have consistency and control on the output. The first improvements are already appearing, but they are still far away from the level a filmmaker would need. If at some point a model will understand what you need, create what you want and offer you options, you can start fine-tuning: “This direction is good, this direction is bad,” pretty much the same way you’re working with human crew members and actors today. Once AI has developed to this stage, everybody can create a film at home by just describing what he or she wants to see. But still the limit is your creativity. An exceptional artistic individual is required in order to describe an artistic vision. However, these are the two paths right now as I see them. The first one, “the virtual production”, from what I’m sensing around, is a little bit more realistic at the moment than the second one.
What about AI aspects when it comes to distribution and financing? How will AI influence the business models?
There are two buzz terms that I often read or hear: “hyper localisation” and “hyper personalisation”. “Hyper localisation” means that, if you are in Mumbai, you’ll see a different version of the film than somebody for instance in Iceland, in terms of product placement or advertising. “Hyper personalisation” means that we no longer produce content for a broad mainstream audience, but for more narrow special interest target groups, and create viewing experiences that are tailored exactly to specific audiences in terms of content, but also with respect to advertising and product placement. A young family with a new-born baby would see other products or advertising than people in their 60s, and so on. The whole creation of content for the audiences will be influenced by this, and even a completed film may contain different images when shown in India, in Germany or any other place around the world. And this feature can be driven by AI as automatic optimisation. There’s no human interference. It runs by numbers. It’s quite wild.
But does the consumer actually want this?
That’s the big question, and I don’t have an answer to it. If you’re in the office the next day and everybody ended up watching a different version of the show the night before, what do you talk about? It’s a question of if it works that way in terms of “lean back entertainment”. There have been some tests regarding interactive production formats, and it turned out the audience wants to lean back after a long working day and just experience the artistic vision. It’s a campfire experience. We all want to sit together in a cinema theatre or at home and watch what the artist wants to tell us. The question is, will all these technical capabilities actually lead to something that the audience will accept? Will the audience embrace “hyper personalisation” and “hyper localisation”? If we see a film made by James Cameron, it’s totally okay for us as viewers that the main part takes place i.e. in an American house, it doesn’t need to be a localised house.
What about the rights issues?
On the one hand, we have to protect our business model, that’s how we make a living. My professor in film school 25 years back once told me: “You’re actually not a film producer, you’re a rights producer.” If you don’t secure your rights, you basically have nothing. So, this is something I’m very much aware of. On the other hand, there’s no way to stop progress. It’s a global phenomenon. Either you surf the wave, or you get overwhelmed by it. You have to be a part of the progress. But one basic thought may lead the way to finding the right solution: The productivity gains should somehow be distributed amongst the people. We are dealing with a general-purpose technology which is much more productive than any technology in the past. That’s great, but who will reap the profit from these productivity gains? I think now is the right time to think about how you let people participate in the productivity gains created by artificial intelligence. And that’s not an easy task. The ultimate compromise would probably be organised through “collection societies”. If these models earn money, a part of this money has to be distributed.
