What If You Could Talk to Your Favorite Character in a Movie? | Christoph Lassner | TED

An image of the Mona Lisa with a twist. TEDAI, on the surface of the Moon. A 3D fantasy world that you can see being explored here. What do they have in common? They weren't created by humans in the traditional sense. They were generated by AI. That is the very beginning of what I call content 3.0. It opens up possibilities we've never had before. Co-creating content together between storytellers, artists, creators, and us viewers. Interacting with their creations, talking with their characters, exploring their worlds and each and every one of us leaning into what we enjoy most in their creations. I expect this to have as profound an impact on the media and entertainment landscape as the shift from scheduled broadcasting to on-demand streaming. A true paradigm shift that will make experiences more engaging than ever. Let's unpack what this could look like. To do so, we'll use a taxonomy of digital content. Content 1.0 is content that is just viewed by you. I use the word “view” today for all types of content we can experience. Content 2.0 is possibly uploaded by you. Think about modern social media. Content 3.0 is generated by you. Professionals may have trained the models, prepared the setting for it, but in the end, you are a part of creating it. Let's have a closer look. Content 1.0 is professional. Often many people come together to create it and the best experiences for their audience. Think about books, articles, music, movies, games, etc. It made and makes a lot of money on the internet. It's also very expensive to make. Entire companies are built around creating it, owning their distribution channels, and aiming to shape pop culture. Content 2.0 is personal. What started with simple home pages for everyone, memes and blogs, evolved to include content with increasing fidelity as consumer recording devices became more prevalent and media-sharing platforms more powerful. This culminates in modern social media. With billions of people creating and sharing content. Between content 1.0 and 2.0, we expect to create over 100 zettabytes of data in 2025 alone. If we could set up an HD camera back in time, recording three gigabytes an hour, a typical HD video stream, this camera would have to start recording 3.8 billion years in the past to record this amount of data. There would not be much motion to watch at first, since that is way before the dawn of complex life on Earth. If this were a 20-episode TV show, the entire period of dinosaurs would just be maybe one episode. And 300,000 years of human history would barely register in the credits. If we are lucky, as a teaser for the next big thing. That is how unbelievably much content that is. And that is just what we're expected to add in 2025 alone. Now, this great amount of data and content is what powers the dawn of AI models and overall content 3.0. It is generative. With these amounts of data and content, AI models can learn to understand it, even to reproduce it and produce it. For instance, large language models need no introduction today. GPT-3 was trained on hundreds of billions of words. Texts from articles, books, social media posts, code. When you prompt it, it will rely on what it has seen before to create a likely continuation. But content 3.0 goes beyond text, and other models can do this for other modalities: code, images, music, all of these are starting to look pretty good. Even video is getting interesting. Let's have a look at what TED AI could look like in the future. (Video) The very first TED conference was held in Monterey, California. Christoph Lassner: And that one, on the Moon. Let's see when we'll get there. But in the meantime, we have this preview thanks to state-of-the-art video and audio models.

Now, can we take this to another level? What if instead of creating a video clip shot from a single viewpoint we could create an entire world where you can explore every angle and you can look around every corner? That is what I'm currently working on as a co-founder of World Labs. At World Labs, we are building systems that let you interact with spatial intelligence, understanding, reasoning about, and generating spatial data. The rooms you see here don't physically exist. They have been generated. Our models can do that from a single image. For artists, creators, developers, and sometimes just to enjoy a beautiful, fantastical environment. Our models can do this because we help them interpret vast amounts of data in a spatial way. For example, this video of a walk through San Francisco you see on the bottom left. Our models then learn to understand what 3D worlds look like, how they can be explored and generated. So now we have defined and understand this taxonomy and the different types of content. But is this everything content 3.0 is ever going to be? When I look at the models we just talked about, I see them as pieces of a puzzle and the beginning of something instead of its conclusion. Of course, we can use content 3.0 models as tools to create more of content 1.0 and 2.0. Just imagine a streaming service where you describe your dream movie and it appears in minutes. A social platform where you upload a video and it's just a prompt, and it is immediately transformed into a professional influencer clip. So all of this may become a reality, but I think it's still not realizing the full potential of these models. And this is where I start to dream about new kinds of storytelling, interactive worlds and experiences, about characters that we can interact with in so much more natural ways than just dialogue boxes, about worlds that we can explore with no artificial boundaries, about narratives that are not predetermined. Since being a teenager and playing 3D games, developing mods, levels, characters, my journey has taken me further in this direction and I think now we have an opportunity to take a huge leap forward thanks to content 3.0. The reason is that we can now create interesting content at or above the rate of consumption. The impact of this fact cannot be overstated. Just imagine Michelangelo painting an image in the blink of an eye, faster than you can look at it. Or Ian Fleming, writing his James Bond novels while you're reading them, and even doing a full movie production for you while we're at it. This means that for every individual viewer, we can develop narrative on the fly. We have never been able to do this before at any point in human history. Improv is so interesting and fun because actors play it out on stage without a plan, improvising what happens next. But this is limited to one narrative per group of actors and shared across the whole audience. We can now have artists and producers create a stage for a story. A villain with an agenda. A hero with an interesting back story. A style for a world to be rendered in. And that's it. From here on out, the story will develop between this world and every individual viewer. Notice how different that is from content 1.0, even content 1.0 games. There, the story is prepared in advance, their characters are scripted. The amount of agency this results in is very different. Even more, characters can break the fourth wall for every individual viewer. This has also never been possible before. Imagine an actor reaching out to every single viewer and having a meaningful conversation with them, or reacting to them. Or the viewer, reaching out to them and telling them where the villain is to be found, or a new idea for a plot twist. So between hunting supervillains and racing sports cars

a future James Bond might turn to you and have a casual chat. Would you ask him about his Wiener schnitzel recipe? I would stick with his martinis. Or strategize with him about where the villain can be found, but you see the point. It opens up an entirely new and so far underexplored medium. So let's have a look. For content 1.0, we have producers come together to create content. I'm talking about this in the most general sense. Writers, actors, directors, artists, everyone who is involved in creating that piece of content. This piece of content is created once to appeal to as many people as possible. It has a high fixed cost, and the producers are paying for production and distribution in advance. So that means that this piece of content has to appeal to as many people as possible. And we have to aim for big successes at low risk, with a very large target audience. Content 2.0 changes these equations. Viewers are producers. We have lots of smaller content productions at lower cost. A social network is shouldering the cost of distribution, matching content to the right audience. Everyone has an incentive to create viral content, but it can be niche. After all, fans of niche content are also producing it. All this goes hand-in-hand perfectly with an ad-driven business model. Content 3.0 has a chance to completely change this, again. The producers can focus on creating a model. This does not even have to be the final content, it can be the setting, the stage, the world for the content to play in, the background stories of the characters and their situation. The content is then co-created with every viewer individually, playing out live, blurring the lines between classic linear media, like movies, and interactive media, like games. And going beyond that. Now, this won't be easy. Initial attempts may fail. Initial attempts may look awkward. But remember, it took us several decades from the invention of movies to be able to make films that stand the test of time. It took us a few decades after the invention of computers, to figure out how to make computer games that stand the test of time, and over the last years, we have only seen the very first experiments with these new tools. Their economics are also fundamentally different. We're seeing cost reductions and static content production thanks to better AI tools. But for fully, on-the-fly generated dynamic experiences, the cost per viewer is much higher than for content 1.0 and 2.0. So even the economics of making hits might look quite different than for content 1.0 and 2.0. This is all currently rapidly evolving, and I'm excited to see how it will develop over the next years. So in short, content 1.0 is just viewed by you. Content 2.0 is possibly uploaded by you. Content 3.0 is generated by you or together with you. Even more, going forward, I expect you'll be in it and a part of the story or interacting with it. The best storytellers of a generation will always tell the best stories with the tools they have. Content 3.0 gives them that. An entirely new set of tools that we are just beginning to understand and explore. These tools will require new skills to master, and a new generation of artists will show us what that can look like. So what do you think, Mr. Bond? (Video) Bond: Hey, Chris. Content 3.0 will be awesome, let's go. CL: Thank you.