In films, TV shows and books — and even in video games where characters are designed to respond to user behavior — we don’t perceive characters as beings with whom we can establish two-way relationships. But that’s poised to change, at least in some use cases.
Interactive characters — fictional, virtual personas capable of personalized interactions — are defining new territory in entertainment. In my guide to the concept of “virtual beings,” I outlined two categories of these characters:
- virtual influencers: fictional characters with real-world social media accounts who build and engage with a mass following of fans.
- virtual companions: AIs oriented toward one-to-one relationships, much like the tech depicted in the films “Her” and “Ex Machina.” They are personalized enough to engage us in entertaining discussions and respond to our behavior (in the physical world or within games) like a human would.
Part 1 of 3: the investor perspective
In a series of three interviews, I’m exploring the startup opportunities in both of these spaces in greater depth. First, Michael Dempsey, a partner at VC firm Compound who has blogged extensively about digital characters, avatars and animation, offers his perspective as an investor hunting for startup opportunities within these spaces.
Shadows’ Dylan Flinn and Kombo’s Kevin Gould on the business of ‘virtual influencers’
Fable Studio founder Edward Saatchi on designing virtual beings
TechCrunch: Virtual influencers got a lot of press and investor attention last year. The most talked about was probably Brud’s Lil Miquela, who looks almost like a real human and has done PR stunts criticizing her creators. There’s also Cameron James Wilson’s The Diigitals, an agency of virtual fashion models who each have social media accounts (the most notable being Shudu).
Auxuman has a group of virtual influencers who release original music. There’s Shadows here in LA, which you’ve incubated and invested in and is focused on virtual influencers who are animated characters. I guess my point is: this space got crowded quickly.
Michael Dempsey: That is a good breakdown. I think the space got crowded quickly because of investor sentiment and many people seeing the same gaps in the market, as well as the increasing value of strong IP. You could argue technology also is enabling some of these as well, either on the creation side and of course on the distribution side. Each of these companies have different goals in how they want to engage audiences with their influencers/characters, the scale at which they want to build IP and the types of fidelity they’re going after. I think that increasingly, we could see these various companies bleed into each other at some point, but thus far it’s been very vertically driven in some form with very few characters being truly multi-threat entertainers.
On the virtual companion side with one-on-one interactions, are you seeing just as much activity among new startups? There are a few of them, but I feel like I’ve seen more companies pop up focused on virtual influencers. Most of what I see are chatbots, like Hugging Face, which has been creating an AI friend to communicate with over text messaging.
The virtual influencer space is a lot sexier. It’s easier to tell a story as to why this one company has raised a lot of money and thus this other company should be funded as well. There’s a lot of fast followers. The virtual influencer space attracts an interesting bridge of both creative technologists and pure creatives.
The virtual companion side has a high barrier, given the technological complexity, and also quite a high barrier for achieving perfection from a consumer perspective. When you have a one-to-one conversation with an AI, you can figure out the edges of its capabilities quickly. Even if the environment is fairly structured, the technical complexity is not yet a solved problem.
There is a lot of progress being made on the machine learning front, but it is much harder than just having an artist use DAZ 3D or Unity or Unreal or Maya to design a character, spin up an Instagram account for it and raise money off of that. A lot of people in venture capital loved the narrative of the virtual influencer concept, and it took hold very quickly.
To what extent do you view virtual influencer startups as technology companies versus content companies?
It depends. Some are pure media companies that are very good at storytelling and the core innovation that they are taking advantage of is just new distribution mechanisms that allow characters to have scale at a faster rate than before. I think you can liken that to a Marvel in some ways because you’re storytelling and trying to build IP.
Others look more like technology companies because they are ultimately utilizing and building core technology in-house to enable their creative. I think of Pixar as a great counter of this and that’s who we think about things at Shadows. Pixar is incredible at telling stories, creating new IP and connecting with types of audiences, but they also built proprietary software that allowed them to do 3D animation at a much faster rate with RenderMan and Presto.
If you look at the history of animation, most of the best software has actually come from these siloed studios or VFX houses that spun out the software to give it more leverage. We may see that, but today most of the startups building virtual influencers are storytelling companies, not technology companies.
Ultimately though, it’s a hits-driven business, so the distinction may not matter if the core goal is to build massive IP, but sometimes quantity is what breeds quality.
How important is making the characters look like real humans? A lot of the startups in this space are focused on that.
It splits between certain audiences. There are some people that just won’t watch cartoons or traditional animation (I think this group of people is shrinking in market share). There are others that really love the CG stuff. That’ll bear true in the influencer economy. For a lot of consumers, the novelty factor of a human-looking animation is cool. As time continues to go on, that won’t be as important, and novelty of high-quality CG will decay.
Part of the difference I see between virtual companions and virtual influencers is that the one-to-one interactions of a virtual companion have to be driven by AI because you can’t have a person in your startup carrying on a conversation with every single user. The virtual influencer startups I’ve seen are all scripted entertainment.
That’s a fair categorization. As it is today, it is highly scripted. For a long period of time that will be the case. Virtual companions can have much more scale and can have potentially deeper connections with their audiences if they are able to nail the mechanics of having real-time conversations with persistence over time — that is, remembering the context from your prior conversations like a human would.
Thus far we’ve seen a lot of niche-driven virtual influencers, and I don’t think that is how traditional Hollywood thinks about telling stories. Influencers are stories that unfold over longer periods of time at a higher cadence — potentially daily or multiple times daily via post. I don’t think that skill is as easily honed as people think if you’re trying to tell high-quality stories where people build affinity towards a character and towards the universe for a long period of time.
What’s the state of technology right now that is required to make a compelling virtual companion? It seems it’s mainly about natural language processing, but also computer vision and GANs when there’s a visual representation of the character.
Natural language processing and being able to generate compelling conversational text is getting better. We’ve seen that across a bunch of different use cases in ML. With the visual stuff, there are different edge cases depending on what types of data people are willing to give up.
There is more and more research around using computer vision to understand the emotions of humans, though it’s fairly controversial. You could theoretically have a character that you video chat with and can infer context or emotion from your body language, not just from the text input. That research is fairly novel and I don’t see it being implemented soon at the virtual influencer level.
When it comes to the character’s voice, we’ve started to see synthetic voice generation and voice cloning become significantly better and more scalable and more productized, though I haven’t heard a remarkably good demo that works off of a moderate amount of training data. The productization of these technologies also has proven to be still quite nascent.
This all starts with natural language understanding, because if we want to build up the one-to-one relationships, you’re going to need a lot of training data. What’s interesting about a lot of these one-to-one communication bots is that unlike a traditional chatbot, where you’re sending a single request like “what’s the weather right now?,” you’re sending tens or hundreds of messages back and forth all day. The training data corpus is significantly higher and scales at a faster rate as your user base expands. It’s just a matter of how do you get that user base that is willing to fork over this data.
How far do you think we are from having virtual companions for entertainment that are really enjoyable? For example, they are really funny to talk with or can have substantive conversations.
If we want to get to a point where it actually feels like you’re talking to a human but you keep the conversation on some sort of rails and have a limited amount of creative freedom, I would say in the three to five-year range.
That’s very different from complete freedom of conversation in terms of topics, and understanding where we can use weird voice inflections that throw off the NLU models. That could be a decade time horizon. The last 5-10% is significantly harder than the first 90%.
What do you envision as the biggest use cases by consumers of virtual companions within entertainment?
I think the “AI friend” thing is really interesting. Loneliness is a very real problem. And also having the idea of having this companion that you can speak to freely is really powerful for people who often don’t have that in their lives. These companions cross the border of entertainment to be a wellness utility.
To go on a lighter side, I do think that characters will be able to bridge the gap and build relationships that are valuable, probably with children to start with. The core of it will just be having fun conversations with something you know is artificial but is able to read you based off of a corpus of input.
And I think that before that, we’ll probably get some version of a companion that feels more like an assistant that is pitched as a companion, because it feels nicer but will be utility driven.
I wonder how much consumers will engage with a virtual companion in an ongoing manner after the novelty wears off. Does the knowledge that it’s not a real person still create a barrier where it makes them more insecure or lonely by the fact that they’re not talking to a real person?
I think it’ll shift by age demographic. Kids can do something continually — their novelty decay is lower than adults’. I think that adults will probably experience that in some way.
I do wonder, as it gets good enough, about AI as the stranger on the internet who you’re playing Fortnite with or you’re in a chat room with because the company wants you to be engaged. Will they need to disclose whether this is a real person or not?
I don’t know if we’ll have a renaissance of real humans again. Where we’re going today is people continually want to be more and more loosely connected, but seemingly are less and less deeply connected to people, at least in person. It used to be that parents found it crazy that kids spent time talking to friends or to strangers online on AIM rather than playing outside with friends, and now we spend far more time engaging with lots of people online and less time bonding with specific people in a deep way in person. Perhaps lots of AI companions online is the next step, and people won’t care that so many of their conversations aren’t with real people.
As a VC that’s ultimately looking for a big exit, how do you think about the scalability of virtual influencer businesses? Can they get to hundreds of millions of dollars in revenue by building a portfolio of virtual influencers?
There’s the studio model and the talent agency model. If you look for venture scale outcomes, you build the studio model in which you create unique IP and scale it with distribution across traditional and new channels. As a studio with a whole diverse portfolio of characters, they could be significantly more profitable on a per-influencer basis than normal influencers like Kylie Jenner are today.
My main view is that if you can own a lot of the IP, if you can build true defensible moats in the mid term, you can monetize at a significantly higher rate down the road than even today’s film studios.
Alternatively, you take a talent agency model and figure out best practices of this space and provide that to others as a service provider, taking some economic cut of their success. That is going to be very difficult to execute on a large scale today because the idea of a virtual influencer is not mass market enough yet. I think it could be over time.
Aside from your investment in Shadows, what investment opportunities are you hunting for in this space?
On the virtual companion side, I’m very interested in how to solve a singular pain point. We’re early investors in a company called TalkSpace, which is an online psychotherapy marketplace, and even the idea of using text as an interface gives a lot of the providers more scale. It also opened up new user behavior, where 70% of people on TalkSpace have never been in therapy before. This idea of like TAM expansion and giving scale to both sides of the marketplace is really interesting by introducing this new medium. We’ve invested in this thesis across genetic counseling, women’s health, nutrition and addiction treatment in the past.
Virtual companions could do both of those things in an even more scaled way for different use cases, and, eventually, once we can get the technology good enough, for pure entertainment. You can build really deep brand affinity towards a virtual companion that builds relationships on a one-to-one basis using the same IP.
Comment