Virtual Humans: AI complete?

Michael Abrash, Chief scientist of OculusVR has recently said that Virtual Humans are the future of VR. It’s nice that some one thinks it’s important because I’ve spent a large part of my life working out how to create virtual humans.

With modern graphics we can create some pretty realistic looking characters, and motion capture lets them move pretty realistically as well, but, as I said in my previous post, just having a character moving in front of you isn’t enough. It needs to respond to you, it needs to feel as if it knows you are there. We need virtual characters that can interact with people.

AI Complete?

Isn’t that really hard? If I’m saying that I want a character that can interact with me, and interact in a way that is convincing, doesn’t that mean I need to simulate a person. That makes the challenge “AI Complete” meaning that you have to solve the entire hard Artificial Intelligence problem just to have a character in VR. If that’s true, despite what Abrash has been saying, convincing VR characters aren’t going to happen any time soon.

Luckily I don’t think that is true.

Illusions

The reason virtual characters aren’t AI complete comes down to the fundamentals of how VR works: Virtual Reality is an illusion. VR doesn’t literally transport us to another place, it creates an (very convincing) illusion that we are some where else. Mel Slater, the VR pioneer, talks about two fundamental illusions created by virtual reality. The first is Place Illusion, the sense that we are in a different place, this is generated by low level features of VR like head tracking.

The second illusion is more challenging. It is called Plausibility Illusion. It is the sense that what is happening in VR is really happening. That it is a convincing event. The idea of plausibility illusion came in large part out of work on virtual characters (which I was lucky enough to collaborate with Mel on).

One of the main factors in plausibility illusion is whether the VR world responds to you. If you are in a world, but nothing you do has any effect it doesn’t feel like a real experience. On the other hand if your actions have an effect (as simple as branches moving as you brush them with your arm) then the world suddenly seems plausible and you feel a real part of it. This is one of the reasons it is so important that virtual characters respond to players and don’t simply play back canned animations.

Willing Suspension of Disbelief

In fact, what we found when we ran experiments with people interacting with characters, was that people were very willing to read intelligence and responsiveness into characters when it wasn’t there. People would think that a character is annoyed with them or flirting with them, even if it isn’t really doing much at all. If it does respond a little bit, e.g. making eye contact, then people are really willing to relate to the character as a real person, even though they know it isn’t. They are willing to suspend disbelief. They want the characters to be real so they act as if they are.

That means that we don’t have to work that hard to create convincing characters, our players are likely to meet us half way (and unless they are deliberately trolling) will be convinced by the character as long as it responds to them a bit and doesn’t do anything obviously wrong.

The Art of Virtual Characters

At the start of this post I suggested that virtual characters could be a hard, scientific AI problem, but luckily I don’t think that is true. In fact, I don’t think that creating Virtual Characters is a scientific challenge at all, it is an artistic challenge.

Creating a character is about creating an illusion that players are willing to buy in to. That is fundamentally the work of an artist. It is what authors, actors, directors and filmakers do: create compelling characters.

A character isn’t a simulation of a person. Hamlet and Ophelia aren’t complete simulations of people. In fact, part of the power of a great artistic work like Hamlet is what is left out. There is so much we don’t know about Hamlet the character, so much ambiguity that could be read many ways. That is why we’ve spent the last 400 years watching the play and arguing about what it means. That is why each new generation can find something new to read into it.

While I think it will take a long time before we have a VR characters as powerful as Hamlet, the artistry will, in many ways, be the same. It won’t be about creating a complete AI simulation, it will be about putting in enough to be convicing, and what is left out will be as important as what we put in.

VR is now turning from a technology into a medium, and it is time for artists to step up and create compelling experiences and in particular, compelling characters. And what about technologists like me, what should we do? We shouldn’t be creating the characters ourselves, we should be creating tools for other people to create them (but that is a topic for another post).