Versu: making NPCs human

versu-5Amidst all the focus on Second Life and the emerging press coverage on its tenth anniversary – with Gamastura being the latest, providing a very short, punchy and positive piece based on the All Things D article which I examined here over the weekend – it is easy to miss the fact that another of LL’s products, Versu, has also gained a little media attention.

Versu, the Lab’s interactive storytelling application for the iPad, launched back in February with little fuss or flag-waving.  Since then, there’s been little news. It doesn’t appear as if the library of titles available for it has grown to any degree, and there has been little in the way of any word from the Lab as to the future of the product. Versu hasn’t been updated since March 2013, the promised Android version has yet to surface, and there has been no news on the Lab’s broader plans to allow users to write and publish their own stories.

However, in June, UK magazine New Scientist published a piece on Versu, AI gets socially savvy, written by Douglas Heaven. The article was born out of Richard Evans, one of Versu’s co-creators, presenting a paper entitled Versu: A Simulationist Interactive Drama, at the Games and Media Event held in May 2013 at the Imperial College London. Heaven’s piece also appeared in New Scientist Online under the title AI makes social game characters all too human.

One of the current Versu titles
One of the current Versu titles

It’s a fascinating read. Not only does it give a greater insight into the power and versatility of Versu – something which is potentially easy to dismiss when first encountering the product on paper -, it also suggests ways in which applications such as Versu could be used to assist with or study a range of real-world social and other situations.

FLIRTY, shy or gossipy… these aren’t the typical traits of a bit of computer code. But a simulation system that gives computer-controlled agents a sense of social propriety could change that, leading to more realistic interactions between humans and characters in games.

So opens Heaven’s piece, before going on to give some background on Evans himself, who is not only one of the two minds behind Versu, but also worked on The Sims 3, before going on to set-up LittleTextPeople with Emily Short, which the Lab acquired in January 2013, and from which Versu was born. Interestingly, one of the motivations to generate such a rich depth to Versu’s non-player characters was the frustration Evans had with the behaviour of characters in The Sims 3 where, Heaven notes, computer-controlled characters would suddenly behave inexplicably, shattering the illusion of realism – such as visiting a neighbour’s house and having a bath…

To avoid this, Versu characters are not scripted, per se, as the New Scientist piece notes:

Versu character screen (click to enlarge): you chose the characte you wish to play, the rest will react to you based on a range of social rules, their own goals, and your actions in the story (image courtesy of Linden Research / Apple Computers)
Versu character screen: you chose the character you wish to play, the rest will react to you based on a range of social rules, their own goals, and your own actions in the story (image courtesy of Linden Research / Apple Computers)

Each of its computer-controlled characters is governed by a deep model of social propriety. They react to rudeness, disapprove of bad manners, and they know a violated social norm when they see one. They gossip, show off and flirt. To make such behaviour possible, Evans streamlined the code that defines characters’ beliefs and desires. This allows them to weigh the consequences of many possible actions before deciding how to behave…

The behaviour of the agents [characters] is not scripted. Instead, they each have a unique set of goals and desires that govern their behaviour. No two play-throughs are likely to be the same.

This approach allows the Versu characters to “play out their own soap opera independently”, which can have some very surprising repercussions, as Evans noted in his presentation. In it, he relates that when testing Versu, he was surprised to find he was being snubbed by one of the characters. It wasn’t until he examined the game’s logs that he found out why. Earlier in the game, he had been rude to another character, who had gone on to tell others of his unpleasantness, with the result that Evans found himself snubbed later in the game.

Because Versu uses social rules to govern the actions and reactions of non-player characters, it is seen as a particularly powerful tool for modelling human interactions, which is in turn bringing it to the attention of academics as well as gamers.

Evans also sees a wider potential for the technology developed for Versu, suggesting that it could be used in a wide variety of activities – such as helping soldiers learn to interact with  civilians in a war zone, or young people to deal with bullies.

Of all the new products launched to date by the Lab, Versu is the one which has piqued my curiosity the most – and it is also the one I’ve found to be the most frustrating, given it is restricted to the iPad and lack of any further news on it coming out of the Lab. Evans’ paper and this New Scientist article therefore both provide welcome insight into the app, and on its potential for the future – assuming the Lab shares in those aims and remains committed to developing and enhancing Versu.

Related Links

3 thoughts on “Versu: making NPCs human

  1. “The behaviour of the agents [characters] is not scripted. Instead, they each have a unique set of goals and desires that govern their behaviour.” Uh-oh. That sounds suspiciously like what I’m researching for my PhD 🙂 My supervisors already told me to hurry up, lol.

    Granted, Versu is 2D and chat, while I’m doing that in 3D and movement/animation, but yes, I can agree that this might be one way of simulating human behaviour which works better than traditional rule-based systems.

    It also raises some very interesting philosophical questions about how the mind works, and, to be honest, when I started my own research — I’m not really a fan of AI, it’s just boring and very complex maths — I had no idea that I’d be questioning current models of consciousness. As the New Scientist notes, Versu characters seem to behave in a human way even if they have no rules to tell them to behave that way. This runs contrary to many current models of the-brain-as-a-computer (not all of which have been abandoned) and makes strong AI proponents grumble and mumble under their breaths…


    1. It’s the range of interactions that the Versu characters appear to enjoy / be capable of performing which also fascinates me, particularly in the level of unpredictable behaviour which can be introduced into such interactions due to the lack of descernable rules (outside, perhaps, of those we would ourselves apply in our own interactions with other people which in turn influence how they respond to us).

      If the model could be mapped into a 3D environment such as (dare I say it) Second Life, or even a purpose-built environment, the end opportunities for simulating human behaviour for the purposes of training, learning, etc., could be quite remarkable.

      Back when LittleTextPeople were first acquired by the Lab, there was speculation as to whether their work would spill-over into Second Life. At the time, the Lab made it pretty clear this would not be the case. However, given we now know the Lab is both investing in new virtual environments (High Fidelity) and more particularly, if we’re all reading Rod Humble correctly, actually developing their own follow-on to Second Life, one can’t help but wonder if we’ll see some of Short and Evans’ work in agent behaviour incorporated into whatever the Lab has in mind for their “future” VWs.

      Good luck with the PhD work!


Comments are closed.