During a revealing Game Developers Conference talk, designer and Lionhead founder Peter Molyneux explained his studio's approach to internal experimentation -- and demonstrated a number of projects. He began by thinking back to the company's origins. "We were sitting in a pub getting drunk -- that's how companies are born," Molyneux remembered. When the team was only comprised of about 30 members and working on Black & White, "innovation was easy and cheap," he said, but "it really was utter and complete chaos." During development, some extraneous ideas -- such as Molyneux's intention for the in-game weather to match the weather outside in the real world -- ended up eating up lots of resources and taking focus away from core mechanics. But still, the designer argued, "If I was a betting man, I'd bet that what our customers really want is innovation. They want to be shocked and surprise and awed every time they spend an untold fortune on that game at retail," and the press "wants innovation, feeds on innovation." Furthermore, developers love to explore new ideas. And if they are unable to do that within their existing place of employment, "What do they do?" Molyneux asked. "I'll tell you what they do -- they leave their company and go somewhere else, because their idea is so smart." All of that is why Lionhead established Lionhead Experiments, a forum for small experimental teams. "If you have a small experimental team that can spend a man-month," you can do a great deal of experimentation, without being financially prohibitive, Molyneux said. Such projects range from one to 12 weeks, averaging a four-week cycle, with teams of one to five developers. Even lower-level developers can propose those ideas, but they must be "sponsored" by a senior member of the staff as a "shepherd" for the project -- that sponsore then pitches the idea to an internal Lionhead board of about eight staff that oversees the entire Lionhead Experiments program. The company maintains a high-level prototype engine expressly intended for rapid creation even by non-programmers, tied into existing game engines. Lionhead also has a structure called "concrete" that allows assets from one game -- say, a tree from Black & White -- to take directly into a totally different game -- such as Fable 2. Further off, Molyneux would like to expand that out to systems like AI, to allow prototypes to be quickly assembled in a modular fashion. Later on, the board assesses the success of the experiment: Why is it being attempted? How much would it cost to make? How often would it be seen in a game? How would it be taught to the player? Ultimately, the final question is whether the idea should be executed and put into a game. Interestingly, in the past, experiments have made it into Lionhead games but ended up not being significantly used by the games' designers, so the studio now puts more emphasis on ensuring designers are familiar with the additions. "Where does the money come from?" an audience member asked. "Well, we do a lot of busking on streets," Molyneux joked, then explained that experiments for a particular game come out of that game's budget, while broader projects come out of the budget for Lionhead's Central Technology Group. "We generally find it rare that you can't justify it," he said, since the teams and development cycles are not extensive. "They're very small amounts of money." Experiment Examples After his overview of the process, Molyneux demonstrated a number of actual experiments. He began by showing an early version of Fable II's dog, which he himself designed and which ended up factoring heavily into the full game. "This is probably one of the most valuable experiments we ever did," he said. Using the original Fable engine, the team asked itself, "Why don't we think how the dog can actually move and be a companion to the player?" They decided to focus on exploring what a dog would do, rather than try to slot a canine into existing typical video game companion tasks. This led to the mechanic of the dog running out in front of the player, rather than beside or behind the player as most game AI companions are positioned, which had a huge impact on the dog's role. A number of experiements were demonstrated in quick succession: one used the graphics chip to generate thousands and thousands of animated creatures (which ended up not making it in time for Fable 2, one that took one coder ten days to develop more advanced fluid physics effects, one about using lighting for shoft shadows, one that attempted emotive facial expression technology. Molyneux then referred back to a talk he once gave about his frustration that more people couldn't experience combat-heavy video games due to the dexterity required to operate multi-button controls. Following the talk, a Lionhead employee told him, "I've got this really cool idea for something called one-button combat," so he was set up with Lionhead's prototype engine to try and design the heavily context-sensitive systems. That ended up being the core of Fable II's combat -- "The cost was enormously high implementing that into Fable II, but the reward was massive," he said. Next up was something called The Room, which Molyneux had previously demonstrated at a past Game Developers Conference and which he said ultimately proved unsuccessful after spending a great deal of time on it. Set in an elegant living room, it allows players to shape objects with small blocks of clay, which the game then turns into nicely-modeled objects, but also contains portals (similar to those of Portal, but developed before the release of that game) that not only transport objects in real-time but also change their size and other properties in proportion to the relative size of the two portal openings. Molyneux noted that the portal technology can even be used to connect two different players over an online connection, but unfortunately, "it wasn't really an experiment that ever went anywhere." Interestingly, some of the developers behind that very project ended up going on to leave Lionhead and form Media Molecule, which released last year's LittleBigPlanet.