When you look at your system in FIFA 10 as compared to the last one, it’s clear where you’ve expanded on shortcuts or scripted sections of that code. Do you look at the current system and see where the expansion will come in the future?
I think I kind of do, yeah. We have all these things which are done physically accurately. So ball physics are absolutely 100%, we tried to model the right formulas and everything. The collision system, again, we’ve got the systems down. But there are other systems, the jostling system, which is more based on an algorithm which we created, an algorithm which manipulates the collision system and the ball physics and whatever to do what we need it to do. And ideally, five years from now, we’d want to use the real physics information to create that, but obviously it puts greater demand on our database with weights and heights, it puts greater demands on our coders, greater demands on our collision system, with the ragdolls or whatever we’d have to use to create it.

For example, when the jostles come together we play an animation to make them look like they’re fighting for the ball, but in real life it’s not really like an animation, it’s like you’re still locomoting and then the collision drives you to change the animation procedurally. And that would open up a lot more opportunity for the visuals and the gameplay inside just that feature itself.

So I do see where those things could go. Even with the ball striking itself, we’ve got it really complicated, I’m quite happy with the way it uses all the context and stuff, but it doesn’t actually kick the ball – the animation doesn’t actually kick the ball the way we calculate it to. So it doesn’t actually use, like, ‘you kick it with 12 pounds of force’ and stuff like that, we’re a level abstracted from that, saying ‘you kick it at 56 miles per hour,’ you know what I mean? Which is different from saying ‘you kick the ball with X newtons of force and you kicked it on the left hand side’ – it’s slightly different, I guess there is another level of complexity and physical simulation that you could do, which would give you diminishing returns, I think, but would still give you returns of a sort.

Is there anything you’re itching to make much more complex?
There is, actually. It springs to mind occasionally when I’m in the office, and I think ‘that’s gonna have to wait until the next next gen.’ I know that there’s this idea of player-level intelligence that I’m keen to work on, so they don’t have this team-level AI telling them roughly what to do, it’s all based on their decisions, that’s something which I’m interested in. I’m hopeful that maybe in the next generation – the last generation was a lot about graphics, and maybe this generation isn’t so much about graphics because the jump isn’t so big. And I’m hoping the next one is maybe more about AI and about how we can use that.

In gaming, there are very few companies that use classic theoretical AI systems, the kind you’d get taught at university. I mean a lot of those things just aren’t feasible on our consoles, but there’s not many people who use proper AI, groundbreaking AI techniques or anything like that. We all kind of just make our own way, and use old techniques and old methods. And some of them are very good for what they’re needed for, but things like neural nets and things like that, academic techniques, are not getting used in videogames. Maybe there’s not a place for them, I don’t know, but I think there’s a place for something there with AI.

Could you explain how you’d see that impacting on the player’s experience?
I think you’d feel it in the variety and the mistakes. My feeling, and obviously I play the game every day, is that there’s lots of way that we can make the CPU do something like, say you play the game and the CPU learns how you play the game so it combats the particular way you play, and then maybe you have to change the way you play. I mean, fighting games do it a little bit, but obviously it’s more complicated for 11 on 11 football.

Maybe it could learn your game so that when you play in the online club system and you can’t be there, it plays for you. Those kind of things are possible. Just in general, the CPU’s on-ball behaviour needs to be improved, the positioning, while I think our positioning is pretty solid, there’s things we can improve within that, or the goalkeeper, the way he makes decisions about coming off his line and going back. I guess it just takes a better understanding of how the human brain works and what kind of things they’re thinking about when they’re making their decisions.

If we could separate the player behaviour into player-level decisions rather than just team-level, it gives the opportunity for a player to make different kinds of mistakes, and that would open up greater variety in gameplay. I think also things like when you press pass or through pass, just the way the AI analyses which option’s better, because we’ve got more processing power and maybe we’ve invested in some AI technology, the way we analyse space, or the way we do something – we can come up with better results, and more realistic results, more football effective results.

And I guess if we simulate physics absolutely properly, every time the ball’s kicked it would behave realistically, right? There’s still a couple of occasions in our game when it doesn’t feel right. Things like that. [laughs] It’s certainly a challenge for us going forwards to work this stuff out.