jbsh

Last night was the launch of the CTN here in Bristol at the Watershed. The keynote (which was fully packed out) came from Michael B Johnson from Pixar.

I’m not sure if Michael’s talk will be up on the CTN or Watershed sites as he had been told by Pixar not to allow recording (so this is from memory rather than live notes). The main points that really resonated was the instruction to ‘fail fast’ and iterate quickly towards a great movie. Josh has a great post on the ‘Fail cheap, fail fast, learn & move on‘ approach from a VC perspective. Ewan has a great cross-over post on the approach (or lack thereof) in education. Michael went in to some detail about how they used technology to allow them to creatively generate new plot nuances and stories that could then be refined, whittled, mashed, etc into the final story. Each film gets made twice, once in story and once for ‘real’. The software that Michael and his small team address pain points in the creative process and redistributed the power in intelligent ways. One example allowed the story artists to very quickly sketch directly into a time line to generate roughly edited scenes with their drawings. This got over a pain point (scanning in hand drawn sketch frames so they could be digitally edited) and sensibly redistributed power so that the artists could create a story (which they wanted to do) and the editors got much richer and complete material to refine (which they wanted), and Pixar got to a compelling story much quicker and with less tension between these key people in the process. Win-win-win; everyone’s a winner!

He also talked about the artists in developing story that had four talents;

1. draws really well
2. draws really fast
3. works well with others
4. always has another idea

There was a load more other great stuff (including footage from Ratatouille, early rushes from the Incredibles, and some interesting voice casting for Buzz Lightyear).
In Q&A someone asked about the divide between creatives and technologists and Michael pointed out that there wasn’t a divide. Great software developers (in his opinion and the general consensus during drinks afterwards) was that great coders can code/develop really well, really fast, are good in a team and always have another idea/option/suggestion. The audience was (from what I could tell) a typically Bristol mix of technology researchers from BBC, HP, Bristol & UWE, independent film & screen, digital media, entrepreneurs and social enterprises. I had a really great chat with Tom Alcott (Social Network Company) about the use of social network mapping to improve internal business operations and also about his partner Katie’s social enterprise Frank Waters.

Peter was quick to point out to those in the audience from academia that he wasn’t doing ‘proper AI’, this was a game and he was interested in results.
Most games don’t care about the player; Peter asked if you can feel loved by a game? In part he proposed achieving this through a great story, recognising that you can generate a feeling of being cared for by good character interactions (NPCs), and they hope to achieve this result through AI. The game (Fable 2) has better memory about player’s presence & actions. You can also gain a family and children giving long term engagement.

The core of the discussion was their choice to focus on dog as the AI caring presence. The reason was quite basic; it’s much simpler to do really well. Peter almost immediately came out with the brilliant statement that ‘…the tongue engine isn’t quite right…’.
The dog has 3 basic rules - 1st rule, I will not aggravate you, I will try to help, I will look after myself. Great impact through use of controller, but you have no direct control over dog. He acts as emotional cue and guide (don’t need, or have mini-map). Peter started to think of in-game presence as ‘us’. You have the ability to hurt your dog and walk away, but it will faithfully try to return to you. In the minor demo walk through, Peter engaged a couple of small baddies but got the controller a bit wrong and shot his dog (by mistake). Instead of healing the dog, which he could do, he switched the point of view to the dog, and showed himself walking away from the limping, whimpering dog that was trying to follow its master. It was a surprisingly powerful demonstration of how blatantly you can manipulate emotion and get away with it (and I don’t even own a dog).

More prosaically, Peter noted that possibly easiest way to generate empathy is for NPCs to remember you and your actions. How many times have we gone into a shop only to be faced with the same 4 default option of buying, selling, chatting, or leaving?