The BCS Machine Intelligence Prize is awarded for a live demonstration of 'Progress Towards Machine Intelligence'.
The annual conference of the BCS Specialist Group on Artificial Intelligence hosts the BCS Machine Intelligence Competition. The competition is definitely not a group of researchers reading out academic papers, it is a step into a world of risk for most competitors. In the words of Rick Magaldi from British Airways, the competition is an opportunity for system developers to fling themselves into the unknown in front of an audience of critical and knowledgeable observers to see if their systems can fly, figuratively speaking.
Each competitor offers a live demonstration of a system that is judged by how much progress it represents to the development of Machine Intelligence. The audience don't know about the technology or the theory, they judge the entry on what it actually does during the demonstration.
This year, four excellent demonstrations produced a close run competition, but the winner, with a late goal, was a system presented by Martin Rhodes and Simon Coupland from De Montfort University, that uses evolutionary algorithms to work out which of thousands of options is the best way to take a free kick. The 15 minute demonstration consisted of several examples, each based on an actual free kick. The demonstrators first showed a real free kick from a recent soccer game and then set their system to work on the same problem faced by the kicker. A screen display showed the system testing out many ideas before suggesting the ones that were, it thought, most likely to result in a goal. In some cases, the system derived the same kick as the one actually used by the kicker before suggesting alternative options. In other cases the system disagreed with the kicker and suggested a different option from the one used.
Whilst the human soccer player probably drew on years of experience and practice to work out the best kick; the computer system had to rely on the physics of ball spin and momentum and a range of other variables to produce a winning kick. One of the kicks tested by the system was a goal scored by David Beckham. Maybe his secret was to use a similar system to learn from, or maybe not.
Other entries were equally impressive. Hemin Omer Latif demonstrated GazeBot, a mobile robot from Nottingham Trent University. This robot managed to locate and track a person on stage and proceeded to follow him wherever he went. The robot followed him out of the lecture theatre and along the corridor. The audience could watch progress on a large screen display, through the eye of the robot. This was all done with a single vision sensor and no other measurements were taken.
Chris Huyck from Middlesex University presented an intelligent agent (CABot2) in a simulated game environment that used `fatiguing Leaky Integrate and Fire' neurons to interpret natural language commands to move about and perform actions in the simulated environment. Finally, David Burden from Daden Limited, demonstrated an avatar called `Halo' that lives in the popular Second Life virtual world. Halo looks just like any other avatar in that environment, all of which seem to be controlled by a human being living somewhere, anywhere, in the world. But not Halo. She is an avatar that is completely machine driven. Other avatars (people) can meet Halo and talk to her just as they would to any other avatar. But would they know that they were, on this occasion, actually talking to a machine? Will machines first take over virtual worlds before turning their attention to our world?
The competition in 2008 was extremely competitive and fascinating to view. It was also highly challenging for each of the competitors. These four competitors were initially selected from a much larger list of entries. Some very good systems were not selected for the final competition, not because they were not good enough, but typically because they may not be so easy to demonstrate effectively in a live situation to an audience that would vote on what the system did and not what it was made of.
The challenge is there. Can anyone do better? Is anyone else prepared to risk their reputations on a 15 minute demonstration of `so called' Machine Intelligence? Everyone will get another opportunity in 2009, at the next Machine Intelligence Competition to be held at AI-2009 in Peterhouse College, Cambridge next December.
The Competition is intended to stimulate work on Machine Intelligence in the United Kingdom. Each year, it seems to be doing a better job. Let's hope that more researchers continue to bravely demonstrate their systems. Maybe Industry in the UK can be encouraged to play a bigger role and really show what they can achieve. More information about the competition can be found at http://www.bcs-sgai.org/micomp/ and the organisation that hosts the conference at http://www.bcs-sgai.org/.
(Photo 1) Simon explains in general how the real world free kicks are set up in the Machine Intelligence system. He then watches in anticipation to see if their system actually comes up with a winning kick, whilst being watched by the knowledgeable audience of conference delegates. Martin, out of shot (so to speak) does the setting up and then launches the application.
(Photo 2) Professor Max Bramer, Chairman of the BCS Specialist Group on Artificial Intelligence (SGAI), presents the winners Simon Coupland and Martin Rhodes (left to right) with the prize. The winners also received a check for £1000 from the SGAI. Martin is to continue his success with further research into the uses of Machine Intelligence in the Sports Arena.