AI computer 'learns' children's game

Machine uses video and audio inputs to determine patterns

A 'learning' computer has been created that can work out the rules of the children's game Scissors, Paper, Stone purely by observation.

The machine, named CogVis, differs from standard artificial intelligence (AI) systems in that no data is inputted by a human operator. Instead a motion tracking camera and audio inputs feed into the machine's data pool, which then works out patterns and makes decisions much like a human brain.

"The philosophy is that you can take a camera and audio inputs and the computer can learn about anything from them," said Derek Magee, university research fellow at the University of Leeds.

"Obviously we haven't solved that one as yet, but this is a first stage. Once CogVis had collected the data it needed, it didn't take long to learn the game."

The computer watched volunteers playing Scissors, Paper, Stone and listened while the players announced whether they had won, drawn or lost.

Within 10 minutes CogVis had collected the data it needed and a few minutes later was able to correctly identify the results of games.

The system was designed as part of the European Union's CogVis Project, which has brought together technologists from six countries to develop a "cognitive vision system", a computer that can recognise behaviour and learn from it.

Most AI systems rely on allowing a computer leeway to develop under the aegis of set rules, but when something outside the rules system is encountered the computer fails to cope.

The CogVis system was recently awarded the British Computer Society's Prize for Progress Towards Machine Intelligence.

"The demonstrations showed clearly that the UK is taking some serious steps forward in machine intelligence and how it will be used," said Professor Max Bramer, chairman of the British Computer Society's specialist AI group.

AI algorithms are already being used to automatically examine, visualise and uncover patterns in data.