MI Banner

The BCS Machine Intelligence Prize is awarded for a live demonstration of 'Progress Towards Machine Intelligence'.

Summary of Finalists - 2010

The SEMAINE Sensitive Artificial Listener

Michel Valstar and Hatice Gunes, Imperial College London

Description of system: The SEMAINE Sensitive Artificial Listener is an artificially intelligent talking head, that is designed to keep up a conversation not unlike one you would have in a noisy bar. Instead of relying on the words spoken by its conversational partner, it analyses non-verbal cues such as facial expression, intonation, and head gestures to decide when to speak and what to say. Just like a real person, it also provides non-verbal feedback, such as a smile, a nod, or uttering things like 'uh-hu'. There are four emotionally archetypal characters: Poppy, Prudence, Obadiah and Spike. Each of them tries to draw you into their own emotional quadrant. We will demonstrate this by holding a live conversation with these agents.

Virtually Roman!

Mario Gongora, Dave Irvine, Gerardo Saucedo, De Montfort University

Description of system: Characters that appear intelligent are common in today’s computer games and other virtual worlds. However, this apparent intelligence is limited to the capacity of the character to tune a limited set of predefined characteristics or skills, all of which are pre-programmed for its specific role. Our new “intelligent” characters go beyond this basic capability by redefining entirely the way each represents its knowledge and roles. We will present an animated Roman virtual person living in a historically accurate version of Roman Leicester. Our Roman does not have a fixed predefined set of characteristics nor a script to follow. The contextual information of the virtual world creates a role for him, but does not limit him to specific descriptions or characteristics. The way he views the world is as generic as any independent individual. We will show him interacting with, and learning from, his environment.

Cleverbot, taking part in a Turing Test

Rollo Carpenter

Description of system: Cleverbot is a conversational AI which every month chats to and learns from millions of people on the web. That data, plus feedback, fuzziness and above all deep context are what set it apart from the rest. During the 2010 presentation, Cleverbot will be taking part in a rapid-fire Turing Test, with the help of audience volunteers. Will occasional user claims of Actual Intelligence happen in Cambridge, too, we wonder...

Monte Pac: A Monte Carlo Tree Search Pac-Man Agent

Simon Lucas, David Robles Contreras and Spyridon Samothrakis, University of Essex

Description of system: Monte Pac is an AI system that uses Monte Carlo Tree Search to provide a form of general intelligence, in this case applied to playing a challenging fast-action arcade game (Ms Pac-Man) in screen-capture mode. The system captures the game window at approximately 20 frames per second, extracts the main game objects, and using its AI tries to plan a good strategy. The plan is executed by sending key events to the game window, and may be changed many times per second as the game unfolds.