Virtual robot takes television interview


Dr. Kristinn R. Thorisson, Associate Professor in the School of Computer Science, along with his colleagues published a paper at the conference Intelligent Systems & Agents in Portugal last month. The article won the prestigious "Outstanding Paper" award as the best paper of the conference.

The paper describes a new phase in the AI ​​research - the first automatic system that can learn a complex task without instruction by observing how it is performed. The paper describes how Kristinn´s and his co-authors artificial diagnosed virtual robot S1 learns to take a TV interview by observing humans in an interview situation without it being given any prior information on syntax, or any other information on how to perform the work.

After the virtual robot observed interview situations for 20 hours, the S1 robot could be interviewed using fully correct sentences and interpret and use hand and head movements in line with what happens when people talk. S1 can either take the role of interviewer or the interviewee. 

The studies are the result HUMANOBS project, which received two million funding from the European Union in 2009, in an effort to develop a system with universal intelligence.