Edinburgh, Fife & East Scotland

AI computers learn to spot the killer in CSI episodes

Ted Danson in CSI: Crime Scene Investigation Image copyright CBS
Image caption Former Cheers star Ted Danson appeared in CSI from 2011-2015 when it was cancelled

Computers have been trained to spot the fictional killer in episodes of a TV crime drama.

Artificial intelligence (AI) machines watched episodes of CSI: Crime Scene Investigation and learned to identify the perpetrator.

The study by Edinburgh University was aimed at training AI to solve problems that are challenging for humans.

Researchers say such devices could tackle real-world tasks that need complex reasoning.

The AI correctly identified the perpetrator during the final part of an episode 60% of the time.

The study found people who watched the same show were able to identify the killer 85% of the time.

'More cautious'

Dr Lea Frermann, one of the authors of the study, said: "Pinpointing the perpetrator in a TV show is a very difficult task for computers, but our model performed encouragingly well.

"We hope our findings will aid the development of machines that can take on board - and make sense of - large streams of information in real time."

Scientists mapped footage, scripts and background sounds from five seasons of the show into a format the computers could read.

The data was fed into a computer model that learned to process the plot as each episode unfolded, continually revising the criminal's identity.

Dr Frermann added: "Although humans are significantly more accurate in identifying perpetrator mentions they are also more cautious in their suspicions.

"Our model suspects participants in the plot more eagerly, and starts predicting perpetrator mentions earlier on in the episode."

The study was published in Transactions of the Association for Computational Linguistics. .

It was funded by the European Research Council and Horizon 2020, which are both EU organisations.

More on this story

Related Internet links

The BBC is not responsible for the content of external Internet sites