AI computers learn to sleuth after watching TV crime show
Computers in a Scottish laboratory have been bingeing on box sets of a popular television crime drama in an effort to learn how to identify the culprit in each case.
The hit US show CSI, or Crime Scene Investigation, began in 2000 and ran for 15 seasons. Now it has been used for a novel experiment.
Scientists at the University of Edinburgh chose the series for a new study aimed at teaching machines how to solve a problem – in this case fingering a fictional killer – by assimilating information from images, audio, dialogue and scene descriptions.
They taught the artificially intelligent machines to approach solving the crimes in the same way people would – by considering which characters might be responsible on the basis of their behaviour in previous scenes.
Information in various forms – spoken, visual and textual – was processed as the plot of each episode unfolded.
They say the results of the study suggest such devices could play a role in developing efficient algorithms for realworld tasks that require complex reasoning.
Dr Lea Frermann, from the University of Edinburgh’s School of Informatics, 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.”
The researchers set out to discover whether artificially intelligent computers can find the answers to puzzles that are challenging for humans.
They designed the computer model to solve arbitrary problems based on acquiring data.
The team mapped footage, script and background sounds from five seasons of the show into a machine-readable format and fed it into the computers, which learned to process the plot as each instalment
0 Researches used hit US TV show CSI to test artificial intelligence unfolded and continually revised the criminal’s identity.
The computers correctly identified the perpetrator during the final part of an episode 60 per cent of the time.
However, although the success rate for AI sleuths is impressive, itseemsthatthey are not quite ready to replace flesh-and-blood detectives.
People who watched the same shows were able to work out who was responsible for the crimes 85 per cent of the time, the study found.