In the speech, delivered in June this year, he revealed that when he became a spy, converting raw audio obtained from phone taps and bugging was “a painstaking task for hundreds of professionals sat wearing headphones”.
But he added: “Today, we want to automate as much as possible of that foundational conversion of audio into searchable text – freeing up our analysts to focus on extracting the intelligence insights that count.
“But the challenging nature of our audio data means that commercial speech-to-text solutions often can’t do what we need – at least not with the precision that high-stakes work rightly demands.
“So our data scientists build, train and deploy our own machine learning models, continually improving them based on real feedback – giving our people a huge productivity boost, enabling them to apply their analytical skills to the true secrets and mysteries.
“This interplay between maths, computing science, engineering and human expertise is a critical dynamic for us.
“Building a model is just the first step; the real test is using it on live operations. Applied maths at its sharpest.”
‘Another grateful nod to Alan Turing’
He then added that AI was being used to monitor whether potential terrorists or individuals on the path to violent radicalisation posed a real risk to society.
He said: “A second example of applying AI is in detecting violence in images. Understanding whether, say, a prolific contributor to extreme Right-wing online forums is also watching graphic beheading videos can help in assessing the level of risk they might pose.
“But we don’t need or want to view all the sport they’re also watching. So with another grateful nod to Alan Turing, we again turn to machine learning.
“We have put in place automated capabilities to detect violent material within large data streams.”
