CIA using deep learning neural networks to predict social unrest five days before it happens

The Central Intelligence Agency (CIA) has upgraded its approach to surveillance by focusing on a new "technology-first" strategy that sees it using deep learning, neural networks to scan big data in order to predict when and where trouble is likely to occur in the US.

Neural networks are large networks of traditional computers that are trained using algorithms to solve complex problems, thus making them artificially intelligent beyond their individual capabilities. They work in a similar way to the human nervous system. Different layers of the neural network examine different parts of the problem and combine their results to produce an answer.

In October 2015, the CIA opened the Directorate for Digital Innovation in order to "accelerate the infusion of advanced digital and cyber capabilities" – the first new directorate to be created by the government agency since 1963.

Looking back over the year, deputy director for digital innovation Andrew Hallman told the Next Tech event in Washington DC on 4 October the agency has boosted its "anticipatory intelligence" using sophisticated algorithms to shift through huge "open data sets" in order to find patterns between seemingly unconnected information – both historical and current.

The term 'open data sets' refers to any information that is freely accessible on the internet, so this can mean anything from open posts on social networks and newspaper articles, to comments left by users on articles, forums and social bookmarking websites, as well as historical data from archived reports.

"We have, in some instances, been able to improve our forecast to the point of being able to anticipate the development of social unrest and societal instability, some I think as near as three to five days out," Hallman said, according to Nextgov, which co-hosted the event together with trade magazine Government Executive.

"What we're trying to do within a unit of my directorate is leverage what we know from social sciences on the development of instability, coups and financial instability, and take what we know from the past six or seven decades and leverage what is becoming the instrumentation of the globe."

In fact, over the summer of 2016, the CIA found the intelligence provided by the neural networks was so useful that it provided the agency with a "tremendous advantage" when dealing with situations, which, if you think about it, probably relates to the various protests in several US states over police brutality.

Hallman said that in the past, US intelligence agencies have faced resistance from policymakers over analytics provided by computers, and that there was a clear preference for intelligence gained from traditional spycraft.

However, since starting the new directorate, he says that intelligence analysts are getting better at understanding which data the neural networks can produce and how to explain their observations to policymakers in a way that helps them make important decisions.