Edward Snowden May be the Last of the Human Spies

In future, the public may never be alerted to NSA-type revelations because surveillance is fast becoming automated

By Christopher Steiner

Kurt Vonnegut once opined: "Human beings are chimpanzees who get crazy drunk on power." That power corrupts is hardly debatable. For that reason, the evolution of espionage has run in parallel with the development of organised tribes of human beings that we now refer to as countries.

Human nature makes it predictable that organisations such as the NSA would be cataloguing phone calls and other electronic interactions between humans. But Edward Snowden's revelations also tell us how far electronic snooping has yet to go. While the din of outrage still resonates, we should be thankful that Snowden – a human being – actually exists. In the future, the world may never be alerted to such breaches of privacy because there will be no humans involved in spying at all. Just as algorithms have conquered our stock markets and our musical tastes, so too will they conquer surveillance. Even the most human of tasks, snooping, will become the province of the bots.

While it's true that the surveillance Snowden spotlighted is of a new and digital variety, it still required human levers to give it any meaning. The NSA, for example, using its call log data, would take an interest in people who repeatedly dialled the phone numbers of known troublemakers. Human agents would query the call-logging database and find out who a prime target in Yemen might be speaking with inside the US. The data is collected passively and electronically, but much of the intelligence and the methods to derive it come straight from human minds. But what will happen when a machine makes the rules?

In the late 1940s, Vonnegut observed how General Electric was replacing human machinists with computer-operated milling machines to cut rotors for jet engines. This passing of duties from humans to bots led Vonnegut to imagine a world where human chores of all manners would cease being the labour of men and become strictly the work of machines. Power and income, then, would be concentrated among the few who controlled the machines. Snowden and the teams of analysts at the NSA, CIA and GCHQ who sit in front of our stores of electronic intelligence will hardly be necessary in 15 years. Algorithms will have replaced them, leaving only a few humans, like General Keith Alexander of the NSA, left to watch the house.

Underneath those top humans will be machine-learning algorithms that dance across the data of humanity like a spider tending a web. They won't be programmed simply to search for call patterns or numbers; they will learn what patterns and numbers are significant by ingesting news, conflicts and terrorist threats in real time, comparing that to activity seen on computer and phone networks. Algorithms that trade stocks at the speed of light already read specially tailored news feeds from Bloomberg and Reuters; the intelligence world, although less lucrative than that of Wall Street and the City of London, will not be far behind.

Algorithms are more efficient than people; they can find relationships within data streams that a human eye couldn't spot in 20 years; they're indefatigable – and they're cheap. Also on the positive side, algorithms aren't much for drama, counter-espionage or leaking. They do their jobs and don't ask questions. But they can make mistakes that border on inexplicable. Just as an algorithm belonging to Knight Capital in 2012 went berserk and lost that firm $440m (£288m) in 45 minutes, an NSA algorithm could finger thousands of innocent people to be targeted for extra surveillance, or worse.

But these things can and do work in what would seem to be incongruous arenas. The CIA has been using algorithms that run on a thread of mathematics called game theory for more than two decades. The man behind these strings of reason and mathematics, Bruce Bueno de Mesquita, a political science professor at New York University, says that analyses driven strictly by human observation are flawed by their very nature. Human analysts, he points out, have appetites for meaningless information such as personal gossip, backstories and tales of failure and conquest. Algorithms couldn't care less about these things, of course – a fact that helps them do their job better than humans. A CIA study found that Bueno de Mesquita's algorithms were right twice as often as its own analysts in making predictions about future intelligence events. The study spanned more than 1,700 predictions made by the algorithms – a task the bots dutifully performed without billing even one hour of overtime.