18 December 2018
Absolutely Integral: the human's role in machine learning

By Mark de Vries, Consultant

Are machines becoming genuinely intelligent beings, or are humans still fully in control?

AI playing chess

The robots are taking over – or so they say. Smart devices are seemingly everywhere we look, we’ve accepted artificially intelligent machines into our homes and workplaces and there are even self-driving cars on the roads. It’s starting to seem like it won’t be long until the world is entirely automated, and we humans are merely passengers going along for the ride.

We’re some way off that being the case, though. The benchmark is for a machine to pass the Turing test, and although it was claimed that the Eugene Goostman chatbot did so in 2014, many have refuted the claims saying that the test was weighted in the machine’s favour. The challenge in beating the 68-year-old test is for someone to talk simultaneously to a robot and a human, and if they are unable to identify which is which then the test has been passed.

It requires a machine to demonstrate true intelligence, to be able to actively make conversation, analyse responses and formulate a genuine answer. But we can’t forget that these robots are being created and programmed by humans, so is passing the Turing test a genuine possibility?

The first attempt

Rather amazingly, trying to pass machines off as being genuinely intelligent dates back to 1770 when Wolfgang von Kempelen tricked the world into thinking he had invented and constructed a robotic chess player. On debut in Austria, the so-called ‘Turk’ would typically win matches within 30 minutes, identify its opponent’s illegal moves and even communicate with players. The world was stunned as the Turk then went on a tour of Europe with its creator, which included matches against Benjamin Franklin and Napoleon Bonaparte.

So, does machine learning really date back almost 250 years? Of course not. It was later discovered – albeit nearly two centuries later – that the chess automaton was in fact operated by a man inside the cabinet, despite von Kempelen opening the doors to show audiences a seemingly empty interior. The very earliest attempt to demonstrate a machine’s intelligence relied on human input, and it’s the same to this very day.

We’re all involved

Since the Turk’s creation, machines that appear to be truly intelligent have relied on human input in order to calculate responses and make decisions; the difference now is that everyone is doing it, including you and me. We’ve all at some point been confronted with a reCaptcha authentication box on the internet, to which we’ve gladly obliged and entered the distorted word and blurry door number. The process may seem trivial, but the data that is procured from each and every response is powering the world’s machine learning.

Requiring internet users to decipher misprinted words has led to books as old as time being digitised, preserved for the ages. The idea behind the enhancements to the original Captcha system was that if users are carrying out tasks that computers find useful, they may as well provide meaningful information in the process. Google’s use of reCaptcha takes that one step further by making users identify features within images taken from Street View, creating data that is then used to programme autonomous vehicles. You can even make money from carrying out menial tasks that require human intelligence with Amazon’s Mechanical Turk, a crowd-sourcing platform launched by Jeff Bezos in 2005 that now gives organisations access to 200,000 workers.

The human’s role in machine learning is nothing short of vital, but if the Turing test is ever genuinely passed then who’s to say machines won’t start teaching other machines? It’s a scary thought, and perhaps an inevitability, but in a time when the headlines read “robots will take our jobs,” perhaps humans are more important than ever.