AoC Annual Conference & Exhibition
I hear a lot of excitement about Deep Learning. Disappointingly, it has no connection with the moment of insight in a student’s eyes as they grasp a fundamental principle. It has nothing to do with the years of dedication to a subject that make craft skills effortless, and lay down foundations of factual knowledge for unconscious recall when required. It’s not even an intensive scuba-diving course.
Deep Learning, in this context, describes a particular type of computer program that finds its own optimum route to a desired goal. The goal you may have heard about recently is winning the ancient game of Go. Google Deepmind are very proud of AlphaGo Zero, which was given little more than the rules of the game and the goal of winning, and played against itself until it was very good indeed at winning Go.
This kind of Machine Learning program is called a Neural Network, because it’s based on the way an animal’s brain is thought to work. By trial and error, a network of components gradually finds the best way to get from the input data to the desired outcome. The engineers call it Reinforcement Learning, and it’s modelled on a behaviourist understanding of how animals learn.
Deep Learning programs are already better than humans at some things, such as playing games and sorting images. Just as machines now do most of the heavy lifting in physical jobs, machine learning will be doing a lot of routine mental tasks over the coming decade or two. AI, Artificial Intelligence, won’t take over the world in the Terminator sense, but it may take over a lot of the work currently done by humans.
It’s a challenge for educators hoping to equip their students for secure and rewarding employment. As manual jobs were decimated by machinery, we migrated to office-based work and service industries. If routine keyboard-pounding jobs can be done by AlphaGo Zero and its ilk, where will we go next?
Previous industrial revolutions saw obsolete industries replaced by new ones. Aviation, technologically impossible 150 years ago, now employs around 10 million people worldwide. There’s no guarantee new industries will arise, as that depends on economics and politics more than technology. But we certainly can’t rely on the old jobs still being there when tomorrow’s FE students enter the world of work.
I won’t disagree with those who say today’s young people need flexible skills and a working knowledge of technology. But the advent of Machine Learning, of Artificial Intelligence, brings new opportunities to think about what human learning is, and what it’s for.
I think it’s dangerous to equate what AlphaGo Zero, or any other advanced computer program does, with the way a human being learns. I said it was based on behaviourist models of how pigeons can be trained to peck a lever for food rewards, or rats to run through a maze. It’s a model of animals as little machines, responding mindlessly to stimuli. Even for animals, it’s not entirely accurate. For humans, it’s a joke.
Yes, computers can beat us at Go, at Chess, at arithmetic, and at spotting patterns in massive datasets. But they don’t know why they do it. They don’t know that they are doing it, in fact. Is this learning? I wouldn’t say so. And though it may be deep, it’s very narrow. Claims that machines are on the verge of out-thinking human beings tell us less about what machines can do, and more about the low esteem in which humans are often held.
In the real world, any FE student can out-think AlphaGo Zero. So, without undue flattery, it may be time to remind teenagers of what makes them unique: their human capacity to think, to learn, and to apply their skills consciously towards shaping the world of tomorrow they want to live in.
Written by Timandra Harkness, science writer, broadcaster and comedian
Timandra Harkness will be speaking on day two of this year’s AoC Annual Conference and Exhibition. Timandra has written for publications including the Telegraph, Guardian, Sunday Times, Evening Standard and WIRED. Her book Big Data: does size matter? gives a history of data collection and collation, how it’s changing the world, and its shortcomings from politics to health to smart cities. As well as looking at big data, she tackles AI and robotics, and considers topics around our relationship with science and technology.