Most of us will have been happy this past week to see rising pressure from Elon Musk and others to the UN to ensure that we never make Terminator a reality by weaponising artificial intelligence. Personally, If Microsoft’s Twitter bot has had any impact I think we should keep it away from the internet too!
Have no doubt, robots are already here. In 1996, Deep Blue was able to beat a grandmaster chess champion by thinking further ahead than could be possible from a human. Just last year, Google’s Alpha Go beat the world champion at Go. The reason behind the 20 year gap comes down to Moore’s law and a little bit of science fiction.
In 1965, Gordon Moore predicted that computing power would continue to increase two-fold roughly every two years. Since that point it has proven its accuracy. A clear example can be found in memory cards which have rapidly progressed from Kilobytes to Gigabytes over the last 20 years.
In chess, there can be an average of 35 possible moves per turn, in Go this increases to around 250. This means in contrast, thinking 5 moves ahead would take over 18,000 times more processing power for Go than for Chess. At 20 moves, this increases to nearly 120 Quadrillion times. This meant that to have any kind of foresight would take massive amounts of computational power, but even with Moore’s law, the same Brute-force method used for chess would be unattainable in Go for decades.
A little bit of Sci-Fi
Aside from exponential computing power, we have also started to use that computing power to develop machine learning algorithms. By setting objectives for programs, we are essentially able to write programs that write themselves. For Alpha Go, this meant playing itself to narrow down it’s options to the most appropriate moves in each situation – in effect, it programmed itself with the required experience to beat grandmasters.
This is just one example, Google has been training their machine learning algorithms to deal with a range of scenarios based on set parameters. This objective rather than hard-coded approach to robotics means that the applications are limitless. Combine this with the development of neural networks and soon we will be looking at robots which can identify their environments and act accordingly, acting in an office as self-driving cars do on the road.
In terms of real world applications, Boston Dynamics is as close as they come. Their quadrupedal robots have proven themselves durable enough to work in the field capable of overcoming both rough terrain and a swift boot to the side. More recent developments are capable of lifting and jumping unaided while balancing on two wheels.
Other developments within robotics have seen robots which can work as a hive-mind to solve problems, spider-bots which can learn to walk again after having limbs disabled and robots which can demonstrate “hand-eye” co-ordination superior to humans.
Broad technological developments in this manner have always generated a wave of innovation across the board. The economist Nikolai Kondratieff recognised similar trends with the inventions of hydro power, steam and then electricity.
Schumpeter later discovered that these innovation shifts were actually increasing in frequency. Arguably as each wave precipitated the development of the next through increases in idea transmission and resources.
Industrial revolutions have always displaced labour as soon it has become economical to do so. Windmills replaced human power; looms replaced human dexterity; cars replaced Horses; electronics displaced elevator operators and vast swathes of clerical staff have been made redundant by data processing software.
This next development then, the start of technology capable of making rational judgement, surely spells the end for work as we know it. As with all technologies, the more it is used, the better it is funded, the greater the growth and the more applications will open up. As more variables can be understood, quantified and compensated for, artificial intelligence and its infinite simulations of logic may yet prove to surpass our own abilities.
How long then until a machine is capable of picking and packing an order in a warehouse scenario? How far are we really from completely automated marketing, which knows more about our habits and preferences than would be economical for a salesperson to know? How far are we from a machine that can supervise other machines?
It is certain that the next few decades will foresee drastic shifts in how we work. The allocation not just of resources but of human labour will surely pose the next big questions for economists of the future. It is likely that social skills will prove the last hurdle for machine learning as we emerge from the uncanny valley; until then, it can’t hurt to know how likely you are to be replaced in the next 20 years.