Is automation really kicking goals?

There are quite a few international competitions where robots compete in roles usually filled by humans, such as various Robosoccer leagues, the Robocup Robot Rescue League, a national robotics competition in Malaysia and the World Robot Olympiad.

" [Robot soccer helps] point to the difficulties any future wave of automation may face."
Mark Lawson, Retired senior journalist

These are mostly for high school students and university AI (artificial intelligence) students exploring their fields rather than major companies using commercial robots and the focus is on improving robot skills.

But they do point to the difficulties any future wave of automation may face, despite fears often expressed in online forums and recent popular books about job-categories set to be swept away by a wave of automation.

Instead the competitions point to the reality of productivity figures: despite vast changes in social and business life including an iPhone in every hand, a laptop on every desk and apps for everything, the rate of increase due to technology has declined noticeably in recent decades.

{CF_IMAGE}

EMPHASIS

Of all the competitions, perhaps the premier one is Robocup Standard Platform League which started in 1999 but took on its present form when the league adopted the Nao robots in 2009. These are half metre-high humanoid figures produced by French company Aldebaran Robotics.

Because the robots are uniform the emphasis is on the software required to control them in five-a-side matches on a tiny indoor field, rather than the hardware used in a major part of other competitions.

The games can be found on Youtube and are entertaining, although not in the same way as people soccer.

The 2016 final was held in Leipzig, Germany between German team B-human and US team UT Austin Villa.

At one point the ball stopped on the line in front of UT Austin Villa’s goal with the combination of white line and black and white ball confusing the visual recognition systems in the robots.

A UT robot stood by the ball while others roamed around the field until one B-human robot worked out where the ball was and booted it in.

Vision systems and object recognition has proved a major problem for computers and robots. There have been great strides in the technology but in tasks such as recognising different objects on a conveyor belt, the human brain and eye remains far above an industrial robot.

Unlike people soccer, Australia is a power in Robosoccer thanks to the University of NSW Computer Science and Engineer Faculty team rUNSWift. This year’s cup campaign was marred by unfortunate glitches which resulted in a loss to the eventual runners up UT Austin Villa but the team won in 2015 in China and the year before.

Team member Jeremey Collette says one problem is the switch to a traditional black-and-white ball, as opposed to the previous year’s bright orange ball, which was hard for the visual system programmed by the team to pick out.

The team went to the Pre-robocup Asia Pacific Competition in Beijing in October and took second place to the UT Austin villa team in the orange ball league (there was a separate black-and-white ball league).

One major advantage held by rUNSWift was a balancing algorithm which let the robots run at a breathtaking 30 centimetres a second (1.2 km/h) without falling over (robots fall over a lot in Robosoccer.)

If the ball was in the Australian half, the robots would kick it and rely on their superior speed to run it down. Again, standing upright and running come naturally to humans but has proved immensely difficult fort robots.

That limits the ability of robots to do tasks outside of standardised factory conditions, such as carrying loads up a set of unfamiliar stairs which any human can do. 

FADING ADVANTAGE

Collette said rUNSWift’s balancing advantage was fading. One of the rules of Robosoccer is after the competition is over the teams have to release the code they have used to all the other teams, as part of the aim of improving overall skill.

As noted the robots do not have the processing capacity or advanced optical systems of an industrial robots.

They are somewhere above a high-end smart phone with two cameras, and that system has to be programmed to recognise the ball on the field, walk, kick, decide where the robot is on the field, act on the team strategy and decide whether the starting whistle has sounded (another surprisingly hard issue for robots).

Any Robocup game is then a minor miracle and the still clumsy on-field action a vast improvement over games in earlier years.

There are plenty of reports pointing to a wave of automation set to eliminate job categories. A working paper The Future of Employment by Oxford University academics Carl Benedikt Frey and Michael Osborne estimated 47 per cent of jobs in the US are at risk of being automated.

Jobs to be robotised, from the detailed analyses undertaken by the two academics, include truck driving and the bulk of administrative support functions.

Self-driving cars are certainly the topic of the moment and bookkeeping systems now routinely do a lot of the reconciliation work which took hours of office time in past decades.

Automation and artificial intelligence is also continually evolving, with one on the horizon called deep learning. As explained by Toby Walsh, a professor of AI at the University of NSW in an article on The Conversation, “deep learning uses a ‘deep’ neural network, loosely modelled on the human brain. It’s deep because it has half a dozen or so of layers.”

Those permit the neural network to pick out features. For example, in recognising images, the intermediate layers recognise features like edges and corners, Walsh said.​

But despite the continual evolution of AI techniques, the near universal use of computers and apps, economists routinely joke they can see the information revolution everywhere but in the productivity figures.

A book The rise and Fall of American Growth by distinguished economics professor Robert J Gordon contends the real change era was between 1870 and 1970 with various phases in that.

He contends nothing really beats the major transformations wrought by, say, the electrification of factories or the mass use of telephones. In offices, do the changes caused by laptops and iPhones really beat the drudge work automated away by the advent of typewriters and later photocopying machines?

Gordon calculates between 1920 and 1970 output per hour in the US increased an average of 2.82 per cent a year. But after 1970 the average increase was 1.62 per cent.

LESS SIGNIFICANT

Professor Gordon’s thesis that major changes wrought by IT systems in the past few years are less significant than previous advances remains controversial and productivity comparisons are difficult beasts. But productivity growth tracked by the Australian Bureau of Statistics (see chart) would suggest in Australia the decline began in the mid-1990s.

There are certainly skills which humans learn naturally which computers have a lot of trouble emulating. Scotland Yard, for example, has set up a ‘super recogniser’ task force, a unit staffed by people with an uncanny ability to recognise faces, even in grainy, poor-quality pictures and videos.

This ability is thought to be present in about 1 per cent of the population. Staff scan CTV images and pictures relevant to open cases looking for faces known to the police. The success of the unit has prompted police in other jurisdictions to set up their own.

Such promising advances aside, as technological changes has created and then mostly swept away whole professions, such as switch-board operators and typesetters, only the brave or foolhardy would forecast what will happen next.

But anyone watching a Robocup game, where the players charge around the field looking for a ball any child can see, may think Gordon has a point and technological change that actually affects productivity has slowed down.

Mark Lawson is a retired senior journalist