August 12, 2016

Article at The Australian

MAGAZINE | March of the robots

Rethink Robotics’ Sawyer and Baxter, designed to work on a production line, learn on the job without programming and use human facial expressions

By Wilson da Silva

Don’t panic about machines getting smarter and behaving more like us – we’re going to need them.

RATHER THAN be worried robots will take our jobs in the next 20 years, we should be more afraid there won’t be enough robots. That’s the prediction made by preeminent roboticist and serial entrepreneur Rodney Brooks.

The pioneering Australian, founder of iRobot – the Boston-based billion-dollar company which has sold more than 14 million robots worldwide (including the Roomba vacuum cleaner and the Packbot bomb-disposal rover) – says the world is on the verge of a major shift that will see the world’s robot population explode into every facet of life.

It’s being driven by dramatic advances in artificial intelligence and robot vision, cheap but very fast computing, widespread wi-fi and a deluge of big data.

“I’m not worried about them taking jobs, I’m worried we’re not going to have enough smart robots to help us when we’re old and infirm,” says Brooks, who made his name as a maverick of robotics at the Massachusetts Institute of Technology. “We’ll all be tech-savvy and we’ll want to retain our independence … and we’ll want to stay in our homes longer.”

He points to the demographic bulge of ageing baby boomers who will live longer and need care, when there won’t be enough workers to care for them. The percentage of working-age adults in the US and Europe, which was stable at around 80 per cent for 50 years, is now rapidly falling: it’s forecast to reach 69 per cent in the US and 64 per cent in Europe by 2050. Even in China it will drop to 67 per cent by then.

To tackle this, we’ll need “social robots”, machines that interact with us using speech and social cues. Luckily, the technology is arriving at just the right time, and Brooks is ready to usher it in: in 2008, he co-founded Rethink Robotics with venture capital from Goldman Sachs and Amazon founder Jeff Bezos, among others.

Rethink has created machines that work alongside people and respond in ways we understand: like Baxter, which requires no programming, learns on the job and produces a quizzical look by raising its digital eyebrows if it doesn’t understand something.

While he isn’t yet up to the task of carrying groceries to your car or taking out the rubbish, it’s only a matter of time. Progress is fast, and the field has surprising goals (literally): like having robots play – and beat – the winners of the FIFA World Cup by 2050. Known as RoboCup, the annual robot championships – won five times by Australia’s University of NSW – attracts 3500 participants from 45 countries who compete across 16 categories, all aimed at advancing artificial intelligence and robotics and creating squads of android soccer champions.

Driverless 500-tonne tip trucks have moved 200 million tonnes of iron ore for Rio Tinto in the Pilbara

Even before then, AI will become ubiquitous. “Within a decade, AI will become the operating system of all our connected devices,” says Toby Walsh, a leading AI researcher at UNSW. “Apps like Siri and Cortana will morph into the way we interact with the connected world. It’ll be the way we interact with our cars, fridges, central heating system and front door. We will be living in an always-on world.”

Some of that future is already evident in the Port of Brisbane, where 27 giant robots – 10m tall and weighing 60 tonnes – scramble to and fro at lightning speed in what is the world’s first fully automated marine container terminal. The eight-wheel behemoths, known as AutoStrads, bestride shipping containers and move them around the terminal entirely autonomously, using radar and high-precision GPS. And they do so 24/7, placing containers into position with better than 2cm accuracy.

AutoStrads are the creation of Hugh Durrant-Whyte, who led much of their development at the Australian Centre for Field Robotics at Sydney University. Considered one of the world’s foremost innovators in field robotics, he was also behind the technology that helps mines move all their ore using robot trucks. The lumbering giants are now used by Rio Tinto, BHP and Fortescue Metals.

Helping robots see and understand their environment is the specialty of Peter Corke, director of the Australian Centre for Robotic Vision at the Queensland University of Technology. His team has created robotic systems for mining and aerial and underwater applications and is developing AgBots, or agricultural robots. AgBot II, a four-wheeled rover, has been seeding, weeding and fertilising crops in trials begun last year, and indications are that it can reduce the cost of energy, labour and chemicals by up to 40 per cent.

Another notable is Bob Williamson, chief scientist of Data61, the IT division of Australia’s national science agency, the CSIRO. He wields influence in machine learning, in which computers develop the ability to learn without programming, teaching themselves to grow and change as they are exposed to new data.

This is a hot area of research, propelling everything from self-driving cars to Netflix recommendations. And it has spawned stunning advances, such as the 4-1 victory by Google’s AlphaGo program over world champion Go player Lee Sedol last year. Go is an ancient, complex board game long thought beyond the reach of computers. But AlphaGo relied on a neural network that learned the game itself, training by studying past games and playing itself. It was created by DeepMind, a secretive artificial intelligence company Google acquired in 2009 for $US650 million.

Machine learning will have a huge impact on business and society, as it allows computers to dive into oceans of big data and find valuable patterns impossible for humans to discern. But there’s still “an awful lot of black art” to it, says Williamson: each algorithm is developed from scratch. “There’s very few standards. You can’t share anything, there’s always different systems.” His ambition is to create standards for machine learning that would allow it to be used across systems, accelerating advances by working off a common, but robust, base. “This is a real challenge for the community, but no one’s cracked it yet.”

When it comes to social robots, there is still a lot to learn. Mary-Anne Williams, an authority on “knowledge representation and reasoning” in AI, is director of the Magic Lab at the University of Technology Sydney, where she explores how robots can develop social intelligence in dealing with humans. The goal is to design autonomous technologies that can learn, adapt and collaborate with people.

“Robots of tomorrow will be social like Rosie of The Jetsons and C3PO in Star Wars,” says Williams. “They will understand human emotion and be able to interact with people, helping them be more productive and live happier, healthier lives.”

Brooks agrees, and he believes the fear that robots are going to take away all the jobs is overplayed. “Over time – and I’m talking decades – many jobs will be replaced by robots. But robots aren’t going to be good enough to do everything that people do, by any means.”