With the deluge of reports lately on how robots are replacing humans at everyday tasks, you may think the end of the world, at least for a certain category of human workers, is nigh. Indeed that may well be the case in certain professions.
The Associated Press, for instance, has a team of robots that generates 3,000 news reports about the quarterly earnings releases of companies. This Japanese hotel that I wrote about recently is staffed entirely with robots. The paralegal may soon be a position of the past as law firms commission bots to sift through vast reams of legal information and synthesise them.
It comes as delightfully refreshing news that our machine friends, despite all the hype, actually suck at being humans. This enlightening report from Quartz details how the machines from the world's most celebrated annual global robot competition, orchestrated by DARPA, the Pentagon's research division, were destined to rule the world couldn't do the simplest tasks without keeling over like hopeless winos.
Inspired by the Fukushima nuclear meltdown in 2011, the DARPA event invites elite robot-building teams from all over the world to strut their stuff, hoping that it will collectively inspire a new breed of machines that can work in environments where humans can't such as nuclear disasters.
Many of these DARPA robots have in the past been celebrities for their awe inspiring skills. The Sand Flea can jump 30-feet. The Cheetah can apparently reach speeds of up to 28mph, and easily chase down super-sprinter and Olympic champion Usain Bolt. The LS3 can saunter off with 400lbs of equipment. DARPA's very own champion, appropriately named Atlas, can apparently drive a car, smash through walls, and scale buildings.
Well, that's all fine and dandy but there seems to be a massive devolution in the robotic line (or were these robot champions of yesteryear over-hyped?) because the latest DARPA competition which fielded the best-of-the-best apparently had participants that keeled over at any given moment: They tipped out of cars and down stairs, collapsed while turning knobs, and sometimes just stood staring at a problem for hours -- not exactly the kind of troops you want replacing you in the event of a natural disaster. The winning team fielded a champion that took 45 minutes to do something that a human could polish off in a just a few. So, for now, we humans can breathe a little easier.
Or can we?
In what may prove to be the most unholiest of alliances, an article in The Atlantic reveals how robot experts and developmental psychologists at the University of Washington recently got together to try and figure out how to make these machines more efficient, and the result was a startling revelation -- design them to imitate babies.
"The secret sauce of babies is that they are born immature with a great gift to learn flexibly from observation and imitation," said Andrew Meltzoff in the Atlantic piece. Meltzoff is a psychology professor at the University of Washington and a co-director of the school's Institute for Learning and Brain Sciences who collaborated with the robotic engineers.
"They see another person and register that the person is 'Like Me.' They devote great attention to the 'Like Me' entities in the world...Roboticists have a lot to learn from babies," he added.
In other words, when you program a machine to follow a specific pattern to move this way or that, in precise steps and movements, the machine can do so just fine -- but is completely unprepared for the unpredictability of real-life situations. Babies, on the other hand, represent the best example of the adaptability through imitation. They are sponges; they learn complex things by watching, imitating, making mistakes, and doing it again and again in different, new ways. Robots, on the other hand, have to be spoon-fed all of this and even then it clearly doesn't work well.
Meltzoff and his interdisciplinary team devised algorithms that allowed machines to do what they are normally programmed not to do: To infer what the intention is behind a goal or an action, and to explore achieving the same result using a different approach; and to see how its actions could affect disparate outcomes. In experiments run to test out their efforts, it turned out the robots in fact did learn how to imitate humans via observation when it came to head movements and moving objects on a table with their hands.
Today, robots are still far behind the developmental ability of even babies in their ability to absorb and learn tasks through a process of trial and error. However, when sensors become more powerful and computational power becomes more robust, as the Atlantic article suggests, the machine learning process could well become very human like.
And that's probably a day that humans will celebrate and rue in equal measure.