Luzius Brodbeck has developed a smart algorithm for a cheap, intelligent robot that can teach itself the task of building smaller “babies” made from plastic cubes.
He has taken something computers do very well – simple manual tasks, performed repetitively – and added an intriguing twist. “If the robot carries out the routine of building a ‘baby’ robot 10 times in a row, each result will be different,” said Brodbeck, a PhD researcher at the Institute of Robotics and Intelligent Systems at ETH Zurich.
The robot is given a set of between one and five cubes to glue together. It is instructed by an algorithm which tells the robot how to move its arm and where to connect the cubes.
But these limited parameters can randomly change. The robot evaluates its own performance and tweaks its creations, independently of any human instruction.
The results of this fine-tuning are constantly different configurations. Brodbeck, who worked under Cambridge University researcher Fumiya Iida on the collaborative project and published his results in the journal PLOS One, described the process with a genetics analogy, saying “The robot passes on desirable traits to its ‘children’ while breeding out undesirable ones.” Often times the robot returned designs that a human could not make, he added.
It is not hard to imagine a day where scaled-up versions of Brodbeck’s robot will work on car assembly lines, progressively finding little defects and fixing them as they go.
But although it only took something like a year and a half to build, the effort to create a more sophisticated bot is not trivial, he said. “Robots are very good at repetitive tasks. They’re precise and reliable. But there’s still a very long way for robots to go to be truly creative.”
Artificial intelligence threat
Brodbeck says he is ready to leave academia to find a job with industry.
He will have plenty of options. Google, Facebook, Microsoft, Amazon and Baidu, to name a few, are on the hunt for artificial intelligence researchers, snaffling up start-ups, and putting hundreds of millions of euros into the race for better algorithms and smarter robots.
This new arms-race does not just have obscure academics fretting over a future where robots steal blue-collar jobs and generally run amok. Serial entrepreneur Elon Musk and physicist Stephen Hawking are two prominent voices to have recently expressed unease about artificial intelligence.
Last year Musk described artificial intelligence as “summoning the demon”, and said any money he may put into research in the field is less an investment than a way to keep a watchful eye on progress of the technology.
When Cambridge University uploaded a video of its robot creation to Youtube with the playful caption, “On the origin of (robot) species”, it prompted one commentator to liken it to watching the first Terminator being born.
Such concerns are greeted with a small shrug by Brodbeck. They are all conversations that work “very well as plots for movies, I suppose.”As he sees it, we are hardly on the precipice of apocalyptic doom. “I’m not personally worried robots will take over any time soon,” he said. “If you see the work that goes into making them, you’ll quickly realise how difficult it is, and feel less worried.”