‘Only movies build bad robots’: Google makes machine-learning software TensorFlow open source in bid to boost AI industry
Alphabet chief executive Schmidt downplays fears of helping to build robot race that may one day go rogue, says new tools help make people smarter
Google will make its machine-learning software TensorFlow available to the public as it hopes to accelerate the development of artificial intelligence, it said this week.
Machine learning is the core of AI. Computers can perform tasks like speech recognition, decision-making and other tasks related to visual perception.
A programme that uses machine learning can follow by example, and improves as more data is made available to it.
Google utilises machine learning in many of its services – including the automatic filtering of spam in Gmail and the search and classification features in Google Photos that automatically categorise pictures by location or people.
The reason for making one of Google’s core pieces of software open source and available to developers everywhere is to accelerate the evolution of AI, according to Alphabet executive chairman Eric Schmidt.
“If we get everyone using TensorFlow … we will get the benefit because it’ll get stronger and better … from more knowledge [and] discovery,” said Schmidt in a live video conference with reporters on Tuesday.
“I’ll say with some confidence that [TensorFlow] is so good that … probably our competitors will start to use it, and I’m very proud of this,” he added.
The US-based company also incorporates machine learning into its Inbox app, which not only automatically categorises e-mails but now also features a Smart Reply tool.
In this era where digital communication is the norm and users receive multiple emails daily, this tool suggests three short, tailored responses to an email based on the content, and aims to help users better handle their email load.
But don’t get too excited just yet. For now, the tool can only recognise short emails which require simple answers such as “yes”, “no” or “maybe”.
“We are just getting to the point where we can barely understand what a short email might be asking, but we don’t understand a whole conversation,” Greg Corrado, a senior research scientist at Google, told the South China Morning Post. He works on machine learning and co-founded the company’s deep learning team.
Corrado said that for this kind of software to understand natural language in longer pieces of text it would require “many research breakthroughs”.
“Understanding what users need when they’re asking something [in a query] is one of the areas that we will see a lot of growth and development … [hopefully] relatively soon,” said Corrado.
Google is not the only technology company eager to develop AI. Major technology companies like Microsoft, Facebook and IBM are also investing heavily in the field.
On Thursday, Microsoft made available an update to its suite of machine-learning services by introducing a new feature that reportedly identifies eight emotions based on facial recognition. These are, namely, anger, contempt, disgust, fear, happiness, “neutral”, sadness and surprise.
IBM’s researchers are also developing an AI computer that can accurately predict Beijing’s pollution levels 72 hours in advance. This would help Chinese authorities take measures such as shutting down factories early or capping the number of vehicles on the roads in a bid to mitigate anticipated smog levels.
For some, the advance of AI is a cause for concern. Could AI-enabled robots eventually turn against their human creators, just like Skynet in the Terminator movie series?
READ MORE: Hong Kong artificial intelligence pioneer on immortality, the age of robots and manufacturing evil
Google has been quietly snapping up robotics start-ups lately. It purchased nine robotics companies in six months in 2013 and several more that deal in AI over the following year, leading pundits to speculate whether it has future plans to combine its AI system with robots.
Tesla CEO Elon Musk has voiced concerns about AI robots potentially going rogue. In May, Musk was said to be worried that Google might “produce something evil by accident”.
But to Schmidt, there is still a long way to go before rogue robots become a remote possibility.
“Humans are not going to build bad robots, only movies are going to build them,” Schmidt said, adding that machine learning focusses on internet-based services and information that helps make users smarter.
“[Machine learning] has nothing to do with robots,” he said.
“The [number one] robot people want is a robot to clean the kitchen … to put dishes away and clean the counters and organise the flowers.”
“We are a long way from that kind of ability and intelligence,” said Schmidt.
“[That kitchen robot] turns out to be an incredibly difficult robot to build.”