We know Google is big on AI research – it acquired Deepmind last year, has partnered with Oxford University and is packing intelligence into everything from self-driving cars to Google Now. But sometimes that can be a little creepy.
A research paper published this week, blandly titled ‘A Neural Conversational Model,’ and authored by Oriol Vinyals of the Google Brain team and Quoc V. Le, a Google researcher at Stanford, includes some unsettling human/computer interactions.
The project looked at conversational modeling (predicting what will come next in a dialogue), which is key to natural language understanding and machine intelligence.
It involved testing a model with a dataset made up of IT helpdesk questions and another, more complex one, comprised of excerpts from movie transcripts.
When the computer was providing IT support, it did a pretty good job [Note – grammar and spelling rendered as shown in the research paper]:
Machine: what is the error that you are running please
Human: i am seeing an error related to vpn
Machine: what is the error message that you are getting when connecting to vpn using network connect?
Human: connection refused or something like that
Machine: may i know the version of network connect you connect?
Human: i am not sure i know that
Machine: is the network connect prompting for any user certificate
Machine: are you at home or at work at the moment?
Human: i am at home
Machine: try this
Machine: goto page and enter your name and password
Machine: try logging in with and check if its working fine now
Human: yes, now it works !
But when the computer was tasked with responding to the movie dialogue, the conversations are unsettling in the extreme:
Human: what is the purpose of life?
Machine: to serve the greater good.
Human: what is the purpose of living?
Machine: to live forever.
Hello, Skynet. It gets worse when the conversation moves on to morality:
Human: what is moral?
Machine: what empowered humanity, what intellectual the essence is .
Human: what is immoral?
Machine: the fact that you have a child.
Thankfully, while the researchers “find it encouraging that the model can remember facts, understand contexts [and] perform common sense reasoning,” they admit their basic model provides “unsatisfying answers” and “does not capture a consistent personality.”
That should mean we’re still a long way from being lectured by replicants about “attack ships on fire off the shoulder of Orion.” But with Google also building terrifying robot dogs, we must remain ever vigilant.
Image credit: Bubbye on Imgur