These days it may seem like Google has an answer for practically everything but there are some questions it still can’t answer.
Ask it “what is 1,374,294 x 384,293,999?” and it will give you an answer within a matter of milliseconds but ask it “will it be dark when I get home tonight?” and it won’t be able to tell you.
Emmanuel Mogenet, head of Google Research Europe, explained at the 3rd Research and Applied AI Summit (RAAIS) in London on Friday that he and his team of 130 Googlers in Zurich, Switzerland, are desperately trying to change this.
Referring to the “will it be dark when I get home tonight?” question, Mogenet said: “The answer is not on the internet anywhere to be found. The sad thing is we have all the data to answer this question. We usually know where the user is. We usually, if the user lets us know, know where they live. We know what time the sun sets at every point on the planet. We can predict movement of the user if he’s on the road for example. So all the pieces of the puzzle are stored somewhere on one of our hard drives.”
He added: “But coming up with the answer is not something we’re capable of because we cannot get to the semantic meaning of this question. This is what we would like to crack.”
Mogenet said that a breakthrough in this space would benefit Google for obvious reasons. “We’re in the search business. We’re in the assistant business. Being able to answer that kind of question for our users would be extraordinary.”
One option, Mogenet said, is to “build a gigantic database of everything there is to know about the world. Petabytes and petabytes of facts.” But Google doesn’t believe in this approach because it’s impossible to scale due to all the various analogies there are in the world and the fact that they can all be combined with each other.
Instead, he and his team are using a range of artificial intelligence techniques to try and get there, adding that it boils down to Google being able to understand natural language.
“We want to solve common sense to the point where a computer would look at a [do not cross] sign and say ‘yeah, of course, why would you do that'”.
He explained that Google needs to try and build a model of the world so that computers know things like “when there is a 747 there’s not going to be a giraffe on top of it.”
One way to go about building that model of the world is to look at all the images and videos on the internet. “We have massive amounts of images of the world available for free on the internet,” said Mogenet. “The amount of images available on internet are two to three orders larger than every image you will ever see in your life.”
He explained that a model could be created by “pumping” an image dataset through a computer vision stack that was capable of annotating it and doing statistical analysis on it. “Building a computer vision stack that’s powerful enough to splice the world in its components and hopefully their ‘relationship’ is something that will get us to a very basic semantic understanding,” said Mogenet. “We need lots of data, but we do have this.”
In the short term, Mogenet said that his group’s research will be used to support new features in Google’s virtual personal assistant. In the long term, Mogenet thinks that his group’s work could pave the way towards general AI, which is when machine intelligence reaches parity with human intelligence.
“I’ll be honest with you, I believe that solving language is equivalent to solving general artificial intelligence. I don’t think one goes without the other. But it’s a different angle of attack. I think we’re going to push towards general AI from a different direction.”
Get the latest Google stock price here.