Google’s Awesome AI work.

google-as-a-giant-robot

Whether you are a fan/user of Android, Apple, whatever… I’d bet we’d end up in agreement that, putting aside the merit of phones & devices for a moment, Google is more likely to usher us into an era of Artificial Intelligence than almost anyone else.

I mean honestly the Google Now service (while far from AI) is already so great.
At the time of this writing it beats Siri without even trying.

Here’s an example of coolness that happens all the time through Google Now – that would have seemed like magic not too long ago.

I recently booked a hotel room for a business trip.
I used my company’s travel site to do this.
Days later I got in my car and said “OK, Google. I need directions to my hotel.”
And BAM! – directions to my recently booked hotel popped up on my phone’s navigation app.

The ability to do that speaks to the reach google has across multiple systems more than it does some magical bit of AI code. Google Now heard my question and then searched my gmail inbox for the most recent hotel confirmation email.

Simple. Easy.
For it and for me. I didn’t have to do anything extra.
And it was fast.

It was pretty fraking great.
But it wasn’t true AI.

When it comes to winning the race to a truly intelligent system… Google simply has more assets at it’s disposal.

  • They have the knowledge graph.
  • They acquired Metaweb and it’s “database of things”
  • They have countless web queries allowing them to analyze how people think when they’re looking for something
  • They have countless web queries showing what people in certain regions/demographics/etc are interested in at any given moment.
  • They acquired DeepMind, a company devoted to artificial intelligence.
  • They acquired several robotics companies (Boston Dynamics and Schaft Inc. are a couple of the big ones)
  • And they have Ray Kurzweil running the show.

And I recently read an article about a new and potentially powerful asset… the Knowledge Vault.

Whereas the aforementioned  knowledge graph is a large and very valuable database of facts and answers – it’s a database that was curated if not populated entirely by humans. The Knowledge Vault though will not need any human to teach it… it will learn on it’s own.

Knowledge Vault automatically gathers information from across the Internet using bots and interprets the results to build a base of facts. It has already put together over a billion of them, with nearly 300 million considered “confident facts” with over a 90 percent chance of being true.

– source: AndroidPolice.com

It’s important to keep in mind these numbers really don’t add up to anything terribly intelligent. It’s only usually right 30% of the time.

Equally important; We aren’t sure how far off it is the 70% of the time it’s doesn’t even make it to usually right.
Is it close?
Is it  just fundamentally wrong, or is it taking as true all the crazy and wrong stuff it finds out there?
(like roughly half the humans on facebook.)

These are important considerations for sure.
But… we must also remember that it’s learning on it’s own, and it’s doing it with advantage of all the assets Google’s ecosystem can provide.

The folks at androidpolice.com consider what a system like this could mean for the Google Now service:

This could result in an improved Google Now experience. Instead of asking the service simple questions – such as, “What is the weather?” – you would theoretically be able to ask it something more complex, like “Why is the weather like this in my area?

Truly cool.
But beyond Google Now it has powerful implications.

Multi-part questions like that are key ways humans like Ray Kurzweil weed out “chat bots” that try to take on the turing test. Here’s a telling exchange between Ray and the chat bot “Eugene Goostman” that claimed to pass the turing test.

Ray Kurzweil: I live in the capital of the country that built the great wall.
Eugene: Tell me more about Capital. I like to learn about different places!

– source: Kurzeil.net

When we get to a place where multi-part questions like this can be answered by a computer system – on the fly – without having a hard coded, and human built fact matrix in place… things are going to get interesting real fast.

It may start with better search and truly capable digital assistants, but it won’t stop there. Great things will  start to happen. Scary things may also happen.
Best of all… Great Scary things will start to happen.

Leave a Reply

Your email address will not be published. Required fields are marked *