Jeff Hawkins: Where open source and machine learning meet big data

The Palm pioneer has turned to neuroscience and big data to create a path to truly intelligent machines -- a path open to the community's contributions

At OSCON in Portland, Ore., last month, I had the chance to meet Jeff Hawkins, the inventor of the Palm Pilot and arguably the father of the smartphone. I learned that he is now pioneering the analysis of huge streams of real-time data using insights gained as a neuroscientist. His company offers a product that can learn the characteristics of data streams, predict their future actions, and identifiy anomalies.

He has just recently taken the core of that product and released it as a GPLv3-licensed open source project on GitHub so that anyone can build machine intelligence into their systems. Below is a video of our discussion, followed by an edited version of the interview.

[ Track the latest trends in open source with InfoWorld's Technology: Open Source newsletter. ]

You don't do Palm Pilots any more. Tell me about your true passion.

My true passion, and it has been for over 30 years, is neuroscience. Understanding how the human brain works and building machines that work on those same principles. The whole period of Palm and Handspring [a company Hawkins co-founded to create Palm-compatible devices] was like a sideshow. I couldn't get the gig I wanted in neuroscience, so I was building mobile computing. I loved it, I enjoyed it, I was totally excited about it. But my real passion was brains, how they work, and building machines that work on the same principles.

So now you've been working most recently on software that can predict the future?

That's right, we've actually been working on modeling, figuring out how parts of the neocortex works -- which is the big region on top of your head -- and we figured out first what it does and then how it does it. But we've been applying it to problems where you can take streams of data, model the data, and then predict the future. We have a product called Grok that takes data from windmills, energy meters, and other machines and can predict future values. It can detect when anomalies are occurring and things like that.

We have a business side to what we're doing, but we're here at OSCON to talk about the new open source project where we take these learning algorithms, which are essentially models of the neocortex, a slice of the neocortex, and put them in an open source project.

It's been up for a couple of months now. It's called NuPIC, the Numenta Platform for Intelligent Computing. It's the same code tree we use in our products, so you can go and see what we're releasing every day. We have an active community already -- we've had our first hackathon. It's only been going for a couple of months, but it's been going pretty well so far, we're pretty excited about it.

What would I use that for as an open source hacker?

There are a number of reasons why you might be interested in this. People are interested in applying it to new problems. We're applying it to machine-generated data, but there's lots of other applications where people might apply it to.

In the same vein as we're using Grok, you might apply it to financial data, or you might apply it to new sources of data that we don't look at. There's a lot of people interested in taking it and building more complex systems -- robotics, vision, music, things like that -- and this requires extending the algorithms. It's a memory system, so making them bigger, putting them in a hierarchy, and so on.

There're people who want to apply it to existing products -- "I want to predict something" -- people who want to build new types of products, people who are interested in language, semantics, and so on. Then there're people who are interested in doing pure research. They're doing mathematical analysis of these algorithms, so it's kind of broad.

We're talking about the beginning of building brains in software and hardware, and that's a big, big field. It's going to be huge, and it's just getting started.

And I should mention we have a number of people interested in doing hardware implementations of these algorithms. There are some big companies -- IBM, Seagate, some others that actually have programs on the way because they're pretty excited about this stuff.

1 2 Page
Recommended
Join the discussion
Be the first to comment on this article. Our Commenting Policies