It turns out that computer science wasn’t the original passion of Jeff
Hawkins, founder of Palm and Handspring. The creator of the first
commercially successful handheld computing devices really wanted to be a
neuroscientist — but only if he could do it his way. Hawkins wanted to
combine biology with computer theory, ideas that were too wacky, he
realized, to let him advance in a scientific career.
So he pursued two passions simultaneously, working in computer science and
putting out products that addressed real-world business needs while studying
theories of how the brain worked.
In 2002, he founded the Redwood Neuroscience Institute in Menlo Park,
Calif., to prove that his ideas on the basis of human intelligence could
produce intelligent machines. His new book, On Intelligence, released
this week by Times Books, explains his theory that
intelligence relies on the brain’s ability to make accurate predictions
about the world based on stored memories. This memory-prediction system
forms the basis of intelligence, perception, creativity and consciousness.
Hawkins took the time to discuss with internetnews.com the capacity
of human intelligence, the capacity of robotic intelligence, cortical function and
being a normal human being.
Q: When you talk about intelligent machines, you don’t necessarily
mean that they will be intelligent in a human way, do you?
Yes and no. I’m not talking about machines that look anything like a
human. We’re not talking about C3PO. On the other hand, they will be
intelligent, because they’ll use the same mechanism for building a model of
the world and understanding it as you and I do. They are using the same way of
being intelligent, but about a slightly different world.
Q: You have a great example in the book of an intelligent machine that
could think about and understand global weather systems the way we
understand objects and people. What would be the advantages of such a global
As humans, there are a lot of phenomena we can’t observe directly. There
are phenomena like weather that is really difficult for us to observe.
It’s only fairly recently that we discovered El Nino. It’s a real thing, we
just couldn’t observe it. I believe we can build machines that live in that
world and see those things, just like I see the trains and cars and
buildings out my window. The advantage is that we can discover things in the
world like El Nino, and we also start saying, “Now we understand why we have
all these hurricanes. And guess what, next year we’re going to have some
We’re so used to thinking about intelligence as a certain set of sensors
like sight and hearing and touch. But there’s no reason why these machines
will have the anywhere near same senses we do. Theyll have senses tuned for
their data world.
Q: You believe the most revolutionary use of your ideas will be in the
creation of machines with exotic senses — alien worlds of sensation. These
machines could employ senses that we don’t have, but how will we design
them, understand what they’re doing and interact with them?
Designing them is pretty easy. You can have two people, one blind and one
deaf, and they end up building the same representation of the world. They
believe in the same objects and perceive the world the same way. The
algorithm of the cortex is so incredibly flexible and generic, it says, “I
don’t care what sense you give me. As long as it senses the world somehow, I
will build a correct model.” The cortex builds a model of the world, not a
model of what it’s seeing. It will be fairly easy to say, “I want to build a
machine that looks at weather,” because you just have to hook it up to the
weather systems in the correct way, and it will build a representation of
The next part of the question is more interesting and a bit trickier.
How will we communicate with them? I haven’t given it a lot of thought. I’m
not worried about it in any sense. All we need to do is find the part of our
environment that is the same. It could be language; it could be graphical
representations. As humans, we can model all kinds of things: mathematical
languages, visual languages, spoken languages. We just have to have
intelligent machines share one of those with us. For example, written and
spoken languages are completely different, but we can associate one with the
It’s possible that what an intelligent machine discovers, it might not be
able to communicate with us. But I think most of the time it will be
problems and questions that we do relate to.
Q: In the nearer term, is there anything in your theories that could
be used to influence what we might call “traditional computing?”
A lot of this work has influenced my design of PDAs and smartphones. It
goes back to Graffiti in the early 90s and continues to today. Brains like predictability. People actually like
learning stuff. If they’re going to spend time learning something, they want
it to be predictable; they want it to be easy to model. They want it to have
a simple hierarchical structure.
Graffiti was a great example. I would rather give you something that, if it
didn’t work, you could know why it didn’t work and correct your behavior.
Traditional handwriting recognition didn’t do that. You’d write something,
and it wouldn’t work, and you didn’t know why.
It also impacted how we designed user interfaces. For example, on a
computer you have a bunch of dialogs. One rule I’ve always had is, you
shouldn’t have two ways of getting to the same dialog box. If you do, users
get confused. They say, “Wait a second, I’ve seen this someplace else
before, but I got here differently. How do I get out of here?”
Designers want to put in links all over the place; we’ll give you
shortcuts to everything. I don’t want to go to any two screens that look
similar from different places. There ought to be a concise and obvious path,
so I can learn what got me to this place and what didn’t, so people can tell
how they got to that place.
Q: You write that within 10 years we may see at least the beginnings
of an industry based in your ideas.
I’m being very cautious. I have a grad student at RNI [Redwood Neuroscience Institute]
who is building this stuff now. It’s demonstrable. It’s working. I know how to build this
stuff. I think 10 years from now this will be huge industry, going
gangbusters. It depends on how quickly I can people get started working on
it. It’s not fundamental work, it’s a lot of practical engineering work. In
fact, people can actually download the source
Q: You changed the face of computing by inventing the Palm. Yet it
sounds a bit like that was never what you really wanted to do. How did you manage to
achieve so much in the field of handheld computing while maintaining an
entirely different field of thought and expertise?
And be a normal human being and raise two teenage kids? Mostly, I had
really great partners. At Palm, although I was the original founder, I
quickly started searching for a CEO. A lot of people didn’t realize at Palm
and PalmOne that I was working part-time. I was there a lot, but doing other
stuff too. Having two major interests kept me fresh. I think it made me
better. It made my products better. On the other hand, it made me a better
scientist, because I could step outside of science and see why research got
stuck, why it was hard to get funding, why people were not working on this
Proposing a new theory of cortical function is really hard. Who’s going
to believe me? How do I sell an idea? Because all ideas have to be sold. The
two lives complement each other in many ways.
Q: What advice do you have for young entrepreneurs?
The rules that worked for me, I’m not sure they’ll work for other people.
But the most important thing is to have a passion for something. Any end
that’s significant is going to be hard. You’ll have setbacks, people
doubting you and telling you you’re wrong. Have passion, otherwise you’ll
never get through it. And stick with it. Don’t get hung up on anything else.
Don’t worry about making money or the trappings of success. You want to make
your passion successful.