An interview with Jeff Hawkins, the founder of Redwood Neuroscience Institute & Numenta Inc., on his career path and research towards Thousand Brains Theory of Intelligence

edited by Ayaka Ando


Jeff Hawkins is a BSc graduate from electrical engineering, Cornell University, 1979. Since then, Jeff has developed an impressive career in mobile computing in the Silicon Valley, where he established Palm and Handspring, two mobile computing companies. However, his thoughts have always circled around science. Eventually, in 2002, Jeff founded the Redwood Neuroscience Institute in Menlo Park, California. Three years later, in 2005, he created Numenta, a company based in Redwood City, California. At Numenta, Jeff and his workers are conducting independent research in the field of computational neuroscience, attempting to reverse-engineer the neocortex. The vision is that learning about the mechanisms of computation in the neocortex will eventually lead to creation of new, more functional machines.

On October 15th, Jeff gave a keynote lecture at the Open Day of the prestigious Human Brain Project summit in Maastricht, where he introduced his theory of grid cells in neocortex (a.k.a. a Thousand Brains Theory of Intelligence) as the framework for cortical computation. This theory posits by that cortical columns learn the structure of objects by pairing location signals with sensory input over time. According to this theory, every column of the neocortex develops a complete model of objects in parallel (hence the nickname).

Even though today Jeff is a successful researcher and entrepreneur, as a visionary with ideas ahead of his times he encountered a range of bottlenecks on the way to where he is now. Being a pioneer is not always easy. For instance, Jeff’s PhD proposal to perform pattern recognition was rejected at University of California, Berkeley - only because there was no one who would be capable of supervising such a thesis at that time. He also attempted to start a new research department at one of his employers, Intel, but he was rejected. However, since he started the Redwood Neuroscience Institute and  established Numenta, Jeff has been developing strong collaborations with academia, giving invited scientific talks at dozens of universities and institutes. His independent research has gained steam in the recent years.

 In this interview, we will find out what keeps Jeff going, and how he managed to stay positive over all these years. We will also find out about Jeff’s recent research on cortical computation.

Jeff, thank you for accepting the interview! Especially given that you must be extremely busy these days, just a few days after announcing your new theory of neocortical computation. One question that first comes to my mind when learning about your personal story is: how did you manage to stay enthusiastic about your research plans for all these years spent in industry?

Recall that I started my career in the computer industry and then left a good paying job to become a graduate student at UC Berkeley to study brains. I was, and remain, extremely motivated to understand how the brain works. I went back to industry reluctantly, and I always viewed it as temporary. My enthusiasm for neuroscience never waned because understanding the brain is the most important scientific question of all time. I felt I could make a difference in brain theory and that nothing else I did in my professional life could be as important.

Setting your own private institute must be associated with extreme levels of perseverance. When did this idea first come to your mind? Did you have any guidance or role models, in the process of creating this institution?

It was easier than you imagine. The idea came from several neuroscientist friends of my mine who said the field of neuroscience needs cortical theory. They encouraged me to start an institute. I agreed only on the condition that they help me, and they did. There were a number of scientists who, like me, wanted to work on cortical theory, so they signed up.

When was the first moment when you felt accomplished as a researcher? Is this only now, after you proposed your theory, or did this happen a long time ago?

I measure my success by theoretical advances. The first big success was when I understood that the function of the neocortex is to learn a predictive model of the world. I came to this realization in 1986 and at that time this was not something many people were talking about. Knowing this gave me an approach to studying the neocortex by asking how neurons can make predictions of future events based on a learned model. My book “On Intelligence” was based on this idea.

The second big advance occurred about seven years ago. This is described in our paper titled “Why Neurons Have Thousands of Synapses, A Theory of Sequence Memory in the Neocortex”. This paper introduced several important new ideas, including a theoretical explanation for the existence of dendritic spikes, the functional role of mini-columns, and how the neocortex represents information use sparse distributed codes. The first paper on this was published in 2016 in Frontiers Neural Circuits and quickly become the number one viewed paper in that journal.

Our latest manuscript, “A Framework for Intelligence and Cortical Function based on Grid Cells in the Neocortex” is by far the most important advance. It builds on all of our previous work, and as the title suggests it describes an overall framework for understanding the neocortex. It predicts several novel and testable concepts.

What do you think is the best thing the members of IT industry can learn from researchers, and what is the best thing researchers can learn from the members of IT industry? Or perhaps, you believe that the boundaries between the two worlds no longer exist?

Many experimental labs need computer scientists to manage their data. In my world of theory we are all computer savvy and programming skills are needed. I have also benefited from general business skills, managing people, communications skills, financial skills, etc. However, the world of science and the world of business are very different. I have not found it possible to do both at the same time.

How do you keep your mind so fresh and positive? Most researchers who finished their higher education three or four decades ago, slowly start thinking about retirement. While in your case, it seems that all the fun is starting only now. Did you discover a fountain of youth?

Thank you. I have always been optimistic and upbeat. This is partly my personality, but I have also chosen projects that I believe to be very important for humanity. When I was designing mobile computers I knew that they would be the center of personal computing in the future and therefore it was important work. Studying the neocortex is similarly important. I have a very long term view of my life and have had the luxury of working on difficult but fascinating problems.

Can you tell us more about the ultimate goal of Numenta? Do you aim to also take part in implementing the algorithms mimicking the neocortical computation, into a new generation of machines? What will be the advantage of this new, better AI for us humans?

Numenta’s number one goal is to understand how the biological tissue of the neocortex works. This is a purely scientific goal. We have a secondary goal which is to apply brain theory to the future of AI and robotics. I believe this is inevitable, our work can accelerate the transition to real AI.

Understanding how the brain works and building machines that work on those principles will be the most important transition of humanity. Just to give one example. Elon Musk and Jeff Bezos want humans to live on other planets. This will not be possible until we have truly intelligent agents that can proceed us and make those environments habitable for us. Robotic engineers and laborers capable of this will have to be based on brain principles. Today’s AI and robotics are still in the stone age.

What are our criteria for finding workers at Numenta? How would you describe your dream worker?

They must love the mission and what we are doing. They have to be capable of learning a lot of neuroscience (if they don’t already know it). Programming skills are a prerequisite. And they have to be able to think theoretically.

Thank you for the interview. You are a role model to a lot of young people in academia, and we wish you all the best in your research life.

Natalia Bielczyk