This time each year we create a plan for the coming fiscal year, ending each June. The plan includes how we allocate our time and budget. This year (fiscal year ending June 30, 2018) we are shifting our priorities more than most years, and I thought I would share those changes.
As some of you know, in the beginning of 2016 we had a major insight/discovery related to brain theory. I sometimes say we discovered “the missing ingredient” of how the neocortex works. If our insight is correct, and I believe it is, it will have a major impact on how neuroscientists think about the neocortex. Briefly, the standard view of how the neocortex works is that sensory data is processed by a hierarchy of neocortical regions, each region extracting more complex features than the last region, until objects can be recognized. We discovered that the standard view is only half the story. The other half, the missing ingredient, is that each region also determines where sensed features are “relative to the object being sensed.” For example, when you touch the lip of a coffee cup, the neocortex knows both what the finger is feeling and where that sensed feature is located relative to the cup.
The implications of this discovery are profound. It explains how the neocortex learns the 3D structure of the world, and it tells us that each neocortical region, indeed each cortical column, is far more powerful than is currently thought. This understanding will also enable us to build machines that learn about their world the way our brains learn. The brain learns via movement. We move our fingers over objects and we move our eyes to look at different parts of an object. Our discovery tells us these movements are essential to learning, and they allow the brain to model the structure of objects in the world as they are in their entirety, not simply as they appear on our senses.
We recently submitted our first manuscript related to this discovery. You can read a pre-publication version here. We have several more manuscripts planned, and we will continue to present this research at various scientific and academic conferences through talks, workshops and posters.
Okay, so what are the implications of this discovery for Numenta’s coming year? Our discovery allows us to accelerate our work on brain theory, and so we have decided to shift most of our resources into research for the coming year. During this time, we will reduce application development and licensing efforts. Instead, we will invest more in research, through our internal efforts, our publications, and our outreach to other researchers and scientists. You may notice these changes reflected in our website and other communications. Matt Taylor, who heads our NuPIC open source community, will separately communicate any impacts there.
Numenta has always had two goals: reverse engineer the neocortex and apply that knowledge to the creation of machine intelligence. Our goals are clear, even though the path to achieve these goals is never as clear as we would like. Last year’s discovery is a big advance and we feel we must follow where it is leading us, at least for the coming year. We will share our progress as we go. We are tremendously excited to continue this work and appreciate your continued interest.