Last month, Numenta released a major new theory for intelligence and cortical computation. As we do with all of our research, the team documented the theory in a research paper and made it available on a pre-print server while kicking off the submission process with a peer-reviewed journal. However, we also did something we’ve never done before: we (Donna and Christy) created a “companion piece” to the research paper. What’s a companion piece and why did we write it? How did two non-neuroscientists write a paper about a neuroscience paper? In this blog post, we’ll take you behind the scenes to show you how and why it happened.
While Numenta has been publishing research papers for several years, this latest research paper is different. Instead of focusing on a particular topic, like sequence memory or sensorimotor inference, it introduces a framework for how the neocortex works. Of course within the framework are detailed proposals. But at the highest level, the “Thousand Brains Theory of Intelligence” offers a new way of thinking about the brain, with implications for understanding intelligence itself.
Given our sense of the breadth and potential implications of this advance, we wanted to make sure that the material was not limited to neuroscientists. We found it interesting, and we expected that others like us might, too. Essentially we wanted to write something that we would be able to read, and thus the idea for a companion paper was born.
Well, we quickly encountered our first problem. In order to write a companion piece, we first had to read and understand the research paper ourselves!
We read an early version of the research paper over the summer and from that, we created our first draft. We debated several points with each other along the way, seeing which one of us could fill in understanding for the other. It was our attempt to say, “Okay, here’s what we think the research paper is saying and why it’s important.” Our goal was to pull out the big ideas and present them in a way that made sense to us. We proposed examples to illustrate the difficult concepts. But we also needed to make sure that we weren’t saying anything incorrect. We reviewed multiple drafts with Jeff and Subutai, the research paper’s authors, who would tweak and massage our examples to make them more accurate. We then sent a draft of the companion piece to friends and colleagues to get feedback on what worked and what was unclear. We found the feedback to be extremely helpful and we integrated much of it into the current version.
Some people have asked whether they should read the companion paper in addition to the research paper. The simple answer is, read in whatever way is helpful to you. Each one is a standalone piece, so we expect most people will choose one or the other, depending on their comfort level in reading scientific papers. But it also depends on how much explanation you want. The companion paper introduces the ideas without explaining how they work or how the researchers arrived at them. In that way, some may find the companion paper to be a helpful primer for the research paper, allowing the reader to get a conceptual understanding of the “what” before diving into the “how.”
The companion paper is not a substitute for the research paper. It’s not a rewrite either. It does not follow along, section by section, translating each paragraph into non-neuroscience speak. It is our interpretation of the theory. We’ve added some details in the companion piece, such as some neuroscience basics, and deleted others, such as some of the neuroanatomy. Our hope is that after reading the companion piece, you will walk away with an understanding of Numenta’s “Thousand Brains Theory of Intelligence” and its implications for understanding the neocortex and intelligence.