In this meetup, neuroscientist and entrepreneur Jeff Hawkins, co-founder of Numenta, discussed his new book A Thousand Brains: A New Theory of Intelligence. This is a fireside chat between Numenta’s researcher Lucas Souza and Jeff, where they covered the main aspects of the theory, what it represents for neuroscience and machine learning and how can we incorporate these breakthrough ideas into our existing learning algorithms. This was followed by a Q&A.
Meetup link: https://www.meetup.com/Brains-Bay/events/277296472/
Brains@Bay Meetups focus on how neuroscience can inspire us to create improved artificial intelligence and machine learning algorithms. Find more details here.
We received an overwhelming number of questions and couldn’t finish answering them. Hence, we set aside some time with Jeff and Subutai to go over some questions.
1) Are design affordances a natural consequence of The Thousand Brains Theory? In other words, does the theory explain why you’ll have an intuitive sense of how the object is supposed to be used even when you’ve never seen it before?
When you pick up an object you’ve never seen before, you will attend to different components of the object – usually components that you recognize. And each component will have its own associate behaviors. Imagine you have to guess an object in the box by touch only. When you touch the object, you recognize its rectangular in shape. You then start attending to the smaller components. You might touch something that feels like a button on the side and intuitively have the thought of pushing it due to your past knowledge of buttons. And when you recognize that the button is on top of the object, you might intuitively know that it is a power button. And the object you are holding is a phone.
2) Daniel Kahneman discusses fast pattern recognition thinking vs slow deductive thinking in his book Thinking, Fast and Slow. How does that relate to the Thousand Brains Theory? How do cortical columns handle “fast thinking” and “slow thinking”?
Disclaimer: We haven’t read the book yet.
This is a complex topic because we have reflex behaviors that occur starting in the spinal cord and also the brain stem. When you touch something hot, your arm retracts. That’s one example of fast thinking. Your cortex finds out about your reaction after you move your hand away.
Another example of fast thinking happens in the neocortex itself. Our neocortex is arranged in some sort of hierarchy. When we are familiar with an object, we can often interact with it without attending to different components (unlike the example in question 1). Athletes train themselves through repetition to improve their “muscle memory” so they don’t have to think about what they’re doing when they perform. If they think (i.e. attend to things) during their performance, they might slow down. Slowing down could be just a matter of hundreds of milliseconds, but that could ruin their performance. When the athletes are performing, all the activity happening in their brains will happen lower down in the cortical hierarchy (as it requires less thinking), which would be faster than if it has to go higher up.
There is not necessarily a major division of how the brain does “fast thinking” or “slow thinking”. It just depends on how much you are attending to the component. If you attend to something more – you will think “slower.” As ideas get more complex, more activity will happen higher up in the cortical hierarchy. And of course, for the reflex reactions, you don’t have a choice. Additionally, multiple cortical columns can recognize things collaboratively extremely fast.
3) Is the concept of amoral intelligence coherent? How can the maps we’ve learned in our brains be bad?
In the book, Jeff talked about how your maps of the world in the neocortex can be heavily influenced by old brain functions (i.e. primitive behaviors). If you were a pirate, you might have maps for different things than if you were a trader. Maps are not all identical, but the point is maps are just maps. The maps themselves don’t imply that you have to steal or trade. It implies that there might be some information that is more useful to the pirate, and some that are more useful to the trader. Maps do not have any moral compass associated with them.
The maps we create in our neocortex can be influenced by what we’ve learned – they could be influenced by our biases, training, and moral compass. However, a biased map does not imply that you have to use it in an amoral way.
4) What kind of computational role do oscillations play in a Thousand Brains Theory of Intelligence?
Oscillations in the brain have been studied for decades. Recently, we discovered that oscillations play a key role in grid cells – at least it is believed to be. We feel like we have a concrete understanding of how certain oscillatory processes play into the creation of path integration, which is a part of our Thousand Brains Theory. We have also speculated on how these oscillations can scale movements and models of the world. Oscillations are not included in any of our papers or outlined in the book. You can watch our previous research meeting on oscillations here.