I will take a human brain with that please

I will take a human brain with that please

As we learn more and more about the intricate world of neurons, their connections and functions, many of us become tempted to simulate neuronal circuits in computational programs. The science and art of building neuronal circuits in silico goes by several names including computational neuroscience, neural networks, artificial intelligence and many others.

An interesting and intriguing version of these efforts was published recently in the journal Science (“A large-scale model of the functioning brain. Science 30:1202-1205 (2012)”). The authors wanted to build a computational model that could implement a variety of different tasks including copying images, recognizing shapes, counting and even some elementary forms of reasoning. They decided to build a model consisting of 2.5 million neurons. This may sound like a lot of neurons. Actually, the human brain is thought to contain on the order of 10¹¹ neurons (about 100,000 million neurons). Not that we need to simulate them all. More on this later.

They endowed those model neurons with several properties that make sense in the context of what we know about Neuroscience. For example, those neurons fire all-or-none “spikes” like real neurons do. Some of those neurons are highly specialized. There are some neurons that would receive the visual input and attempt to mimic the retina. Other neurons direct a simulated arm that executes simulated movements. And lots of neurons are involved in the nitty-gritty computational details of cognition, of implementing the tasks, making sure that the input is interpreted in the right way to generate the desired motor output. Although several of the different pieces and components of the model were well known, it was nice to put it all together and take initial steps towards actually trying to simulate a brain.

The authors have a neat web site where you can watch videos of their simulations and the tasks that the model can solve. The simulations are so cute that the authors even decided to assign a name to their brainchild: “Spaun”, which stands for a rather fancy Semantic Pointer Architecture Unified Network. Spaun even passed some questions in an IQ test (which may make several people wonder about the actual utility of such tests!).

There is strong interest in the Computational Neuroscience community to better understand the algorithms by which the brain can solve complex tasks. Related to endowing computers with the capability of solving specific tasks is the notion that we can teach computers how to learn. Part of the magic performed by our brains is the capability of rapidly learning, adapting and inventing new ways to get things done. Storing information and rote memorization is not enough. Digressing quite a bit, many teachers and curricula still do not get it: computers are already much better than we are at repeating information. But this will be the subject of another blog.

What will a model that implements human cognition look like? We do not know yet.

Forgetting questions about computational muscles, it is not the case that we can simply simulate 10¹¹ neurons and start selling human brains. A realistic simulation of every single nanometer of cortex is not the goal. The beauty and power of models comes from abstraction and capturing the critical rules for computation. A beautiful short story from Argentinean fiction writer Borges illustrates this point when describing how useful maps are (as a two dimensional model) and how pointless it would be to create real scale maps where one mile is represented by one mile. We still have a long way to go to understand how neuronal circuits perform the magic tricks that they do in a seemingly effortless manner. However, as Spaun illustrates, there is rapid progress and a lot of excitement in the field and many of us work hard to try to get those answers.

Will we be able to actually “build a human brain”? What about “simulating a human brain”? The question is not really where but rather when. Imagine the possibilities.

Mind the quantum?

Mind the quantum? 

One of the greatest adventures of all times for Science involves trying to understand how our brains work. Our capacity to perceive the beautiful colors of the rainbow, to build machines to take us to the moon, to prove mathematical theorems and to fall in love depends on the intricate circuitry of many neurons in our brains. Francis Crick pungently wrote: "You are nothing but a pack of neurons". 

I have recently read a great book entitled "Physics in Mind" by Werner Loewenstein, W. (2012). The book takes us on a delightful journey that starts with the Big Bang and ends up discussing the extent to which quantum mechanics plays a role in brain computations.
With precise, enticing and often provocative prose, Lowenstein dares delve into fundamental questions at the boundary of Physics, Biology, Neuroscience and Philosophy. 

The author argues that understanding consciousness will require interpreting cognitive phenomena under the microscope of quantum formalism. He is not alone in these claims. Giants of the caliber of Roger Penrose and many others have made related claims. 
Yet, I wonder. Physics made great advances by considering simple models. Einstein famously argued that models should be as simple as possible. There is beauty, elegance and enormous power in thinking about an elephant's weight as concentrated in a single point.

Theoretical Physics is a well established field with several centuries of demonstrated successes. In comparison, Neuroscience and, in particular Theoretical Neuroscience, is the new kid on the block. Before postulating complex models, we should push the simpler, classical ones and see where they take us. We should start with neurons, synapses and circuits and try to capture the mind. It is likely that we will take many steps in the wrong direction and we will fall several times. Yet, what we stand to gain is enormous. If we can develop a "classical theory" of the mind based on neuronal ensembles and their interactions, we will be able to better understand who we are, to treat devastating neurological conditions, to build intelligent machines.