Failing to use a better analogy, various scientists and philosophers compare the brain to a computer. Maybe you object to that view and believe that it’s more than that? What I’ll now show you are insights coming from my book and based on the latest neuroscience studies,1 which will help you make up your idea on the matter.
For sure, it isn’t the type of computer you are familiar with, nor even something that already exists. However, it is the type of hardware and software that companies like Google, Microsoft, IBM or the American agency Darpa are working on.
The neocortex is an IT system
About 200 million years ago, mammalians suffered a modification of their brain’s anatomy, which led to cortical growth. The neocortex, which evolved differently throughout the mammalian species, ended up giving this bumpy looking brain in humans. If we were to flatten these bumps, we would be left with some kind of organic tissue the size and the width of a napkin. This tissue, made of around 200 specialized cortical areas, performs dedicated tasks such as data treatment of sounds and images.
The fact that we perceive the world as a unique experience, combining all inputs from these specialized cortical regions, confirms that the neocortex controls the flow of information traveling between these areas. Consequently, we can conveniently apply the image of an IT network manager to the neocortex.
Cortical columns constitute your brain’s data processing centers
When observing through a microscope these cortical regions, we can identify columns, which look quite similar vertically and horizontally. This resemblance is one of the reasons why Vernon Mountcastle pushed the idea in 1978 that they had to perform the same basic operations.2 We need to look even deeper in the tissue to understand what this function is. When doing so, we see six layers made of neurons communicating vertically between them or longitudinally with neurons from other columns.
What several studies show, is that each neuron sends and receives data from a higher and lower layer, just like in Artificial Neural Networks (i.e., forward and back-propagation). Upper layers send more abstract data, which we confront with reality in the lower tiers, that is, at the level of our senses.
For instance, after seeing lightning, a brain will expect to hear thunder, based on acquired experience and knowledge. If this pattern is confirmed, the feedback is likely to be immediately assimilated at the lower layers, and a person will not even be aware of this process. If it isn’t, data will flow back and forth (as micro-volts accounting for the data variance), till the issue is consciously solved and the pattern confirmed or modified.
Neurons together with their connections are the processing units
Neurons switch on and off via action potentials and could mistakenly be compared to a transistor. However, these cells play a greater role, such as storing memories and communicating through their axons and synapses.
When exposed to a particular environmental condition requiring a thoughtful response, this neuron and its neural circuit involved in such response gets fired up. If a similar event happens later, the same answer is likely to be given, though probably more quickly and efficiently. Consequently, it is the evolving network of connected neurons (i.e., phenomena that are regrouped under the term brain’s plasticity), which can best describe your mind’s processing units.
It is also a common error to compare a synapse solely to a network router. On top of this message redirecting function, it also possesses memory storage capability and plays this transistor role. In fact, it contains more than a thousand molecular scale On and Off switches. Consequently, if neurons with their neural circuits constitute microprocessors, then their synapses are the CPUs.
The brain uses a distributed architecture
Since the brain is located in the head, you could think that this network manager uses a centralized architecture, but it doesn’t. Its architecture isn’t decentralized either, like in the case of an octopus, where brains in the head and each one of their eight suckles may decide on their own course of action.
In fact, your mind uses a Peer-to-Peer (P2P) architecture, very commonly used on the internet for file sharing. Each node works as both “client” (i.e., it sends) and “server” (i.e., it receives) to the other nodes on the network. There are no centralized units or other auxiliary mechanisms, which would coordinate the operations among peers (e.g., resource location, replication). No single neuron stores all the data, as it is distributed across various neurons, ensuring tolerance to failures. Furthermore, any node may have a view of the decision (through excitatory and inhibitory signals) and obtain the complete system understanding. All these particularities characterize a distributed architecture.
Your brain’s logical topology
If we push further this analogy with P2P, besides being distributed and having a physical hierarchical (i.e., multi-tier) structure, the human brain architecture has two logical topologies. Low-level queries, mainly associated with our senses, are structured in non-random connections. A recent study shows even that there is an almost universal equation,3 which calculates the number of neural associations. Very similar to the computing equation (2n), it is also based on the power of two: N = 2i – 1, where i is the number of sensory inputs.
The second topology is based around a random connection strategy, associated with neural networks responsible for more abstract notions. Though there might be no strict mapping between a neuron’s prediction and those of its peers, the fact that it mostly interconnects with neighboring neurons (95% of all neural connections lie within about 2 mm of the injection site), largely compensates for the unstructured nature of these queries.
With those two connectivity strategies, our brain gets the best of two worlds. It benefits through non-random connections of stable evolutionary solutions. On the other hand, randomness in superficial layers maximizes the capacity to discover possible new combinations across a broader range of sensory cortical regions and thus can more easily extract, discriminate, and categorize new patterns.
Your brain’s coding techniques
So how are neurons listening to some of their inputs and not to others, at any given point in time?
The answer lies in the brain’s temporal processing function. Neurons do not add up all their thousands of inputs but instead, are influenced by the inputs’ timing. If two or more neurons are firing simultaneously to another neuron down the pathway, it is more likely that the information will be passed on by this neuron further down the system. The downstream neuron perceives signals, which do not correlate in time, as noise. Thus, the brain listens to the firing up synchronization from up-stream neurons. The greater the synchrony, the stronger the attention and the faster the reaction.
The brain can take into consideration this synchronization by using different encoding modes to store and retrieve memories or to detect patterns:
- Rate coding describes fluctuations in the frequency of produced neural spikes.
- Temporal coding describes a code based on temporal relationships in the neural response.
- Population coding associates the firing patterns of various neurons coming from the same stimulus.
- Sparse coding focuses on the strong activation of a few neurons when reacting to specific stimuli.
Your brain uses big data processing
Neurons use changes in firing speed, amplitude, or shape to carry information, just as the telegraph uses a Morse code to transmit messages. For example, sensory neurons modify their activities by firing sequences of action potentials in various temporal patterns that vary according to the input properties (e.g., more or less light).
However, and unlike for the telegraph, the brain must integrate the activities of millions of neurons. It is where the analogy with big data processing can help understand this complex process. Though we still don’t understand what the brain’s Morse code letters are, we know that a system using some logic can discriminate inputs based on patterns, making sense of data in noisy environments.
To illustrate how the cortex can do that, imagine a sequence of excel spreadsheets, which capture the millions of activated neurons and their links (represented here with just five neurons, in four distinct timeframes) within a cortical column.
In my simplistic example, the observed pattern in the timeframe is composed of the neurons A, C, and E firing together according to specific features (e.g., amplitude or speed) and connection sequence. Consequently, and supposing that this pattern is associated with the image of a cup, any prediction or real sensory inputs containing a cup would show this activation code.
Now imagine that these spreadsheets are filed in a series of repositories, like a Russian doll. Each higher level cortical processing, would give a more abstract and complete picture of this cup, demonstrating that time and location are essential to understand how our brain codes its environment.
Brains can be copied to create intelligent machines
Are you now convinced that your brain is a sophisticated computer system? If not, it’s maybe because I’ve only shown its available network resources, connectivity strategies, and programming techniques. Explanations about the mind’s source code and its computing instructions are still missing, and without these rules it maybe hard to get a full understanding of how the brain works.
However, these rules exist and neurologists and philosophers are developing models, which can explain how the brain becomes a prediction engine. The computer industry is looking with great interest at these developments. In fact, many voices in the industry are now saying that the best way to create ever more intelligent machines is to invest in technologies mimicking ever more precisely the brain’s rules and functionalities.
If it becomes a trend, do you believe the next computer generations will look functionally like the human brain? Are you now convinced by the analogy between computers and brains? Let me know.
- What’s on their mind? Biological and Artificial Intelligence; Serge Van Themsche – 04-2018
- The mindful brain Cortical organization and the selective Theory of higher brain, V. B. Mountcastle and G. M. Edelman, MIT Press -1978.
- Brain Computation Is Organized via Power-of-Two-Based Permutation Logic: Author, Kun Xie, Grace E. Fox, Jun Liu, Cheng Lyu, Jason C. Lee, Hui Kuang, Stephanie Jacobs, Meng Li, Tianming Liu, Sen Song and Joe Z. Tsien; Front. Syst. Neurosci – 15 November 2016
- The Biological Path Towards Strong AI by Matt Taylor from Numenta (AI Singapore Meetup) https://www.youtube.com/watch?v=Sirm-xFPiOA – June 1th 2018
Other Matt Taylor video: HTM Hackers’ Hangout – Jul 6, 2018
