The human brain is the most complicated object we know of in the entire universe. Understanding how the human brain works will likely lead to the most significant advancements in the human race and quite possibly the genesis of a new intelligent species.
One of the most promising paths to understand the human brain is via full brain emulation through neural networks software. It’s a fascinating new idea and actually seems attainable in the relatively near future.
The Vastness of the Human Neural Graph
The human brain is a vast network of interconnected neurons. The human brain is composed of 86 billion neurons. Each neuron on average is connected to 10,000 other neurons through synapses. Synapses enable neurons to communicate with other neurons through electrical and chemical signals. In total, the human brain comprises a neural network of roughly 100 trillion synapse connections.
By comparison, the last known size of the entire Google Knowledge Graph was only a mere 570 million nodes connected through 18 billion facts (or connections). That’s only a mere 0.000018% of the neurological graph size of the human brain.
The Human Connectome Project
The human brains 100 trillion neural connections are not, however, connected randomly. How each of the 86 billion neurons is connected to 10,000 other neurons is what actually defines the mapping of the brain. These connections determine our behavioral health, intelligence, emotions, and are definition of consciousness.
Researchers are attempting to determine how each individual neuron is actually connected to build that map. That is the goal of The Human Connectome Project and the NIH’s BRAIN initiative announced by president Obama in April of 2013.
Mapping the human brains neural network is a massive undertaking but similar to the Human Genome project it’s simply a matter of engineering scale. The basic principle involves the automated slicing and scanning of the brain. Not an easy task and significant bioengineering challenges remain unknown. However, it’s a task that could be conceivably completed in next decade or two.
Emulating a Human Brain
Once we actually have the pathways of the human brains neural network defined, what do we do with it? That’s where emulation comes into the picture.
I’m a huge classic 80’s arcade game fan. I remember the excitement when I found out I could actually play almost all of those old games on my Mac using a program known as MAME.
It turns out the actual programming of all of those old large arcade cabinet games are stored on something called a ROM chip. These chips contained the entire code of the game. MAME is a software program that permits these games to be run on todays computers by simulating the old arcade computer that surrounded the original ROM chips. This is the process of emulation.
That same process of emulation can also be applied to the neural network model of the brain. By simulating the brains inputs of vision, hearing, and touch to the input neurons the neurological model acts and behaves just like a real human brain.
It’s a process that has already been tested at least in terms of scale. On August 8, 2013 the Japanese super computer “K” simulated a network of 1.73 billion neurons and 10.4 trillion synapses (a mere 1% of the actual human brain). The project used randomly generated neurons, since the actual human mapping is not known yet.
However, the scale of the emulation problem of the human brain is the important part. In order to simulate a mere 1 second of “biological” brain time, the K super computer (currently the fourth fastest computer in the world) took 39 minutes of actual processing time. The emulation used over 80,000 processors and consumed almost 1 petabyte of memory.
The Path to Artificial Intelligence
The results of such emulations demonstrate the incredible power and complexity of the human brain given the enormous amount of processing time required to emulate simply 1 second of 1% of the human brain. Full-scale computational emulation will require exa-scale super computers, which should be realized by 2020.
Assuming that that the human neural network can be fully mapped by then we may be less than a decade away from full brain emulation. If the Japanese “K” supercomputer can simulate 1 second of 1 percent of the human brain in 39 minutes, then all we need to do is to build a super computer that is 234,000 (100*39*60) times faster than “K” to simply achieve human standards. Given the growth of super computing power, that’s rather likely to achieve at some point not too far off.
While the benefits of full brain emulation to human health and cognitive abilities are clear, it raises some very important questions such as “Is it alive?” and “Does it think?”.
What if it can be made smarter? Why not “tune” it up a bit? What if it can just think faster than us?
Full brain emulation is currently the most promising path toward true artificial intelligence. Indeed, it’s almost straightforward which gives it such a high likelihood to succeed. The critical question we need to truly understand is will we survive it?