The silicon brain

"Will it ever be possible to emulate the human brain in a machine?"

This is a question that every scientist, engineer, and philosopher should ask at one point or another. Whether it is posed as a question about the limitations of science, the limitations of machinery, or the limitations of the mind itself, it's a question that has immense implications for the future of mankind. Ultimately, the only real answer will be made when the feat of brain replication has been accomplished, and not a moment sooner. In the meantime, though, it's a fun discussion to waste an afternoon on.

First of all, let's distinguish between the brain and the mind. The argument of whether consciousness can ever arise in a machine is a diverting one, but it is not the one I wish to pursue here. For the moment, I'm focusing on the brain in particular - in other words, a machine emulating the brain should be able to act, respond, and interface with the body in the same way as a brain. It doesn't necessarily need to produce consciousness - after all, there's no definite way to tell if a biological brain produces consciousness. It does, however, need to be able to replace a brain in a human body, and take over all of the functions of a brain, including mental tasks such as memory, recognition, and abstract reasoning. Whether this machine incorporates some allegedly nonphysical component of the mind is unimportant, as long as it can reproduce exactly the functions of the human brain. In other words, there is no need for there to be an actual conscious mind located in this machine, as long as it gives the appearance of having one.

I feel that this is an important point because there are an alarming number of people who seem to believe that it is impossible to locate the mind in the brain. I won't attempt to defend that belief here, because it is one that I do not share and which I might easily misrepresent. At the same time, I don't want to attack a belief that I don't fully understand. Instead I would like to circumvent the entire argument by being satisfied with a brain facsimile that merely functions in every way as if it had a conscious mind, rather than attempting to prove that a conscious mind exists within it. As far as I am concerned, there is no functional difference between a reasonable emulation and the real thing.

Now that that's out of the way, let's talk about the fun part: the technical details. As I see it, there are 3 fundamental obstacles in creating a functional replica of the human brain:

1) Size constraints
2) Interfacing
3) Architecture and dynamics

Let's start with size constraints. The human brain contains around 100 billion neurons, and takes up a volume of 1700 mL, including blood and fluids. That sounds fairly compact - 17 nanoliters per neuron. For comparison, though, Intel's recently announced Tukwila chip fits 2 billion transistors onto a 21.5x32.5mm chip. Assuming a height of 4mm, that's around 2 mL per 2 billion transistors, or 1 pL per transistor. That's less than one ten-thousandth of the size of a neuron, meaning that even if it takes 10,000 transistors to simulate a single neuron, size constraints are not the limiting factor.

A more difficult constraint is that of interfacing. There are two problems with interfacing that must be addressed: biocompatibility and signal transmission.

I recently read the book Neural Engineering, edited by Bin He, which discussed among other things the practicality and state of the art of internal electrodes, i.e. electrodes which are surgically inserted into the body. Unfortunately, even the most modern and high quality gold and iridium electrodes degrade rapidly, lasting only a few months before they are degraded either by the environment of the body or fatigue induced by electrical current. Therefore, I believe that biocompatibility is an important hurdle to overcome in the simulation of the brain.

It is also important to consider that, although the brain could be simulated in whatever way we choose, it needs to interface with the body, which expects the same outputs as the biological brain. Therefore we must be able to adequately translate into the natural neural language to communicate with the body. This leads me to the final problem I perceive in the construction of an artificial brain, which is the choice of architecture and dynamics of the system itself.

So far, attempts to mimic the brain have been for the most part entirely artificial, using a standard computer architecture and emulating the brain only loosely - for example most work in artificial neural networks makes the (in my opinion incorrect) assumption that neural information is carried strictly by the rate, and that weighting of time-independent scalars can produce brain-like function. Unfortunately, ANNs do not come close to emulating the brain. This is because we do not understand how the brain works on a fundamental level (as I have pointed out in an earlier post, but please indulge my persistence on the point).

It is my belief that an emulation of the brain will necessarily incorporate a brain-like architecture, with all of the rich signal dynamics available in the biological neuron taken into account. Although it may be overoptimistic of me to say so, I believe that the 10,000 transistors per neuron currently available is sufficient to encode these dynamics, if used properly. I most certainly do not believe that the brain will be adequately emulated in software alone under these constraints - rather, I think a new computer architecture is necessitated, one which incorporates more rich dynamics than that of a simple binary datum. A pseudoanalog computer may be sufficient to carry out the fundamental operations of the brain, and I think a brainlike computer architecture will be one of the most significant advancements of this century.

So when do I think we'll be able to replace our brains? My guess is circa 2100 AD. The fundamental limiting factor at this point is the interface, which must be biologically compatible. However, in nonbiological (for example, robotic) applications, I think we'll see brain emulation as soon as an adequate architecture is designed and implemented, which I hope happens before 2050 AD. In any case, it is certainly within the realm of possibility, and perhaps even probability, within our lifetimes.

We've got chemistry, baby

It's Valentine's Day again, which means I'm wondering about the same thing I wonder every year at this time: the neurobiology of love.

As far as I know there are two chemicals especially involved in love: oxytocin and vasopressin. That's not to say that they're the only two chemicals involved; dopamine and serotonin are involved in the pleasurable rewards of love, norepinephrine in the stress of love, and testosterone and estrogen in the chemistry of sex. Chemicals called pheromones might play a part in attraction, if we're anything like fruit flies. But in bonding - which, all things considered, is the hallmark of love - oxytocin and vasopressin seem to be the culprits.

Like most findings in neurochemistry, the discovery that oxytocin and vasopressin are involved in love was made through experiments with model organisms rather than humans. In studies with rats, oxytocin was found to be instrumental in bonding. Studies with voles found that vasopressin (or its receptor) was instrumental in fidelity, at least in males. Similar studies have been done in humans, although due to the squishy nature of genetic modification in humans, these studies are more correlational and less causal.

The chemistry of love is fascinating, but it raises an even bigger question: how much of our conscious thought is influenced by our brain chemistry? Certainly there is a large emotional component to love, and it's a bit easier to swallow that chemistry can affect emotions, over which we have no conscious control. But the idea that chemistry can influence things like attachment and fidelity, which are generally conscious decisions, raises an important ethical and philosophical question: if chemistry affects our thoughts, how responsible are we for our own actions?

My answer is that we are still completely responsible. Actions will always render consequences, and to pretend that because we are not in control we have not caused an effect is patently absurd. For example, if a man were to cheat on his wife and claim that his vasopressin levels made him do it, it would not undo the fact of his infidelity. I believe that, philosophically, we cannot excuse a transgression simply because of the circumstances leading up to it. As neuroscience discovers more and more about the workings of the mind, this is a stance that I believe becomes more and more valid, because ultimately nobody is completely in control.

The chemistry of love and the ethical debate about free will have two things in common: they're both still hot topics of discussion. Is love "just" chemistry? Does free will exist and, if not, can we reconcile it with traditional ethics?

What do you think?