Critique of the Idea of Artificial Neuron Replacement

There is a popular idea that has been around for a while now, the concept that we can augment our biological brain with nanomachines that take the form and function of artificial neurons.


Let's examine this idea in detail, see how it holds up. The following summarize well this concept:

"drastically alter our selves, by making purposeful changes to the way we operate, combining our selves with engineered systems (including decision-systems), and converting our selves to superintelligent agents."

"Using nanomachines to gradually replace the organic brain cells with synthetic neurons will probably be the first step in truly bridging the gap between man and machine (think Ship of Theseus analogy)."

I have been considering this. Although I appreciate how this route may seem to resolve certain issues such as the conscious identities paradox, there are certain extremely serious problems here.

I will set aside for the moment the flawed notion that digital computers are “superintelligent”, whilst collective computers are somehow an inferior form of computational architecture. I have explained elsewhere why this is not the case, and will suffice to say for now that collective computers are not only inherently massively parallel but also massively interconnected, while digital computers are inherently neither. Digital computers are inherently sequential – they can through arduous effort be made to have some (usually small) degree of parallelism, but they in no wise have anything that reflects massive interconnectedness.

However, that is not the ground I want to cover here. In this post I want to examine in detail the idea that miniaturized electromechanical nanobots would make good piecemeal replacements for biological neurons, that we can dust into our domes and have them slowly replace (or augment) said neurons into a resulting synthetic brain of some kind.

Let’s go back to our newly gained understanding of the pieces of collective computation, and how they are evinced in our own brains. As we now understand, there are two main pieces here. First, the topology of the network, ie how each neuron with its thousands of inputs and outputs is connected to every other. Second, the resistance at each synaptic connection point, which is not a static value in our own brains, but a complex variable function that we have only begun to understand.
These components both individually and together affect the computational result of the collective computer – they are both very important.

Now, let’s introduce one of our hypothetical artificial neurons into the mix. Let’s assume a realistic scenario, where we dust these into someone’s brain and they must adapt to the local conditions of the neuron they end up replacing. In other words, we are not designing specific artificial neurons for specific biological neurons in someone’s individual brain. What must one of these artificial neurons be able to do?
Well, several things, all of them extremely challenging. Once it identifies a neuron that it is going to replace, as a first step it must match its topology exactly, if it is to replace this neuron. That means it must determine how many tendrils it has, and deploy up to 10,000 of these long, thin tendrils itself. Bear in mind that the number of neuronal connections varies tremendously, but 1,000 to 10,000 should cover most cases, to the best of our understanding. And the termini of these “arms” are extraordinarily tiny – the animations showing this replacement, where a big fat artificial neuron replaces a big fat biological neuron are laughably incomplete to capture the complexity of the physical dimension to this problem.

Turns out, that’s probably the easy part.

The next challenge will be to seamlessly replace all of the synaptic connections of our soon-to-be-replaced biological neuron with the artificial synapses of our synthetic neuron – and each of these are interfacing with biological neurons, that have not of course been replaced yet. Bear in mind that how they convey signals is entirely different from what our artificial neuron is likely to be able to do – they are chemical processes, not electronic in the sense we know from our digital technology – complex biochemical neurotransmitters that must be released and/or taken up with incredible precision in order to accurately match the effective neuronal resistance of that synapse in a working brain. The ability to adjust the synaptic resistance with exquisite precision is what gives our bio brain’s collective computer its “programmability” – the collective computer equivalent of what we call software in the digital realm.

Therefore, your artificial neuron must have so much biological capability that it is really hard to imagine what it would bring to the table in terms of simply its being artificial.

Now, assume it can do all of these things – which I do not, but I understand some of you might. What specifically does this newly hooked up artificial neuron bring to the brain that somehow conveys anything special above what our normal neuron could? It can’t fire faster, because it is interfacing with biological neurons that can’t take this. And it is very important to understand that “clock speed” for a collective computer is not nearly as important as clock speed is for a digital computer – precisely because a collective computer can do in 5 or 6 clock cycles what would require a digital computer millions, billions, or even more cycles (it depends on the computational task) to achieve.

Not trying to be negative, not saying all this is impossible or anything. But it’s fun to occasionally think these things through.