It appears that FaceBook’s AI Directory agrees with me on the implausibility of trying “to copy every detail that we know of about how neurons and synapses work, and then turn on a gigantic simulation of a large neural network inside a supercomputer, and hope that AI will emerge.”

“I’m going to get a lot of heat for this, but basically a big chunk of the Human Brain Project in Europe is based on the idea that we should build chips that reproduce the functioning of neurons as closely as possible, and then use them to build a gigantic computer, and somehow when we turn it on with some learning rule, AI will emerge. I think it’s nuts.

Now, what I just said is a caricature of the Human Brain Project, to be sure. And I don’t want to include in my criticism everyone who is involved in the project. A lot of participants are involved simply because it’s a very good source of funding that they can’t afford to pass up.”

Quotes from Yann LeCunn in Facebook’s AI Director on His Quest to Make Machines Smarter With Deep Learning in IEEE Spectrum.

Implicit somewhere in the assumption of AI spontaneously emerging from a big enough neural net is the idea that it must have happened that way in nature since evolution is presumed to be true.

Every so often, the research community comes alive with excitement over some new announcement of animal brain simulation, and tantalizes us with promises of a full human brain simulation.

First, there was the rat brain, simulating 10 thousand neurons and 10 million connections at a molecular level, and with an apparent ability to self-organize.

Then, there was the cat brain, with 1 billion neurons and 10 trillion synapses.  Or not, according to the rat brain camp, which denounced the claims as “shameful and unethical.”  But the cat brain camp stressed that they had “not built a biologically realistic simulation of the complete human brain.”  Pity those who think research is devoid of high drama.

Somewhere along the way came a simulation of the human visual cortex, with 1.6 billion neurons, and 9 trillion synapses.  Simulating the entire cortex, assuming the availability of sufficient computing power, would require a nuclear reactor just to provide the roughly 1 billion Watts of electrical power.  The brain makes do on about 20 Watts.  I leave it to our best and brightest to figure out how we would cool such a behemoth.

Call me skeptical, but I don’t think we really understand intelligence yet.  We create enormous simulations of things that have some ability to self-organize, in the hopes of stumbling upon intelligence.  But would we even recognize it if we saw it?  The premise is that if we just make the simulation big enough, it will begin to work.  On the other hand, researchers don’t learn without trying.

Artificial Intelligence (AI) research is not entirely without merit, because it has certainly produced useful tools for us, including simulated annealing, genetic algorithms, neural networks, and more.  But I don’t believe we have a good understanding of what intelligence is.  I don’t even think we fully understand what computing is, at a fundamental physical level.  We certainly know how to perform computation, but we lack a rigorous physical understanding of what information is, let along how it interacts with physical hardware of any kind.

When it comes to brain simulation, I’m not the only skeptic.  Researchers do not know what level of simulation—molecular or neuronal—is appropriate for brain simulation.  I suspect that intelligence and sentience are radically different than what the AI researchers assume.  And while there is certainly value in learning from nature, mimicry may not be a good long term goal. Nature and humans both build structures, but do so in very different ways, taking trees and houses as a point of comparison.  And humans are frequently inspired by nature, but ultimately seem to have different objectives than nature, and therefore build differently.

So are these brain simulations worth the enormous funding that they command?  Time will tell.  I think we probably will eventually develop artificial intelligence, but I also think artificial intelligence will look as different from natural intelligence as houses do from trees.  I’d like to see someone start building a “house.”