This is a copy of http://heybryan.org/buildingbrains.html.

Okay, so while the 292 used GB of the 500 GB LaCie hdd edmini (a "network drive" -- a POS, IMHO) is dumping itself on to the Maxtor OneTouch 750 GB hdd (an /actual/ external hdd, reformated to ext3 fs), I thought I'd take a few moments to write out what it is that I am talking about when I say I'm working on building brains.

---

One of my interests is self-augmentation, particularly when it's enhancing rather than dehabilitating and particularly when it's regarding the brain. Many have proposed a variety of methods of augmenting the brains -- the neural implants, the Mind Uploading Research Group, nootropics and drugs, cryogenics, and so on -- but the problem that we keep running into is that the brain is behind the skull, it's delicate and the best we can do is shove some wires and electrical probes with an implant into it and hope that we can get something out of it. I am optimistic of the prospects of brain implants, but whether or not they are going to be solving the fundamental problem is another issue entirely.

So, an interest of mine is writing science fiction, and one of the concepts that I ran into a long time ago was the idea of "divergence". Not even twins have identical brains at birth. You can clone yourself, yes, but there's going to be significant divergence from the original copy. I implemented this concept into the story [but you'll have to wait for that [or something]]. Back on topic: this divergence means that copying the entire brain, atom-by-atom, though it would be interesting if ever possible, it's just too lengthy and too complex of a procedure to implement at this time. So the alternative is to consider divergence and then ask just what is it that we can preserve from copy to copy? From person to person?

Turns out that this has something to do with the macrocolumnar organizational basis of the brain, and it's interesting the different types of architectures that can be constructed from this (see here + a good vid). So, to some extent this information is within the genome. The information on cellular metabolism, how long the neurons are going to stretch their dendrites and axons, various rules for synaptogenesis, the differential models can be recovered from these interfaces from presynaptic clefts to postsynaptic clefts and so on. And from this we can preserve some 'identifying' information while acknowledging divergence but still building brains that to some extent are an expression of what the author intends.

Anybody in the know would realize just how hard it is to grow a brain, much less a noncontaminated tissue culture in the lab. It sucks immensely. Growing a full, beating brain would force us to face some issues of vascularization and other aspects of tissue engineering. Organ farms are going to have to solve this one day too, but let's ignore it. What's important about the brain is the interaction between the neurons and the growth of the neurons to construct the physical dendritic chunk of matter that basically represents our brain. This doesn't have to be in the typical form of a brain. It'd certainly be nice. But it means that we can play around with some other organizations, like the neuron culture in a dish. Many researchers have experimented on neural tissues with patch clamp techniques, Markram built a robot for it and probably some others have done the same by now. Other researchers opt for the microelectrode array (MEA) that makes up the stereotypical brain implant. Because of this, it is possible to use computer networking to interface the different MEA chips together and to have different, on-the-fly reorganizations of brain matter. At the same time, that information from the genome could be 'randomly' mutated and new neuron cultures could be grown. The lag is going to be terrible, but if a slice of neurons can learn to fly a (simulated) jet, why can't they learn to wait for their neighbors? Zindell's moonbrains were terribly slow as well ... anyway, that's the basic idea.

There's a lot of information in the brain. It gets more information than it generates because of the nature of the neuron (more axons than dendrites). And given that there's only a limited number of motor outputs in the human body, the brain is only able to pass on so much information. What could we do if we could recover some of that information? By this I refer to inserting brain implants into hunks of neurons and then recovering information that is otherwise not used to generate the final motor output of the human body. This could be useful for automated grammar processing, visualization, programming, for so many countless tasks. And in the case of a completely compartmentalized brain farm with GAs running all over the place trying to come up with new, more interesting genomes for slices of neurons, this 'unused information' can be used to debug the whole thing, or applied towards other areas (grammar, visualization, programming, data dumps, ...). I'm not saying that the information is going to have a one-to-one correspondence with spoken English or something, it's going to be pretty weird data, but surely there's ways to feed this off to some mindbots and other computational processes that can then go do something reliable with that information -- how'd you like to be able to spawn of a few thousand Google searches every few seconds? And have the results automatically sent through a few of your own home-grown neural slices? I know I would.

And just how am I going to go about building brains, much less brain farms? One interesting possibility that I'm exploring is the biotech toolkit project (which includes a complementary DIY neurochemicals kit) and overall it's related to SKDB, the societal-engineering knowledge database, which the 'bioreactor' project is a subset demonstration of. There's also my attentional augmentation system that I explain below.

(see the HTML page for the notes/references and that explanation)

- Bryan

Of late, there has been a ballooning blog storm on self-directed neuroplasticity (1, 2, 3, 4, 5, 6, etc.), which has interesting implications in light of toposophy, which "deals with the theoretical problems and possibilities of attempts to extend and amplify one's mental potential." And the page, over at Orion's Arm, continues, "... most mental enhancement is incremental, involving merely adding on new capabilities and integrating them with the already existing framework. Typical cyborgisation procedures as memory enhancement, skill libraries, coprocessors, extended neural networks and pidgin lobes fall in this category. While such additions may cause mental shifts and re-evaluations of identity, they merely extend the basic architecture of the underlying mind. This kind of bootstrapping can be self-supporting, each improvement making it easier to add new improvements, producing an accelerating mental expansion, a singularity."

So the basic idea of recursive self-directed neuroplasticity can be broken down into a few key ingredients. First, we need synaptic plasticity from modern neuroscience. This includes your typical bunch of functions: long-term potentiation (LTP), long-term depression (LTD), spike timing dependent plasticity (STDP), and neural facilitation (short-term plasticity). These have a molecular basis.

And next in the recipe is the self-directionality, otherwise known as "attentional effort" and sometimes sustained attention. Many prominent physicists, philosophers, poets, mathematicians and others have commented on how insight works, such as Asimov, Poincare, Feynman, Tesla, Helmholtz, etc. "Incubation sometimes requires a very long break: Feynman noted that "You have to do six months of very hard work first and get all the components bumping around in your head, and then you have to be idle for a couple of weeks, and then - ping - it suddenly falls into place ..." (Csikszentmihalyi and Sawyer, 1995, p. 350)." So the incubation theory of insight and learning, especially in the opportunistic-assimilation hypothesis (failure indices left in LTM, retrieval from LTM to STM just gives the original problem space without the crappy stuff), can help mold neuroplasticity over a long period of time -- simply by directed attention, such as changing the cholinergic and GABAergic afferent/efferent inputs into the prefrontal cortex and the visual cortex. That starts to get into the neuroimaging sciences -- anybody have an fMRI machine that I can borrow? No, really.

Next is the recursive aspect of it all. As you learn, you can make more sense of it all, and then direct learning towards particular other tidbits; perhaps that's why I am so interested in focusing/attention -- it can lead to recursive self-improvement in the extropic transhuman sense.

It's interesting, really. You don't have to sit back and watch this all go to waste, since you can do scientific experimentation in your own home with neural slices. It's not too hard to get biopsies from biotech companies, then wire up the neural slices into electrode arrays, patch clamps, microfluidic systems, etc. Things get difficult with signal analysis, although I can't vouch on that front quite yet. And what we learn from training organotypical slices, or small patches of neurons in a dish, could tell us something more about our own plasticity. A good place to start would be here on that front. Next step is a cheap, nearly free neurochemical kit. The price of neurochemicals on chemcial supplies websites is way, way too high to do anything interesting with ($1/mg), so we're going to have to engineer bacteria to produce our neurotransmitters for us.

- Bryan