A recursion has become somewhat of a ritual of mine, much like those rituals found in obsessive compulsion or Asperger's and other parts of the autistic spectrum. I define a recursion as the cognitively computationally-complex (usually O(n)) methodically iterative behavior (program) that results in informatic digestion, otherwise known as one of the suggested mechanisms of the opportunistic-assimilation hypothesis in insight incubation theory; this holds for the Wp definition. An example of a recursion would be the times that I go through fifty papers in a night, or do a depth-five recursion through Wikipedia starting at some particular page. As a programmer, it is interesting to program new recursions, and perhaps the most interesting of all of the possible recursions is the hypothetical feedback recursion loop that establishes programmer-programmed unification, as studied in theoretical metarecursion and to some extent in self-actualization, transhumanism, immedeism, etc. Simply put, unification exists where the programmer and whatever that is programmed are one in the same, unlike the typical condition of the hu brain where systematic incoherence propogates and reloads involuntary programs and interrupts the programmer process, blocking unification. Although there are ways known to moleculary increase repetitious behaviors in ASD-folks, incoherence still propagates. Maybe there are other ways to reach the hypothetical status of unification -- other ways to do brain modification? So, supposedly (recursion-path minimization/optimization), there is some recursion-path that can take the programmer from lengthy recursions to those recursions that end up "back where they started" (the programmer) even more quickly, minimizing the length of such a recursion, while maximizing the covered 'surface area' and 'volume' of personal possibility space. There are a few ways that possibility space has to be restricted in the first place; first, no supernatural phenomena, second, no network-wide byproduct utilization effects (i.e., Egan's complementary hyperdimensional aliens encoded into some basic fundamental numbering system within the fabric of reality, except expanded into the brain); so this limits everything to cells, metabolism (not just fMRI studies of metabolism), GRNs, proteins, genes, mRNA, stochastic chemical reaction networks (signal-transduction pathways), neurotransmitter circuits, axon-dendrite connection circuits (physical neuro connectionism), what else can be investigated for the reduction of the gap between programmer and programmed? Another approach is the same as the minimal genome project, a minimal programmer project, starting with the comparison of simpler bilateria brains to hu brains, perhaps the in vitro synthetic engineering of brains. The Allen Institute has recently been given the task of creating the human brain map to detail where in the brain certain mRNA molecules are found, indicating gene expression. What the Allen Institute does is take a brain and slice it up into cross-sections and then sample the local concentration of mRNA (microarraying), indicating gene activation in that region of the brain, supposedly during the activity immediately before death. In these methods, changes would be implemented through viral gene therapy (bacteriophages) or through DNA synthesis/splicing; in the case of viral gene therapy, there is a limited range and limited amount of viruses that can be synthesized within the brain (which is a good way to do it with an implant), but out-of-body synthesis might be interesting; also, neuropharmacological drugs could be an option. In the case of DNA synthesis, results can only be tested out in the next generation of implementation of hu. Maybe stem cell replacement therapies and restarting neurogenesis, but flat out killing existing brain structures and hoping neurogenesis will fill in the gaps does not sound like a good idea Other possibilities include brain implants as a method of stimulation and perseveration, data transfer, brain-computer interfacing, etc. This is also where the submind idea comes in at, as expressed on the roadmap; basically the concept is to expand attention by growing brains in a vat to help process additional information. Along with brain implants comes the idea of using rTMS to modify the state of regions of the brain. Or the physical grafting of new neurotissue structures into the brain, once the tissues have been grown and targetted for a certain region of a well-characterized brain. Neurotissue engineering also allows for other possibilities, perhaps not necessarily 'subminds', but definitely aspects of attention processing and so on.

One of the methods in programming is called sandboxing, the practice of putting unit modules and their unit tests into a sandbox instead of immediately implementing them, thus protecting from possible damage. This method can be done via theoretical simulations and experimenting with other brain tissues that are used to represent the current brain structure and programming of the hu. What changes are you going to make? This is where a healthy interest in computational neurosci starts becoming useful, or, alternatively, an interest in neurofarms. In chemistry, there is something known as combinatorial library testing, where a large supply of molecules are dumped together to see if there is some molecule within the population that can react in some way, any way, to come up with a solution that matches a selection (kind of like microbe selection experiments and directed evolution). The problem with doing this with human drugs, or making modifications to hu systems, is that when you start scaling it up to the trillions of possible connections that a human brain can have, you start testing some of the upper bounds of the energetic and physical provisions of the universe (what, with the 1E77 particle existence count bound). Although not all trillion factorial possibilities can be explored, significant constraints on possibility space can be established. Even then, the hu effort required to make these experimental setups, even the computational ones, are far out. Thus my interest in kinematic self-replicating machines -- either to build the supercomputers necessary for Markram simulations, or for the physical basis of the neurofarm [of self-improving, recursive brains]. And more generally, such self-replicating machines offer physical expression of computational recursion, an interesting property.

Self-replicating machines already exist, as intensely study in biology. Solid-state replication isn't quite the same. Having been studied by von Neumann, Freitas, Merkle, Drexler, and many others, perhaps a new way of coming about self-replication is needed, namely a brute-force tactic: skdb/oscomak.The societal engineering knowledge database (skdb) is kind of like aptitude, yum, CPAN, rpm, etc., in that it hosts modules (blackboxes) for projects, whether digital or physical. These project packages include simulation tools for tweaking, software packages to rewrite the package for a particular use, explicit well-documented links and dependencies to other projects, etc. Skdb started as a need for self-replication: the idea is to come up with as much social knowledge as possible (stuff that can't be derived from first principles on the spot) and then connect all of the components together and fish out the Hamiltonian paths that supposedly represent the d-loops, dependency loops that are thought to indicate an ability to self-replicate. In essence this is a massive manufacturing knowledge compression project, David Gingery was only the beginning (and Kevin Kelly knows it). My approach is to serve up information content in natural language that users can process and then encode into formalized programs, file formats, specifications, etc., which can then be automatically processed by computational simulations to come up with a self-replicating machine; and along the way we can do human processing to come up with other projects, anything from engines, houses, microprocessors, compressors to particle accelerators -- note however, this is not a replacement for human creativity and using heuristical recombination and mutation of parts to build something novel and useful.The implementation involves ghost-layering file format io metadata/spec files on top of software packages and well-structured flatfiles for data on various materials or whatever the package deals with; obviously my focus would be on manufacturing knowledge and fabricational complexity.


2008-05-17: Recursive self-improvement via open source rTMS
2008-05-18: Slashdot post re: brain hacking, sandboxing, open source rTMS, etc.
2008-05-27: Recursive self-improving neuroplasticity