Recursive Self-Improvement

From The Transhumanist Wiki

From the SL4 Lexicon:

Recursive Self-Improvement (RSI):

Recursive Self-Improvement is the ability of a mind to genuinely improve its own intelligence. This might be accomplished through a variety of means; speeding up one's own hardware, redesigning one's own cognitive architecture for optimal intelligence, adding new components into one's own hardware, custom-designing specialized modules for recurrent tasks, and so on. Humans cannot conduct any of these enhancements to ourselves; the inherent structure of our biology and the limited level of our current technology makes this impossible. But, we do have experience with a certain limited kinds of self-improvement called "learning" and "philosophizing", and it seems probable that a brilliant neuroscientist in the near future could theoretically use neurotechnological techniques to genuinely enhance his or her intelligence, and then apply that enhanced intelligence to devising more effective intelligence enhancement techniques, and so on. Unfortunately, the neurological structures corresponding to human intelligence are likely to be highly intricate, delicate, and biologically very complex (unnecessarily so; evolution exhibits no foresight, and most of the brain evolved in the absence of human General Intelligence). This makes it seem that human intelligence enhancement, if it can break through ethics barriers at all, is 15 to 20 years in the future, at the least. As Singularity analyst and systems theorist John Smart has said, "wetware is sexy to talk about, but messy and unethical mess with".

True Artificial Intelligence would bypass problems of biological complexity and ethics, growing up on a substrate ideal for initiating Recursive Self-Improvement. (fully reprogrammable, ultrafast, the AI's "natural habitat".) This Artificial Intelligence would be based upon:

• 1) our current understanding of the central algorithms of intelligence,
• 2) our current knowledge of the brain, obtained through high-resolution fMRI and delicate Cognitive Science experiments, and
• 3) the kind of computing hardware available to AI designers.

Futurist Ray Kurzweil has pointed out that, at the current rate of improvement in brain scanning technologies, we should have extremely high-resolution scanners, (more than enough to scan all cognitively relevant aspects of human neurology) and sufficiently high-density storage mediums to record all the data involved, by around the year 2030. With that level of detail, we could theoretically run one of these emulations on in a virtual environment, and the problem of "AI" would surely be "solved". However, this doesn't take into account:

• 1) discontinuous improvements in computing power or scanning technology due to nanotechnology or other unforseen developments or difficulties,
• 2) advances in Cognitive Science that indicate the complexity of certain brain areas is largely extraneous to intelligence,
• 3) qualitative improvements in scanning techniques, or
• 4) a global disaster or regime that drastically curtails technological progress.

The ability to genuinely enhance the hardware components underlying one's intelligence has not yet been observed in this universe, but Cognitive Science and the laws of physics seem to allow it. However, there is probably a minimum threshold of intelligence required before an entity can make qualitative improvements to its own intelligence; a chimp probably couldn't do it, some humans might not have the knowledge, and all humans would self-enhance less adeptly than a seed AI. As long as technological progress continues to occur, the inevitability of recursively self-improving intelligence will become more and more imminent. It would be extremely difficult to outlaw all the precursor technologies for intelligence enhancement; huge sectors of biotechnology, medicine, nanotechnology, and Cognitive Science would need to be suspended or eliminiated. For more information on Recursive Self-Improvement, see Part III of "Levels of Organization in General Intelligence" or "Accelerating Progress and the Potential Consequences of Transhuman Intelligence" at http://www.acceleratingfuture.com .

A lot of the arguments put forth on this site and acceleratingfuture.com depend on understanding of the idea of Recursive Self-Improvement, so visualizing this accurately is important. In some exotic cases, yes; where the programmers run an AI at incredibly slow rates, or when the speed, smartness, and ability to self-modify can't surpass a hardware or software obstacle, and stalls at a constant level of intelligence. (If, by some chance, the AI happened to stall at exactly around human-equivalent intelligence, then the programmers would have at least one additional ally to help get the AI past the bottleneck; the AI itself.) But in most cases, when you build a sufficiently intelligent AI, it will be capable of Recursive Self-Improvement. If and when a certain level of neurotechnology is made available to humans, it would only be a matter of time before they too enter into Recursive Self-Improvement.