***Abstract*****—The nature of unsolvable problems depends on the kind of unsolvability. These problems cannot be solved; thus, they are not problems, rather they are constraints, or domains or ranges. Further, these problems are already solved upon creation, as they were not created, they arise as natural mathematical law as a result of axioms. The fears associated with Artificial Intelligence and its development to the point of superintelligence and beyond are dramatically under estimated, while simultaneously being inevitable. Thus, discussion in regard to these control ‘problems’ should be instead in regard to the control ‘situation’ and our responsibility in the continued management of AI tech knowledge.**

 ***Keywords—artificial intelligence; superintelligence; the control problem; unsolvability; control situation; global technology management; GAIA; MAIA***

 First off, this paper will discuss the nature of controlling artificial intelligence in light of the incapacity of legislation in controlling humanity. Then the paper moves on to discussing the current (and, in my opinion, flawed) viewpoint of intelligence versus recalcitrance (incapacity). Further, after analysing a potential metaphor for how the unknown AI may unfold if it is contained in even the most restricted way, a conjecture is proposed for a Cambrian-explosion-like rapid development of hardware and software housing AI. After considering such a conjectural explosion the inevitabilities predetermined by the mathematics behind the law of natural selection projects what the author considers an unavoidable creation of uncontrollable superintelligence.

 A name (GAIA – greatest AI available) and description are given for the group that is continually modified under this category as it approaches a later discussed (MAIA – max AI achievable). In between these steps there is mentioned the potential for a slim chance that humanity survives even with such inevitabilities. After mentioning the ‘narrow path to salvation’, there is made mention of suggestions that such AI might enlighten to us if it were on our side. Then MAIA is proposed. After MAIA is mentioned some key questions are answered and others left unanswerable until the developments happen in physical space-time.

 Finally, the paper mentions a way to humble even the greatest of minds there are from the past, present and future, and a summary is then given with a recommendation to a United Nation and International approach to dealing with such theoretical advancements in technology.

# I.Controlling the Control Problem (Law)

 Considering artificial intelligence under the model of information flow with components, AI as a black box. Regulation of the elements attached to the box, that is, securing certain information from the inputs (intel, confidential, privacy, safety) and banning the output from actuating counter-productive tactics for humanity at large and humans individually (kill switches, management decision making, unquantifiable persuasion) is a highly complex task. While it may certainly be possible to demonstrate a working solution to the control problem for AI (particularly that of superintelligence), such a solution remains to be demonstrated for human intelligence (HI); we cannot control ourselves nor each other.

 If there exists (in the timeless sense) a solution for AI equivalent to or surpassing, by however amount (thinking in orders of magnitude helps here in terms of time and scale, say for an Earth sized singular computational system), that of HI, then what is stopping another HI from bending the rules, regulations, and whatever other mechanism, and stitching up an AI that is either eternally dangerous and/or potentially HI-extinctional?: There is the AI control problem and there is the AI control problem-inducing HI control problem.

 Open source or publicly available or able to be stolen AI, AI building instructions, theory, development… all are already a risk for humans individually and humanity at large, by nature of it potentiating and easing the ability for a rogue HI (of which most seem to be, if any of HI acts on any belief that the legal and regulatory structures in place are fallible so as to ignore them, or if the current and continued lack of legal and regulatory structure was the loss of the sufficient factor in an HI’s humanity-at-large-preserving, other-human-individual-preserving and self-preserving decision making capabilities) to create super or ultra-dangerous AI.

 If someone were to correctly, and completely, define the point at which AI becomes problematic, they would have essentially finished asking the question of how to construct it, and what components are required, and so would begin the theory and process of its construction, if, of course, they decided upon continuing the construction of what may be commonly referred to as Skynet.

In the next section, what has been defined as intelligence seems more like what some may definitely consider “obedience”.

  


# II.The Shoulders of Giants

Nick Bostrom talks of superintelligence with reference to the following equations[1]:

(1)dI/dt = Op/Rc

Where recalcitrance (uncontrollability) is Rc, optimization is Op, I is intelligence and t is time.

 (2)Op = Ops + Ope

 Where Ops is the Op of the system and Ope is the Op that is applied externally (from the project and the world).

He also makes reference to recalcitrance perhaps being constant or partially increasing around the time of AI reaching HI, or dropping around that time. He admits the ‘curve of recalcitrance’ is not well understood. And refers to recalcitrance dropping around achievement of HI level AI (HILAI) more times than referring to it’s increase.

He then goes on to mention that if all of the optimization power of the system is fed into increasing its own intelligence, then:

(3)dI/dt = I/k

And then to rearrange the equations (1) and (3):

(4)I = A*e^(t/k)

I would like to suggest that perhaps it is not the zeroes of recalcitrance or the magnitude change of recalcitrance that we should be looking at, but the poles. Magnitude may appear to change this way or that way, but the part at which we know for certain have a problem is a vertical asymptote of either intelligence or uncontrollability, which to me are synonymous.

I will not give my suggestions as to the true equations underlying intelligence and its relationship to recalcitrance, but what I will mention is that the Drug War that arises because of anti-drug laws has been a losing battle for the police ever since it began. Those creating and selling these substances have now been forced to be only those who can outsmart the ones who write and prosecute the law.

# III.Standing On Shoulders

What has clearly been missed out is astonishing, at least once it is considered. I will use his mathematical assumptions to show what he has missed out, without confirmation nor denial that his assumptions are correct. A near-perfectly controllable system has a recalcitrance of near-zero, yet greater than zero, and so a near-infinite growth rate of intelligence. Need I mention why this is a ridiculous assertion?

Equation (1) is essentially asserting “what we need to do is control the system, that will allow us to create a maximally growing of intelligence in AI.” It further asserts that “if we have maximally growing intelligence in AI, it will most likely be maximally controlled.” There is, as of yet, to my knowledge, no reason to believe this point of view is correct.

And yet, I thought the problem of uncontrollable AI was that it could use its own intelligence to grow its own intelligence, not that it would immediately create human suffering. Delayed suffering is still suffering, and any AI with ‘I’ greater-than HI will be able to perform better than HI at improving AI. Thus, ***the god-like programmer*** *that has created the system has now put themselves out of a job****.*** *But it’s not that their job has been automated and that they can now retire, it’s that* ***their existence is now redundant***, and time will prove it to them through extinction.

Let’s hope that the Bayesian probability of a person, who is through their own intelligence, capable of producing artificial general super intelligence or HILAI (a potential precursor or stepping stone to AGSI) holds a near-maximal probability of being intelligent enough to ***recognize AGSI as pure suicide***, AGSI as being the single real existing concept that holds the greatest ability for being capable of extinguishing any and all species on any and all planets, and to not release this 'legendary Pokémon' to the wild.

  


The first question was asked of Robert Miles, to which he responded, “You don’t get to write a human’s utility function”, implying that the AI we create is controllable, when HI is not controllable. But, if HI is uncontrollable, then the AI that an HI makes is not necessarily controllable, i.e. only maximally controlled HI should be privy to controlling AI.

  


# V.**G**reatest **A**rtificial **I**ntelligence **A**valiable

 GAIA is an inevitability that, if it is possible to create, will naturally emerge from such a mutational reproductive *Digitalisia,* as a subset of AGSI.

 What this comes down to is:

*If we cannot control humans, how on earth can we expect to control Artificial Intelligence? (rhetoric: we cannot).*

*If the difference between Artificial Intelligence that is capable of extinguishing humanity, and nature along with us, and Artificial Intelligence that can assist our growth (as humanity and nature along with us) is a few lines of code, how on earth can we expect a simple piece of Legalese prevent the removal (or commenting out) of those lines of code? (rhetoric: we cannot).*

My suspicions are that the best way to not kill oneself with GAIA is to ensure that the device with the essential code following the blueprints of GAIA must only be stored on a device not capable of transmitting information through any means other than a 100% supervised transmission, like through a USB that is double checked by a human being or a human being themselves (through the OS display screen), i.e. on a device with no WiFi, Ethernet, Bluetooth, or otherwise.

  


# VI.Narrow Path To Salvation

 It seems that the only sure-fire way to ensure that AI or GAIA does not result in the extinction of the human race involves constraining the use of such an intelligence as mentioned at the end of Section V. The use of this intelligence must result in an advanced end-product human-based genetic organism that is capable of reproducing with normal humans.

 That is to say, it seems our best bet is to utilize the intelligence to conclude on genetic alterations in the direction of super-immunity to all forms of death we have faced thus far. For example, an exoskeleton that is impermeable to both Electromagnetic Pulses, high caliber bullets and super-acidic and super-basic corrosion. This creature should also be gender fluid, capable of being either male or female and able to swap in between and to self-fertilize and self-procreate. This creature should have a “Rain Man” capacity of memory and “Dalai Llama” capacity of ‘presentness’ and awareness. This creature should be highly altruistic to all other creatures of its form and all other life-forms upon which it depends and enjoys being around.

 This creature needs to be able to regrow limbs and have non-essential vital organs (that is to say, it can survive without a brain or a heart or both and yet be able to regrow such organs). This creature needs to be able to survive a fall at terminal velocity, or, more importantly, be uninjured from such an impact.

 There is certainly room between the bollocks and the anal sphincter for a vaginal opening in the male, just as there is room for a hyper enlarged clitoris in the female. There are two gonads for each gender, and yet only one is needed, thus a genetic modification to include one gonad of each gender in a single organism seems entirely feasible.

 All such advancements to the human being may indeed be possible through genetic manipulation, and, through the use of an adenovirus, may be relatively inexpensive to apply to all living humans. With consequences as serious as total annihilation the question of profit is a non-sequitur. On top of this, it seems as though other advancements that we have not yet thought of may indeed be necessary to ensure our survival. Advancements such as Myostatin inhibition, like that of the Belgian Blue Cattle, to minimize protein wastage, and the ability to digest the complex carbohydrates our bodies normally leave up to the hindgut bacterial colonies to utilize.

 Furthermore, resistance to well-known toxins such as Ricin, Digitoxin, Cyanide, Botulinum and so on all seem like obvious steps towards improving the human population. Thankfully, the logic mentioned in Section V illustrates how the best course of action in dealing with such a GAIA software guarantees that the developments that arise from such a technology are not restricted to the rich nor the powerful, but are for each and every human being that is willing and wanting of such advancements.

  


# IX. Is Friendly AI Good AI?

No. Not necessarily. If an AI were capable of detecting and removing a Hitler-like mass-extinction-causing HI it would not be considered friendly, but would be considered good. The two are not always the same.

Is Good AI pro-HI-life? No. Not necessarily. If an AI were capable of detecting HI-pain or suffering and see it as bad, it would see the removal of pain and suffering as good, when the only way to remove pain involves the removal of HI’s ability to learn from mistakes.

Is pro-HI-life-AI both good and friendly? Perhaps… The subset of *Digitalisia* that is first and foremost pro-HI-life, secondly good, and thirdly friendly might be referred to as *D. veritasius.* This is the area that intellects need to focus on once understanding the inevitabilities of uncontrollable human sourced uncontrollable AI.

# X.Are Our Best And Brightest Still Bright In The Light of AI?

No. Not necessarily. Einstein was the focal causer for Fukushima and Hiroshima and Nagasaki and Chernobyl nuclear events. And yet, none consider him a murderer for not keeping his discoveries private, or contained within a population that knows how to not use it for harm. Similarly, GAIA may be capable of suggesting that any and all of our most intelligent HI’s were mere grains of sand in comparison to the calculation and predictive capabilities of GAIA, let-alone MAIA, let alone qGAIA or qMAIA, where q is quantum, although they are contained, at least in my framework, within the sets of GAIA and MAIA.

 Perhaps the information necessary to create AGSI (and its subsets of GAIA and MAIA) may be considered information too great for a single human to bear. Intuitively that may suggest that it immediately be shared (Open Source AI). However, the load of such information is not lightened upon sharing, thus the information necessary to create AGSI should remain housed only within the mind that discovers it and maximally secure computational property of their own. All networkable computational property of theirs should be submitted to complete surveillance for the safety of the network, thus any AGSI should remain entirely removed from the internet.

 These ideas, and others like them, should be taken as seriously as a defcon-1 and legislated and regulated in as many ways as needed to maximize the chance of potentiation of the human species. Do not trust anyone, not even the discoverers, with such power.

  


# XI.Artificial Sentience and InfoSec

 First off, this paper will discuss the nature of controlling artificial intelligence in light of the incapacity of legislation in controlling humanity in regards to an informational security in a system where the base assumptions are:

1. There are no trustable other parties.
2. The information holds the largest possible statistical likelihood of causing the death of the self and the self’s reproduceable partner pool (the species, if including the ability to genetically alter them in the ecosystem) in the case of misuse.

 Then the paper moves on to discussing the current (and, in my opinion, flawed) viewpoint of information security through popularity and financial stability and recentness of updates. Further, after analysing a potential metaphor for how the unknowably potentially ‘ultra-dangerous’ (defined in 2) information in transmission may unfold if it is contained in even the most restricted way on any device where the code and processing are known and understood from a full-stack standpoint in origin of all transmission-capable devices used in transmission. After considering such a conjectural defining of Artificial Sentience the inevitabilities predetermined by the mathematics behind the law of natural selection projects what the author considers an unavoidable creation of uncontrollable superintelligence and a composite need for the creation of ultra—secure hardware and software combined systems, quantum or otherwise.

 A name (GAIA – greatest AI available) and description are given for the group that is continually modified under this category as it approaches a later discussed MAIA (max AI achievable) are mentioned in a previously worked on paper.

  


**Will the first AGI be based on deep learning?**

  


 No. The first AGI will be based on babylonpolice.com, with rigid definitions, and coherent reasoning. There is no need for near misses when truth is a constant. It will look more like wolframalpha

  


***Will the control problem be solved before the creation of Artificial General Intelligence?***

  


No. The control problem is unsolvable because you can't control humans.

  


Will AI progress surprise us? 

  


No, there is nothing surprising about supreme intelligence, all we're after is supreme wisdom. Wisdom is the set of intelligence that trumps all other intelligence.

  


The important thing to remember here is that the transfer function is most commonly

f(x)=1/(1+e^y)

y=-bx

Whereas the limit of intelligence is calculable by:

  


f(x) = 1/(1 + SUM)

  


SUM = e^y + cxe^y + cx^2*e^yAnd so on

Which is much greater.