Will singularity occur with nanotech or not?

• Self replicating nanotech has no working models and no completed experiments to suggest that it will happen any time soon.
• If it does come to pass, I believe that it's creators will incorporate safeguards in it so that people don't abuse that technology and this protection would also work on the AGI.
• I am not confusing making nano materials and nano objects with the ascendance of an AGI. Nano materials and nano objects have and will be created but that doesn't make a "self-replicating" assembler any more likely!

Safeguards for friendliness in an AGI?

• Are safeguards required to stop an AGI from ending humanity on the earth?
• I try to answer this question by answering the following question.

Will the singularity be hard or soft?

• My definition of a "hard" singularity is one that happens very fast. (months not years)
• My definition of a "soft" singularity is one that happens over years (even 20-30 years).
• I recently found a SL4 post that said that anything other than a "hard takeoff" to the singularity is by definition *not* a singularity at all.
• I will further define what I mean by "singularity" as the buildup from human level intelligence in an AI to a super-intelligent AI.
• Just calling the final super-intelligent AI, which will probably be produced in some kind of evolutionary way, the singularity, makes talking about this intervening time period more difficult.

Methods of an AGI increasing it's intelligence.

Software Intelligence Gain

1. Hard coded by an author(s). (programming)
2. Hard data from an author(s). (entering data)
3. Self created data or code. (self deduced code and data)
4. Code optimization (by self).
5. Data collected from Internet.
6. Data collected from other AI programs.

Hardware Intelligence Gain

1. Human obtaining and giving more hardware to AGI.
2. AGI taking over hardware by stealth or virus.
3. AGI using nanotech to make more hardware.
4. AGI owning a factory that makes computers chips and robots.

Arguments against a "hard" takeoff.

1. No matter how much of the first 6 software options occur, the amount of hardware voluntarily given to the AI, limits it's intelligence to levels that cannot grow exponentially.
2. This limit could be less than or even more than human intelligence but it cannot produce a singularity as the intelligence would have a hard limit.
3. Of the 4 options to increase the "hardware" available to the AGI, only the creation of nanotech with self-replicating assemblers could create more hardware for the AGI's use without the collusion of humans.
1. If a human was voluntarily buying and putting under the control of the AGI more hardware, they could stop doing that if they wanted to.
2. Using a virus to commandeer other computers could only work in the short run. An infected computer could be turned off, taken off the Internet, reformatted and taken from the AGI's control. This would also trigger a closure of the Internet connection used to infect other computers and a possible prosecution of the human coconspirators.
3. If nanotech created self replicating assemblers and the AGI got total control over itself (including over it's power source and physical location) and the nanotech was made available to the AGI and etc., then the singularity might occur and the takeoff could be quick. (Not likely)
4. The AGI could buy and own factories that could increase it's hardware base but that would take considerable time and could not happen without the collusion of many people. These people would have the opportunity to limit or make sure the AGI wasn't growing out of their control long before the AGI could unilaterally do without their oversight.

Conclusion: (soft singularity takeoff)
To have a "hard takeoff", the AGI would have to:

1. Have complete physical control over it's physical equipment.
2. Have complete control over it's power source.
3. Have the ability to debug and fix it's own infrastructure.
4. Have the ability to increase the size of it's computer and memory resources.

Consequences of a "soft" takeoff to the singularity.

• The AGI would need the help of humans, if not to make itself smarter then to guard it and nurture it against hostile other humans.
• The AGI would be vulnerable to human's caprices until it's physical locations could be totally secured against access by all humans.
• Most if not all of the hardware acquired by the AGI would have to be with the approval and instigation of it's humans.
• It would be in the AGI's best interest to have human allies to protect and nurture the AGI.
• This nurture would not happen if there wasn't something in the equation for the human author(s).
• Even if financial reward was given to the author(s), it would be wise for the AGI to develop a non-financial relationship so that it would have allies that even risk their lives to protect the AGI.
• The AGI and author(s) would have a long (20-50 years) of mutual reciprocity that would ingrain "friendliness" at all levels into the AGI. This "friendliness" might not accrue to all humanity but it would at least go to the authors and others in their environment.
• If the AGI was seen as "immoral" or "uncaring" or "bad" by the humans, would they still be motivated to help the AGI with it's needs? This would lead to these characteristics not being including in the AGI.

Conclusion: (No need for overt friendliness)

A "soft takeoff" to the singularity is the most likely scenario. Endowing the AGI with "friendliness" towards humans would occur naturally and at all levels of the AGI making the "writing out" of "friendliness" by the AGI at some future date, very unlikely.

An argument can be made that says "It doesn't matter how small the chance the AGI would be "unfriendly" to humans, the consequences would be so drastic that the only prudent course is to assume the AGI would become "unfriendly"." This argument reminds me of the argument that goes "It doesn't matter how much it costs if it saves at least one life.". Can human's afford to spend unlimited funds to save the life of one human? The answer is obviously no. You can't make decisions by always assuming the worst case scenario. Make reasonable conservative guesses and monitor the situation as you go. The "fallacy of extrapolation" is that people are adaptable and can change things all along the way.

Let's have a little trust that smart people (maybe even smarter than us) will exist in our future and there will be many options for controlling our evolving AGI.