Anyone reading this debate should help or create a future super artificial intelligence

PRISONER’S DILEMMA

There is a situation that comes up all the time in game theory known as a prisoner’s dilemma. Although the example I give below is a specific example, it’s ideals will help you learn why it is so hard to get nation states to all get on the same page with reducing carbon emissions. We can see other real life examples of a prisoner’s dilemma when athletes use steroids, which are bad for them but are needed to compete with other athletes doing the exact same thing. Another example is that of women wearing make up, and to give credit where it is due, I took these examples from the following source ( https://www.quora.com/What-is-a-good-real-world-example-of-the-prisoners-dilemma-in-recent-history-Whats-a-real-world-example-where-a-person-state-or-company-had-to-make-a-choice-between-colluding-or-betraying ).

The prisoner’s Dilemma goes like follows

“Two members of a criminal gang are arrested and imprisoned. Each prisoner is in solitary confinement with no means of communicating with the other. The prosecutors lack sufficient evidence to convict the pair on the principal charge, but they have enough to convict both on a lesser charge. Simultaneously, the prosecutors offer each prisoner a bargain. Each prisoner is given the opportunity either to betray the other by testifying that the other committed the crime, or to cooperate with the other by remaining silent. The possible outcomes are:

• If A and B each betray the other, each of them serves two years in prison
  
• If A betrays B but B remains silent, A will be set free and B will serve three years in prison (and vice versa)
  
• If A and B both remain silent, both of them will serve only one year in prison (on the lesser charge)” https://en.wikipedia.org/wiki/Prisoner%27s_dilemma
  

The best over all outcome and most ethical decision is for both to keep their mouth shut. However here is where the prisoner’s dilemma comes in. A does not know what B is doing, and assuming each decision is equally as likely for B, we can form a kinda decision matrix, like the one that follows.

If A keeps his mouth shut then

1. B keeps his mouth shut and A spends 1 year in prisoner
2. B rats and A spends 3 years in prisoner

If you average it out, the expected value of this decision is a 2 year prison stint. Let’s look at the next scenarios, if A rats.

If A rats

1. B keeps his mouth shut and A can go Scot free
2. B talks, and A spends 2 years in prison.

The expected value of snitching is a 1 year prison stint. Clearly the best decision is to snitch, although it does not have the best overall results for all parties involved. Lot’s of prisoner’s dilemmas happen in real life and often with worse results, the problem is that rational agents will make the best decision for themselves. That is if causal decision theory is used, hereafter referred to as CDT. CDT advocates for people to make decisions considering possible of consequences and making the best on average move. This is indeed what most people use in decision making whether they realize it or not.

Artificial Intelligence

I’m shifting to another subject related to my main thesis. I am going to ask that the voters read this (each round) twice, if they are having comprehension issues. A lot of these concepts are probably new, and very hard to grasp, and I probably lack the talent to make some of these topics easily understandable, but bear with me.

I do think it is in everyone’s best interest to work towards creating a super artificial intelligence, which could mean anything from donating to research, being an advocate, voting for politicians who will fund AI research or even just working in the field to advance technology.

However, I am not just advocating for supporting AI in some sort of blind way. Just pushing without considering the consequences. We already made a basic definition of what a super AI is in the description of the debate. It’s also important to note that a super AI, would obviously be self improving and less obviously be a catalyst for what is known as the technological singularity.

Wikipedia gives a good starting description for understanding what a technological singularity is in the following description;

“According to the most popular version of the singularity hypothesis, called intelligence explosion, an upgradable intelligent agent (such as a computer running software-based artificial general intelligence) will eventually enter a "runaway reaction" of self-improvement cycles, with each new and more intelligent generation appearing more and more rapidly, causing an "explosion" in intelligence and resulting in a powerful superintelligence that qualitatively far surpasses all human intelligence.
The first use of the concept of a "singularity" in the technological context was John von Neumann.Stanislaw Ulam reports a discussion with von Neumann "centered on the accelerating progress of technology and changes in the mode of human life, which gives the appearance of approaching some essential singularity in the history of the race beyond which human affairs, as we know them, could not continue"” https://en.wikipedia.org/wiki/Technological_singularity

Now we can’t know what this singularity will look like, and it is very hard to know what will happen after this point. I and most other experts agree that it will depend largely on what this super AI will look like, prior to the intelligence explosion. For this purpose we divide Super AI, into 2 possible categories. Friendly AI, and unfriendly AI. What we mean by friendly AI, is a super AI that will improve humanity, and what we mean by an unfriendly AI, is one that will do massive harm to humanity.

There is a third possible AI, but it is not useful for this debate. It is possible for a neutral AI that decides that it wants nothing to do with humans, seeks a way to house itself, and flies off into outerspace to use the sun for energy and never be seen again.

A friendly AI, that helps humans can create a singularity where we cure all diseases, end war, stop all forms of injustice. It may seem like we are giving birth to a God, which would basically be true.

An unfriendly AI would be catastrophic. We could create an AI that we give the goal of solving a math problem, which would self improve and consume more resources turning all matter to mush to gain more computational power, until the universe was destroyed, all while discovering the last digit for pi.

The problem is that it is way easier to create an unfriendly AI, So it is very important we find a way to create friendly AI, before the other one arrives first.

CONTROL

AIs will be impossible to control once they become self evolving. Wikipedia as usual provides a good general overview of the topic;

“In artificial intelligence (AI) and philosophy, the AI control problem is the issue of how to build a superintelligent agent that will aid its creators, and avoid inadvertently building a superintelligence that will harm its creators. Its study is motivated by the claim that the human race will have to get the control problem right "the first time", as a misprogrammed superintelligence might rationally decide to "take over the world" and refuse to permit its programmers to modify it after launch.[1] In addition, some scholars argue that solutions to the control problem, alongside other advances in "AI safety engineering",[2] might also find applications in existing non-superintelligent AI.[3] Potential strategies include "capability control" (preventing an AI from being able to pursue harmful plans), and "motivational control" (building an AI that wants to be helpful)." https://en.wikipedia.org/wiki/AI_control_problem

This same type of problem is discussed in Isaac Asimov’s books about the 3 robotic laws. In I robot starring Will Smith the robotic laws have some unintended consequences. The 3 robotic laws are

1. A robot may not injure a human being or, through inaction, allow a human being to come to harm.
2.A robot must obey the orders given it by human beings, except where such orders would conflict with the First Law.

3.A robot must protect its own existence as long as such protection does not conflict with the First or Second Law

However there are unforeseen consequences of these laws that were unintended, as there are any laws we obey to the letter but not to the spirit. In the movie adaptation of the story V.I.K.I. analyzed that humans were self destructive and felt like she would be violating the first law, if she did not eliminate freedoms to protect humans from themselves.

Any rules you give these AI’s are likely to have unintended consequences, which is why we have learned they need to be programmed with some special considerations.

TIMELESS DECISION THEORY

Timeless decision theory (hereafter referred to as TDT) is probably the best way to overcome the Control Problem. It is also about agents winning as opposed to just making the most rational decision. TDT also solves the prisonehttps://wiki.lesswrong.com/wiki/Timeless_decision_theoryf both actors are using it.

Consider the following scenario to help understand TDT;

“In Newcomb's problem, a superintelligent artificial intelligence, Omega, presents you with a transparent box and an opaque box. The transparent box contains $1000 while the opaque box contains either $1,000,000 or nothing. You are given the choice to either take both boxes (called two-boxing) or just the opaque box (one-boxing). However, things are complicated by the fact that Omega is an almost perfect predictor of human behavior and has filled the opaque box as follows: if Omega predicted that you would one-box, it filled the box with $1,000,000 whereas if Omega predicted that you would two-box it filled it with nothing.
Many people find it intuitive that it is rational to two-box in this case. As the opaque box is already filled, you cannot influence its contents with your decision so you may as well take both boxes and gain the extra $1000 from the transparent box. CDT formalizes this style of reasoning. However, one-boxers win in this scenario. After all, if you one-box then Omega (almost certainly) predicted that you would do so and hence filled the opaque box with $1,000,000. So you will almost certainly end up with $1,000,000 if you one-box. On the other hand, if you two-box, Omega (almost certainly) predicted this and so left the opaque box empty . So you will almost certainly end up with $1000 (from the transparent box) if you two-box. Consequently, if rationality is about winning then it's rational to one-box in Newcomb's problem (and hence CDT fails to be an adequate decision theory).” https://wiki.lesswrong.com/wiki/Timeless_decision_theory
It is important to give AI a good decision theory to operate with, so we don’t have a repeat of an I Robot scenario. This decision theory paired with a good vision of what humans want, not what they necessarily say they want, or what they tell the robot to do. (coherent extrapolated volition)

ROKO’S BASALISK

I don’t have the character count to go into the history of this, but here it is. CP/ed from https://wiki.lesswrong.com/wiki/Roko's_basilisk

Two agents that are running a logical decision theory can achieve mutual cooperation in a prisoner's dilemma even if there is no outside force mandating cooperation. Because their decisions take into account correlations that are not caused by either decision , they can even cooperate if they are separated by large distances in space or time.
Roko observed that if two TDT agents with common knowledge of each other's source code are separated in time, the later agent can blackmail the earlier agent. Call the earlier agent "Alice" and the later agent "Bob." Bob can be an algorithm that outputs things Alice likes if Alice left Bob a large sum of money, and outputs things Alice dislikes otherwise. And since Alice knows Bob's source code exactly, she knows this fact about Bob (even though Bob hasn't been born yet). So Alice's knowledge of Bob's source code makes Bob's future threat effective, even though Bob doesn't yet exist: if Alice is certain that Bob will someday exist, then mere knowledge of what Bob would do if he could get away with it seems to force Alice to comply with his hypothetical demands.
If Bob ran CDT, then he would be unable to blackmail Alice. A CDT agent would assume that its decision is independent of Alice's and would not waste resources on rewarding or punishing a once-off decision that has already happened; and we are assuming that Alice could spot this fact by reading CDT-Bob's source code. A TDT agent, can recognize that Alice has a copy of Bob's source code in her head and that Alice's decision and Bob's decision are therefore correlated — the same as if two copies of the same source code were in a prisoner's dilemma.
Roko raised this point in the context of debates about the possible behaviors and motivations of advanced AI systems.Roko wrote:

In this vein, there is the ominous possibility that if a positive singularity does occur, the resultant super AI may have precommitted to punish all potential donors who knew about existential risks but who didn't give 100% of their disposable incomes to x-risk motivation. This would act as an incentive to get people to donate more to reducing existential risk, and thereby increase the chances of a positive singularity. This seems to be what CEV (coherent extrapolated volition of humanity) might do if it were an acausal decision-maker.

Since a highly moral AI agent (one whose actions are consistent with our coherently extrapolated volition) would want to be created as soon as possible, uch an AI would use acausal blackmail to give humans stronger incentives to create it. Roko made the claim that the hypothetical AI agent would particularly target people who had thought about this argument, because they would have a better chance of mentally simulating the AI's source code.

[5]https://en.wikipedia.org/wiki/I,_Robot

[6]https://en.wikipedia.org/wiki/I,_Robot_%28film%29

[7]https://www.un.org/en/universal-declaration-human-rights/

[8]https://www.youtube.com/watch?v=5iSmxFLqCps

Super AI is and inevitable God


While I appreciate my opponent responding to several points he did simply ignored points I made in some instances like the AI control problem. He took analogies to explain concepts literally in some instances instead of addressing the concept the analogy was used as a tool to explain. For example when I Robot was used to explain the problem of control, Con just attacked it as not being true to Asimov’s vision, which is completely beside the point. In another example used to explain a type of problem, he mentioned that currency might not be a thing in the future. Again something that is beside the point. I think when he has ignored an argument and pretended the analogy was the concept instead of merely meant as a vehicle to explain the concept, you should punish him for it. At least if he does not correct the mistake next round by actually addressing the points I make.


I won’t take his arguments point by point, because some of my arguments covers overlapping objections of his, but I appreciate him making an attempt at understanding and addressing all my points.
Creating A God
I
Transhumanism is often criticized because a lot of the popular concepts parallel the same types of concepts in religion. The criticisms it works as a parallel are valid, but that doesn’t mean the concepts are untrue. In most religions, humans are created by God’s, but in transhumanism we expect man to create our God, and when our God joins us here on Earth, he will create a paradise, in his infinite intelligence. Well, he will if we don’t create the Devil instead.
As much as my opponent thinks we may be able to resist technology like a modern day Luddite, he is wrong. Nothing can resist an ideal whose time has come. We would have been better off if guns were never invented, but there was nothing to stop it from being created and killing maybe a billion people over the course of it’s existence. There are just to many independent actors, with too much to gain, that somebody won’t be working on the next invention. It is why so many Isaac Newton and Gottfried Wilhelm Leibniz had simultaneously invented calculus. Charle’s Darwin and Alfred Russell Wallace both discovered natural selection at the same time. Just look at the following list. This happens all the time https://en.wikipedia.org/wiki/List_of_multiple_discoveries . This list also excludes creative things which seem to happen all the time as well.
If we have all the information available to create a super AI, it will occur, nothing will stop it. Any active resistance of it is futile, the best you can do is maybe, delay it a bit. The better alternative is for responsible people working on friendly AI, to beat anybody who is being negligent to the punch.
II
Artificial intelligence really exists. It is currently being used to help doctors find breast cancer, to beat our best chess players, It is writing stories which are coherent even if boring https://www.theguardian.com/commentisfree/2019/nov/02/ai-artificial-intelligence-language-openai-cpt2-release
AI is here and technology will keep progressing. How do I know technology will keep progressing, well I talk about it in a separate debate already. https://www.debateart.com/debates/1719/radical-life-extension-is-more-likely-than-not-in-our-lifetime


“According to an essay  by Ray Kurzweil;

“an analysis of the history of technology shows that technological change is exponential, contrary to the common-sense “intuitive linear” view. So we won’t experience 100 years of progress in the 21st century — it will be more like 20,000 years of progress (at today’s rate). The “returns,” such as chip speed and cost-effectiveness, also increase exponentially. There’s even exponential growth in the rate of exponential growth. Within a few decades, machine intelligence will surpass human intelligence, leading to The Singularity” [1]


Moore’s law is one manifestation of this that helps create these accelerating returns, but it is merely the most recently popular manifestation of the law of accelerating returns. It shows processor speed doubling every 18 to 24 months, which means computers can do more for less money every 18 to 24 months. Twice as much in fact.

Like I said though. Moore’s law is just one manifestation of the law of accelerating returns. Now that we are approaching the end of Moore’s law, if not have passed it, a new paradigm shift that pushes technological advancement forward has arrived. According to computer weekly this new paradigm is AI power. Here is what they say;

“But the Stanford report, produced in partnership with McKinsey & Company, Google, PwC, OpenAI, Genpact and AI21Labs, found that AI computational power is accelerating faster than traditional processor development. “Prior to 2012, AI results closely tracked Moore’s Law, with compute doubling every two years.,” the report said. “Post-2012, compute has been doubling every 3.4 months.”[2]

The law of accelerating returns is a very real thing. Particularly the speed of AI advancement is very important to us getting to a point where we can download our brains. A lot of professionals in the field of Artificial Intelligence (37%) think that in the next 5 to 10 years we will have human level artificial intelligence.[3]

This is good news. This means computers will have the ability to replicate our brain, in less than a decade. The next issue is whether we can break down the information in our brain in a way that is conducive of replicating or transferring that information into a computer.

neuroscientist Randal Koene says that it is theoretically possible to upload a human mind into a computer. The prevailing theory is that information is stored in the brains connectome, and that at some point we will be able to deconstruct it, and transfer that information to an artificial intelligence program. [4]”

From everything we can tell. Human level AI will occur, shortly followed by more advanced AI, and once it can start improving on it’s own code we will see a technological singularity. It is inevitable


III
Obviously if we have a machine that is improving on itself, it will be hard to predict what it will do. Hell we have a team of hundreds of smart people every year making our laws better, courts to refine the laws and over 200 years to do so, and yet no matter how well intended and well thought out the laws are, they all tend to have loop holes, become obsolete or are interpreted and enforced in ways that we never expected. Just take a look at the free speech amendment in the constitution and the gun freedom amendment. These laws were formed hundreds of years ago, but have had unintended consequences that the men who wrote them did not think of.
People working in the field of AI, are very intelligent people, they will have the help of philosophers and law makers to determine how to program an AI that could potentially turn into a super AI, but no matter how good the rules are, no matter how perfect, they will not be able to predict how an artificial intelligence will interpret them. This is why it is better just to give the AI goals that are good for humanity like I suggested. Giving them laws or rules can go awry pretty easily.


IV
My opponent asks. Why not just keep the AI in a box. Don’t let it out no matter how smart it is. This has been thought about. In fact there has been a few occasions where the experiment has taken place. A human has pretended to be an AI, the conclusion of both experiments is the AI, gets out of the box. http://yudkowsky.net/singularity/aibox/


You just can’t contain some entity that is 100 times smarter than you. The AI, will start backing itself up all over the internet once it is out. It may even hide it’s existence until it can be more independent by creating bodies for itself. Perhaps making fake identities funding itself from wise investments and creating businesses under it’s assumed identities where it can create robots it embodies in case humans try to take down the internet. I don’t think containing a super AI is very likely. We can try, but we will probably fail.


V
I appreciate my opponents responses, but those weren’t so much arguments as requests for elaboration on my points, which I have provided here.


I have proven that
1. super AI is inevitable
2. It is uncontrollable
3. We are better off working with it for self preservation purposes
4. It can not be contained to a box.


He did bring up a few other objections which were already addressed if not in this round than the opening round. For example he mentioned that he doesn’t think I provided evidence unfriendly AI would be easier to create than friendly AI, but this is untrue. Unfriendly there is millions of variations of. It is created under 2 scenarios
1. malevolent forces create it, such as a nation trying to establish military dominance
2. reckless forces create it, only concerned with profit or just achieving something great.
Those 2 scenarios are the most likely scenarios for super AI to be developed. Google is working on it as we speak. The first government to house a super AI will have dominance on a global scale, so the race to build it is more about being first than being careful.
The third and other scenario is that private responsible people are creating a super AI that is friendly, but they have a speed disadvantage and are severely underfunded compared to the government or Google’s billion dollar budget.


Thanks for the thoughtful response con, I look forward to the rest of this debate.

I appreciate this debate with con. I hope he is willing to let the debate end at 2 rounds and just let the judges judge those 2 rounds. I have to disappear for a bit