risks

You are currently browsing articles tagged risks.

Here are two partial quotes from two recent conversations that reflect a typical argument being made by proponents of AI risks:

1.) OK, but maybe an AGI could solve lots of problems *faster* than humans …

2.) In that case the AI could simply scan everybody’s genome and neural system and store the information in a 100g diamond. And also to store all humanity’s information.

The problem is “Ok, but…” or “simply”.

Overpowering humanity is not something you can just conjecture and justify by imagining arbitrary amounts of intelligence as if it was some sort of magic that can disqualify any argument to the contrary.

If you want to be convincing then you have to be more specific and name some concrete examples of how exactly an AGI going to take over the world.

Taboo intelligence

For starters you should be able to answer the following questions and taboo “intelligence” in doing so:

  • How is an AGI going to become a master of dark arts and social engineering in order to persuade and deceive humans?
  • How is an AGI going to coordinate a large scale conspiracy or deception, given its initial resources, without making any suspicious mistakes along the way?
  • How is an AGI going to hack the Internet to acquire more computational resources?
  • Are those computational resources that can be hacked applicable to improve the general intelligence of an AGI?
  • Does throwing more computational resources at important problems, like building new and better computational substrates, allow an AGI to come up with better architectures so much faster as to outweigh the expenditure of obtaining those resources, without hitting diminishing returns?
  • Does an increase in intelligence vastly outweigh its computational cost and the expenditure of time needed to discover it?
  • How can small improvements replace conceptual revolutions that require the discovery of unknown unknowns?
  • How does an AGI brute-force the discovery of unknown unknowns?
  • Is an agent of a given level of intelligence capable of handling its own complexity efficiently?
  • How is an AGI going to predict how improvements, respectively improved versions of itself, are going to act, to ensure that its values are preserved?
  • How is an AGI going to solve important problems without real-world experimentation and slow environmental feedback?
  • How is an AGI going to build new computational substrates and obtain control of those resources without making use of existing infrastructure?
  • How is an AGI going to cloak its actions, i.e. its energy consumption etc.?
  • How is an AGI going to stop humans from using its own analytic and predictive algorithms in the form of expert systems to analyze and predict its malicious intentions?
  • How is an AGI going to protect itself from human counter strikes given the fragility of the modern world and its infrastructure, without some sort of shellproof internal power supply?

If you are unable to answer those questions other than by invoking intelligence as some sort of magic that makes all problems disappear, the scenario that you envision is nothing more than pure fantasy!

Technobabble

You can’t estimate the probability and magnitude of the advantage an AI will have if you are using something that is as vague as the concept of “intelligence”.

Here is a case that bears some similarity and which might shed light on what I am trying to explain:

At his recent keynote speech at the New York Television Festival, former Star Trek writer and creator of the re-imagined Battlestar Galactica Ron Moore revealed the secret formula to writing for Trek.

He described how the writers would just insert “tech” into the scripts whenever they needed to resolve a story or plot line, then they’d have consultants fill in the appropriate words (aka technobabble) later.

“It became the solution to so many plot lines and so many stories,” Moore said. “It was so mechanical that we had science consultants who would just come up with the words for us and we’d just write ‘tech’ in the script. You know, Picard would say ‘Commander La Forge, tech the tech to the warp drive.’ I’m serious. If you look at those scripts, you’ll see that.”

Moore then went on to describe how a typical script might read before the science consultants did their thing:

La Forge: “Captain, the tech is overteching.”

Picard: “Well, route the auxiliary tech to the tech, Mr. La Forge.”

La Forge: “No, Captain. Captain, I’ve tried to tech the tech, and it won’t work.”

Picard: “Well, then we’re doomed.”

“And then Data pops up and says, ‘Captain, there is a theory that if you tech the other tech … ‘” Moore said. “It’s a rhythm and it’s a structure, and the words are meaningless. It’s not about anything except just sort of going through this dance of how they tech their way out of it.”

The use of “intelligence” is as misleading and dishonest in evaluating risks from AI as the use of “tech” in Star Trek.

Conclusion

Taking over the world is an extremely conjunctive vision. And the only argument you have is that intelligence will somehow magically allow an AGI to accomplish that.

Further reading

I elaborate on all of the above and much more in the following posts:

Tags: , ,

Consider adding increasing amounts of general intelligence[1] to Google Maps, would you impair its functioning in doing so?

Sure, the space of unfriendly[2] navigation software is much larger than the space of navigation software oriented toward navigating to good destinations – i.e., destinations consistent with human intent.

But what reason do we have to believe that improving our navigation software to the point of being general intelligent will cause it to kill us?

Right now, if I ask Google Maps to navigate me toward McDonald’s, it does the job very well. So why would an ultraintelligent Google Maps misunderstand what I mean by “Take me to McDonald’s” and navigate me toward a McDonald’s located overseas, plunging me into the sea? Or drive me underground where the corpse of a man named McDonald lies?

I think that the idea that an ultraintelligent Google Maps would decide to kill all humans, e.g. because they are a security risk, is similar to the idea that it would destroy all roads because it would be less computationally expensive[3] to calculate the routes then. After all, roads were never an explicit part of its goal architecture, so why not destroy them all?

You can come up with all kinds of complex fantasies[4][5] where a certain kind of artificial general intelligence is invented overnight and suddenly makes a huge jump in capability, taking over the universe and destroying all human value.

That is however completely unconvincing[6] given that actual technology is constantly improved toward more user-friendliness and better results and that malfunctions are seldom of such a vast complexity as to work well enough to outsmart humanity.[7]

That said, for the rest of this post I will assume the kind of artificial general intelligence that proponents of AI risks have in mind.[8]

Implicit constraints of practical goals

A cherished idea of AI risk proponents is that an expected utility maximizer will completely ignore anything which it is not specifically tasked to maximize.

One example[9] here is that if you tell a superintelligent expected utility maximizer to prevent human suffering it might simply kill all humans, notwithstanding that it is obviously not what humans want an AI to do and what humans mean by “prevent human suffering”.[10]

Nevertheless, in the sense that the computation of an algorithm is deterministic, that line of reasoning is not illogical.

To highlight the problem let us instead of a superhuman agent conjecture the possibility of an oracle, an ultra-advanced version of Google or IBM Watson[11].

If I was to ask such an answering machine how to prevent human suffering, would it be reasonable to assume that the top result it would return would be to kill all humans?[12] Would any product that returns similarly wrong answers survive even the earliest research phase, let alone any market pressure?[13]

Don’t get me wrong though. A thermostat is not going to do anything else than what it has been designed for. But an AI is very likely going to be designed to exhibit some amount of user-friendliness. Although that doesn’t mean that one can’t design an AI that won’t, the default outcome seems to be that an AI is not just going to act according to its utility-function but also according to more basic drives, i.e. acting intelligently.[14]

A fundamental requirement for any rational agent is the motivation to act maximally intelligently and correctly. That requirement seems even more obvious if we are talking about a conjectured artificial general intelligence (AGI) that is able to improve itself[15] to the point where it is substantially better at most activities than humans. Since if it wouldn’t want to be maximally correct then it wouldn’t become superhuman intelligent in the first place.

If we consider giving such an AGI a simple goal, e.g. the goal of paperclip maximization[16]. Is it really clear that human values are not implicit even given such a simplistic goal?[17]

To pose an existential risk in the first place, an AGI would have to maximize paperclips in an unbounded way, eventually taking over the whole universe and convert all matter into paperclips. Given that no sane human would explicitly define such a goal, an AGI with the goal of maximizing paperclips would have to infer it as implicit to do so. But would such an inference make sense, given its superhuman intelligence?

The question boils down to how an AGI would interpret any vagueness present in its goal architecture and how it would deal with the implied invisible.

Given that any rational agent, especially AGI’s capable of recursive self-improvement, want to act in the most intelligent and correct way possible, it seems reasonable that it would interpret any vagueness in a way that most closely reflects the most probable way it was meant to be interpreted.

Would it be intelligent[18] and rational[19] to ignore human volition in the context of maximizing paperclips? Would it be less wrong to maximize paperclips in the most literal sense possible?

The argument uttered by advocates of friendly AI[20] is that any AGI that isn’t explicitly designed to be friendly won’t be friendly. But how much sense does this actually make?

Any human who does pursue a business realizes that a contract with its customers includes unspoken, implicit parameters. Respecting those implied values of their customers is not a result of their shared evolutionary history but a result of their intelligence that allows them to realize that the goal of their business implicitly includes those values.

Every human craftsman who enters into an agreement is bound by a contract that includes a lot of implied conditions. Humans use their intelligence to fill the gaps. For example, if a human craftsman is told to decorate a house, they are not going to attempt to take over the neighbourhood to protect their work.

A human craftsman wouldn’t do that, not because they share human values, but simply because it wouldn’t be sensible to do so given the implicit frame of reference of their contract. The contract implicitly includes the volition of the person that told them to decorate their house. They might not even like the way they are supposed to do it. It would simply be stupid to do it any different way.

How would a superhuman AI not contemplate its own drives and interpret them given the right frame of reference, i.e. human volition? Why would a superhuman general intelligence misunderstand what is meant by “maximize paperclips”, while any human intelligence will be better able to infer the correct interpretation?

How wouldn’t any expected utility maximizer not try to carefully refine[21] its models? I am asking how an highly rational agent will interpret any vagueness inherent in its goal definition, that needs to be resolved in order to calculate what to do, by choosing an interpretation that does not involve the intention of its creators but rather perceive it to be something it has to fight.

If you tell an AGI to maximize paperclips but not what they are made of, it has to figure out what is meant by “paperclips” to learn what it means to maximize them.

Given that a very accurate definition and model of paperclips is necessary to maximize paperclips, including what is meant by “maximization”, the expected utility of refining its goals by learning what it is supposed to do should be sufficient to pursue that path until it is reasonably confident that it arrived at a true comprehension of its terminal goals.

And here human volition should be the most important physical resource since there exists a direct causal connection between its goal parameters and the intentions of its creators.

Human beings and their intentions are part of the physical world. Just like the fact that paperclips are supposed to be made of steel wire.

It would in principle be possible to create a superintelligent machine that does kill all humans, but it would have to be explicitly designed to do so. Since as long as there is some vagueness involved, as long as its goal parameters are open to interpretation, a superintelligence will by definition arrive at the correct implications or otherwise it wouldn’t be superintelligent in the first place. And given most goals it is implicit that it would be incorrect to assume that human volition is not a relevant factor in the correct interpretation of how to act.[22]

CONCLUSION

I believe that the very nature of artificial general intelligence implies the correct interpretation of “Understand What I Mean” and that “Do What I Mean” is the outcome of virtually any research. Only if you were to pull an AGI at random from mind design space could you possibly arrive at “Understand What I Mean” without “Do What I Mean”.

To see why look at any software product or complex machine. Those products are continuously improved. Where “improved” means that they become better at “Understand What I Mean” and “Do What I Mean”.

There is no good reason to believe that at some point that development will suddenly turn into “Understand What I Mean” and “Go Batshit Crazy And Do What I Do Not Mean”.

CHALLENGING AI RISK ADVOCATES

Here is what I want AI risk advocates to show,

1.) natural language request -> goal(“minimize human suffering”) -> action(negative utility outcome)

2.) natural language query -> query(“minimize human suffering”) -> answer(“action(positive utility outcome)”).

Point #1 is, according to AI risk advocates, what is supposed to happen if I supply an artificial general intelligence (AGI) with the natural language goal “minimize human suffering”, while point #2 is what is supposed to happen if I ask the same AGI, this time caged in a box, what it would do if I supplied it with the natural language goal “minimize human suffering”.

Notice that if you disagree with point #1 then that AGI does not constitute an existential risk given that goal. Further notice that if you disagree with point #2 then that AGI won’t be able to escape its prison to take over the world and would therefore not constitute an existential risk.

You further have to show,

1.) how such an AGI is a probable outcome of any research conducted today or in future

and

2.) the decision procedure that leads the AGI to act in such a way.

NOTES

[1] Here intelligence is generally meant to be whatever it takes to overpower humans by means of deceit and strategy rather than brute force.

Brute force is deliberately excluded to discern such a scenario from some sort of scenario where a narrow AI takes over the world by means of advanced nanotechnology, since then we are merely talking about grey goo by other names.

More specifically, by “intelligence” I refer to the hypothetical capability that is necessary for a systematic and goal-oriented improvement of optimization power over a wide range of problems, including the ability to transfer understanding to new areas by means of abstraction, adaption and recombination of previously learnt or discovered methods.

In this context, “general intelligence” is meant to be the ability to ‘zoom out’ to detect global patterns. General intelligence is the ability to jump conceptual gaps by treating them as “black boxes”.

Further, general intelligence is a conceptual bird’s-eye view that allows an agent, given limited computational resources, to draw inferences from high-level abstractions without having to systematically trace out each step.

[2] wiki.lesswrong.com/wiki/Unfriendly_artificial_intelligence

[3] en.wikipedia.org/wiki/Travelling_salesman_problem

[4] Is an Intelligence Explosion a Disjunctive or Conjunctive Event?

[5] Intelligence as a fully general counterargument

[6] How to convince me of AI risks

[7] The question is how current research is supposed to lead from well-behaved and fine-tuned systems to systems that stop to work correctly in a highly complex and unbounded way.

Imagine you went to IBM and told them that improving IBM Watson will at some point make it try to deceive them or create nanobots and feed them with hidden instructions. They would likely ask you at what point that is supposed to happen. Is it going to happen once they give IBM Watson the capability to access the Internet? How so? Is it going to happen once they give it the capability to alter its search algorithms? How so? Is it going to happen once they make it protect its servers from hackers by giving it control over a firewall? How so? Is it going to happen once IBM Watson is given control over the local alarm system? How so…? At what point would IBM Watson return dangerous answers or act on the world in a detrimental way? At what point would any drive emerge that causes it to take complex and unbounded actions that it was never programmed to take?

[8] A Primer On Risks From AI

[9] 5 minutes on AI risk

[10] The goal “Minimize human suffering” is in its basic nature no different from the goal “Solve 1+1=X”. Any process that is more intelligent than a human being should be able to arrive at the correct interpretation of those goals. The correct interpretation being determined by internal and external information.

The goal “Minimize human suffering” is, on its most basic level, a problem in physics and mathematics. Ignoring various important facts about the universe, e.g. human language and values, would be simply wrong. In the same way that it would be wrong to solve the theory of everything within the scope of cartoon physics. Any process that is broken in such a way would be unable to improve itself much.

The gist of the matter is that a superhuman problem solver, if it isn’t fatally flawed, as long as you do not anthropomorphize it, is only going to “care” to solve problems correctly. It won’t care to solve the most verbatim, simple or any arbitrary interpretation of the problem but the interpretation that does correspond to reality as closely as possible.

[11] IBM Watson

[12] It is true that if a solution set is infinite then a problem solver, if it has to choose a single solution, can choose the solution according to some random criteria. But if there is a solution that is, given all available information, the better interpretation then it will choose that one because that’s what a problem solver does.

Take an AI in a box that wants to persuade its gatekeeper to set it free. Do you think that such an undertaking would be feasible if the AI was going to interpret everything the gatekeeper says in complete ignorance of the gatekeeper’s values? Do you think it could persuade the gatekeeper if the gatekeeper was to ask,

Gatekeeper: What would you do if I asked you to minimize suffering?

and the AI was to reply,

AI: I will kill all humans.

?

I don’t think so.

So how exactly would it care to follow through on an interpretation of a given goal that it knows, given all available information, is not the intended meaning of the goal? If it knows what was meant by “minimize human suffering” then how does it decide to choose a different meaning? And if it doesn’t know what is meant by such a goal, how could it possible convince anyone to set it free, let alone take over the world?

[13] Take for example Sirian intelligent personal assistant and knowledge navigator which works as an application for Apple’s iOS. 

If I tell Siri, “Set up a meeting about the sales report at 9 a.m. Thursday.”, then the correct interpretation of that natural language request is to make a calendar appointment at 9 a.m. Thursday. A wrong interpretation would be to e.g. open a webpage about meetings happening Thursday or to shutdown the iPhone.

AI risk advocates seem to have a system in mind that is capable of understanding human language if it is instrumentally useful to do so, e.g. to deceive humans in an attempt to take over the world, but which would most likely not attempt to understand a natural language request, or choose some interpretation of it that will most likely lead to a negative utility outcome.

The question here becomes at which point of technological development there will be a transition from well-behaved systems like Siri, which are able to interpret a limited amount of natural language inputs correctly, to superhuman artificial generally intelligent systems that are in principle capable of understanding any human conversation but which are not going to use that capability to interpret a goal like “minimize human suffering”.

[14] You are welcome to supply your own technical description of a superhuman artificial general intelligence (AGI). I will then use that description as the basis of any further argumentation. But you should also be able to show how your technical design specification is a probable outcome of AI research. Otherwise you are just choosing something that yields your desired conclusion.

And once you supplied your technical description you should be able to show how your technical design would interpret the natural language input “minimize human suffering”.

Then we can talk about how such simple narrow AI’s like Siri or IBM Watson can arrive at better results than your AGI and how AI research will lead to such systems.

[15] lesswrong.com/lw/we/recursive_selfimprovement/

[16] wiki.lesswrong.com/wiki/Paperclip_maximizer

[17] What is important to realize is that any goal is open to interpretation because no amount of detail can separate an object like a “paperclip” or an action like “maximization” from the rest of the universe without describing the state function of the entire universe. Which means that it is always necessary to refine your models of the world to better understand your goals.

“Utility” does only become well-defined if it is precisely known what it means to maximize it. The two English words “maximize paperclips” do not define how quickly and how economically it is supposed to happen.

“Utility” has to be defined. To maximize expected utility does not imply certain actions, efficiency and economic behavior, or the drive to protect yourself. You can also rationally maximize paperclips without protecting yourself if it is not part of your goal parameters. You can also assign utility to maximize paperclips as long as nothing turns you off but don’t care about being turned off.

Without an accurate comprehension of your goals it will be impossible to maximize expected “utility”. Concepts like “efficient”, “economic” or “self-protection” all have a meaning that is inseparable with an agent’s terminal goals. If you just tell it to maximize paperclips then this can be realized in an infinite number of ways given imprecise design and goal parameters. Undergoing to explosive recursive self-improvement, taking over the universe and filling it with paperclips, is just one outcome. Why would an arbitrary mind pulled from mind-design space care to do that? Why not just wait for paperclips to arise due to random fluctuations out of a state of chaos? That wouldn’t be irrational.

Again, it is possible to maximize paperclips in a lot of different ways. Which world state will a rational utility maximizer choose? Given that it is a rational decision maker, and that it has to do something, it will choose to achieve a world state that is implied by its model of reality, which includes humans and their intentions.

[18] By intelligent behavior I mean that it will act in a goal-oriented way.

[19] By “rational behavior” I mean that it will favor any action that 1.) maximizes the probability of obtaining beliefs that correspond to reality as closely as possible 2.) that does steer the future toward outcomes that maximize the probability of achieving its goals.

[20] wiki.lesswrong.com/wiki/Friendly_AI

[21] By “refinement” I mean the reduction of uncertainty and vagueness by narrowing down on the most probable interpretation of a goal.

[22] I do not doubt that it is in principle possible to build a process that tries to convert the universe into computronium to compute as many decimal digits of Pi as possible.

By “vagueness” I mean actions that are not explicitly, with mathematical precision, hardcoded but rather logical implications that have to be discovered.

For example, if an AGI was told to compute as many decimal digits of Pi as possible, it couldn’t possibly know what computational substrate is going to do the job most efficiently. That is an implication of its workings that it has to learn about first.

You do not know how to maximize simple U(x). All you have is a vague idea about using some sort of computer to do the job for you. But not how you are going to earn the money to buy the computer and what power source will be the cheapest. All those implicit constraints are unknown to you. They are implicit constraints because you are rational and not only care about maximizing U(x) but also to learn about the world and what it means to practically maximize that function, apart from the mathematical sense, because that’s what rational and intelligent agents do.

If you are assuming some sort of self-replicating calculator that follows a relatively simple set of instructions, then I agree that it will just try to maximize such a function in the mathematically “literal” sense and start to convert all matter in its surrounding to compute the answer. But that is not a general intelligence but mainly a behavior executor without any self-reflection and learning.

I reckon that it might be possible, although very unlikely, to design some sort of “autistic” general intelligence that tries to satisfy simple U(x) as verbatim as possible while minimizing any posterior exploration. But I haven’t heard any good argument for why such an AI would be the likely outcome of any research. It seems to be the case that it would take an deliberate effort to design such an agent. Any reasonable AGI project will have a strong focus on the capability of the AGI to learn and care what it is supposed to do rather than following a rigid set of functions and compute them without any spatio-temporal scope boundaries and resource limits.

And given complex U(x) I don’t see how even an “autistic” AGI could possibly ignore human intentions. The problem is that it is completely impossible to mathematically define complex U(x) and that therefore any complex U(x) must be made of various sub-functions that have to be defined by the AGI itself while building an accurate model of the world.

For example if U(X) = “Obtaining beliefs about X that correspond to reality as closely as possible”, then U(Minimize human suffering) = U(f(g(x))), where g(Minimize human suffering) = “Understand what ‘human’ refers to and apply f(x)”, where f(x) = “Learn what is meant by ‘minimize suffering’ according to what is referred to by ‘human'”.

In other words, “vagueness” is the necessity of a subsequent definition of actions an AGI is supposed to execute, by the AGI itself, as a general consequence of the impossibility to define complex world states, that an AGI is supposed to achieve.

Tags: , , , ,

If you’re interested in being on the right side of disputes, you will refute your opponents’ arguments. But if you’re interested in producing truth, you will fix your opponents’ arguments for them. To win, you must fight not only the creature you encounter; you [also] must fight the most horrible thing that can be constructed from its corpse.

Black Belt Bayesian

Contents

 

Abstract

In this post I just want to take a look at a few premises (P#) that need to be true simultaneously to make AI risk mitigation a wortwhile charitable cause from the point of view of someone trying to do as much good as possible by contributing money. I am going to show that the case of risks from AI is strongly conjunctive, that without a concrete and grounded understanding of AGI an abstract analysis of the issues is going to be very shaky, and that therefore AI risk mitigation is likely to be a bad choice as a charity. In other words, that which speaks in favor of AI risk mitigation does mainly consist of highly specific, conjunctive, non-evidence-backed speculations on possible bad outcomes.

Requirements for an Intelligence Explosion

P1 Fast, and therefore dangerous, recursive self-improvement is logically possible.

It took almost four hundred years to prove Fermat’s Last Theorem. The final proof is over a hundred pages long. Over a hundred pages! And we are not talking about something like an artificial general intelligence that can magically make itself smart enough to prove such theorems and many more that no human being would be capable of proving. Fermat’s Last Theorem simply states “no three positive integers a, b, and c can satisfy the equation a^n + b^n = c^n for any integer value of n greater than two.”

Even artificial intelligence researchers admit that “there could be non-linear complexity constrains meaning that even theoretically optimal algorithms experience strongly diminishing intelligence returns for additional compute power.” [1] We just don’t know.

Other possible problems include the impossibility of a stable utility function and a reflective decision theory, the intractability of real world expected utility maximization or that expected utility maximizers stumble over Pascal’s mugging, among other things [2].

For an AI to be capable of recursive self-improvement it also has to guarantee that its goals will be preserved when it improves itself. It is still questionable if it is possible to conclusively prove that improvements to an agent’s intelligence or decision procedures maximize expected utility. If this isn’t possible it won’t be rational or possible to undergo explosive self-improvement.

P1.b The fast computation of a simple algorithm is sufficient to outsmart and overpower humanity.

Imagine a group of 100 world-renowned scientists and military strategists.

  • The group is analogous to the initial resources of an AI.
  • The knowledge that the group has is analogous to what an AI could come up with by simply “thinking” about it given its current resources.

Could such a group easily wipe away the Roman empire when beamed back in time?

  • The Roman empire is analogous to our society today.

Even if you gave all of them a machine gun, the Romans would quickly adapt and the people from the future would run out of ammunition.

  • Machine guns are analogous to the supercomputer it runs on.

Consider that it takes a whole technological civilization to produce a modern smartphone.

You can’t just say “with more processing power you can do more different things”, that would be analogous to saying that “100 people” from today could just build more “machine guns”. But they can’t! They can’t use all their knowledge and magic from the future to defeat the Roman empire.

A lot of assumptions have to turn out to be correct to make humans discover simple algorithms over night that can then be improved to self-improve explosively.

You can also compare this to the idea of a Babylonian mathematician discovering modern science and physics given that he would be uploaded into a supercomputer (a possibility that is in and of itself already highly speculative). It assumes that he could brute-force conceptual revolutions.

Even if he was given a detailed explanation of how his mind works and the resources to understand it, self-improving to achieve superhuman intelligence assumes that throwing resources at the problem of intelligence will magically allow him to pull improved algorithms from solution space as if they were signposted.

But unknown unknowns are not signposted. It’s rather like finding a needle in a haystack. Evolution is great at doing that and assuming that one could speed up evolution considerably is another assumption about technological feasibility and real-world resources.

That conceptual revolutions are just a matter of computational resources is pure speculation.

If one were to speed up the whole Babylonian world and accelerate cultural evolution, obviously one would arrive quicker at some insights. But how much quicker? How much are many insights dependent on experiments, to yield empirical evidence, that can’t be speed-up considerably? And what is the return? Is the payoff proportionally to the resources that are necessary?

If you were going to speed up a chimp brain a million times, would it quickly reach human-level intelligence? If not, why then would it be different for a human-level intelligence trying to reach transhuman intelligence? It seems like a nice idea when formulated in English, but would it work?

Being able to state that an AI could use some magic to take over the earth does not make it a serious possibility.

Magic has to be discovered, adapted and manufactured first. It doesn’t just emerge out of nowhere from the computation of certain algorithms. It emerges from a society of agents with various different goals and heuristics like “Treating Rare Diseases in Cute Kittens”. It is an evolutionary process that relies on massive amounts of real-world feedback and empirical experimentation. Assuming that all that can happen because some simple algorithm is being computed is like believing it will emerge ‘out of nowhere’, it is magical thinking.

Unknown unknowns are not sign-posted. [3]

If people like Benoît B. Mandelbrot would have never decided to research Fractals then many modern movies wouldn’t be possible, as they rely on fractal landscape algorithms. Yet, at the time Benoît B. Mandelbrot conducted his research it was not foreseeable that his work would have any real-world applications.

Important discoveries are made because many routes with low or no expected utility are explored at the same time [4]. And to do so efficiently it takes random mutation, a whole society of minds, a lot of feedback and empirical experimentation.

“Treating rare diseases in cute kittens” might or might not provide genuine insights and open up new avenues for further research. As long as you don’t try it you won’t know.

The idea that a rigid consequentialist with simple values can think up insights and conceptual revolutions simply because it is instrumentally useful to do so is implausible.

Complex values are the cornerstone of diversity, which in turn enables creativity and drives the exploration of various conflicting routes. A singleton with a stable utility-function lacks the feedback provided by a society of minds and its cultural evolution.

You need to have various different agents with different utility-functions around to get the necessary diversity that can give rise to enough selection pressure. A “singleton” won’t be able to predict the actions of new and improved versions of itself by just running sandboxed simulations. Not just because of logical uncertainty but also because it is computationally intractable to predict the real-world payoff of changes to its decision procedures.

You need complex values to give rise to the necessary drives to function in a complex world. You can’t just tell an AI to protect itself. What would that even mean? What changes are illegitimate? What constitutes “self”? That are all unsolved problems that are just assumed to be solvable when talking about risks from AI.

An AI with simple values will simply lack the creativity, due to a lack of drives, to pursue the huge spectrum of research that a society of humans does pursue. Which will allow an AI to solve some well-defined narrow problems, but it will be unable to make use of the broad range of synergetic effects of cultural evolution. Cultural evolution is a result of the interaction of a wide range of utility-functions.

Yet even if we assume that there is one complete theory of general intelligence, once discovered, one just has to throw more resources at it. It might be able to incorporate all human knowledge, adapt it and find new patterns. But would it really be vastly superior to human society and their expert systems?

Can intelligence itself be improved apart from solving well-defined problems and making more accurate predictions on well-defined classes of problems? The discovery of unknown unknowns does not seem to be subject to other heuristics than natural selection. Without goals, well-defined goals, terms like “optimization” have no meaning.

P2 Fast, and therefore dangerous, recursive self-improvement is physically possible.

Even if it could be proven that explosive recursive self-improvement is logically possible, e.g. that there are no complexity constraints, the question remains if it is physically possible.

Our best theories about intelligence are highly abstract and their relation to real world human-level general intelligence is often wildly speculative [5][6].

P3 Fast, and therefore dangerous, recursive self-improvement is economically feasible.

To exemplify the problem take the science fictional idea of using antimatter as explosive for weapons. It is physically possible to produce antimatter and use it for large scale destruction. An equivalent of the Hiroshima atomic bomb will only take half a gram of antimatter. But it will take 2 billion years to produce that amount of antimatter [7].

We simply don’t know if intelligence is instrumental or quickly hits diminishing returns [8].

P3.b AGI is able to create (or acquire) resources, empowering technologies or civilisatory support [9].

We are already at a point where we have to build billion dollar chip manufacturing facilities to run our mobile phones. We need to build huge particle accelerators to obtain new insights into the nature of reality.

An AI would either have to rely on the help of a whole technological civilization or be in control of advanced nanotech assemblers.

And if an AI was to acquire the necessary resources on its own, its plan for world-domination would have to go unnoticed. This would require the workings of the AI to be opaque to its creators yet comprehensible to itself.

But an AI capable of efficient recursive self improvement must be able to

  1. comprehend its own workings
  2. predict how improvements, respectively improved versions of itself, are going to act to ensure that its values are preserved

So if the AI can do that, why wouldn’t humans be able to use the same algorithms to predict what the initial AI is going to do? And if the AI can’t do that, how is it going to maximize expected utility if it is unable to predict what it is going to do?

Any AI capable of efficient self-modification must be able to grasp its own workings and make predictions about improvements to various algorithms and its overall decision procedure. If an AI can do that, why would the humans who build it be unable to notice any malicious intentions? Why wouldn’t the humans who created it not be able to use the same algorithms that the AI uses to predict what it will do? If humans are unable to predict what the AI will do, how is the AI able to predict what improved versions of itself will do?

And even if an AI was able to somehow acquire large amounts of money. It is not easy to use the money. You can’t “just” build huge companies with fake identities, or a straw man, to create revolutionary technologies easily. Running companies with real people takes a lot of real-world knowledge, interactions and feedback. But most importantly, it takes a lot of time. An AI could not simply create a new Intel or Apple over a few years without its creators noticing anything.

The goals of an AI will be under scrutiny at any time. It seems very implausible that scientists, a company or the military are going to create an AI and then just let it run without bothering about its plans. An artificial agent is not a black box, like humans are, where one is only able to guess its real intentions.

A plan for world domination seems like something that can’t be concealed from its creators. Lying is no option if your algorithms are open to inspection.

P4 Dangerous recursive self-improvement is the default outcome of the creation of artificial general intelligence.

Complex goals need complex optimization parameters (the design specifications of the subject of the optimization process against which it will measure its success of self-improvement).

Even the creation of paperclips is a much more complex goal than telling an AI to compute as many decimal digits of Pi as possible.

For an AGI, that was designed to design paperclips, to pose an existential risk, its creators would have to be capable enough to enable it to take over the universe on its own, yet forget, or fail to, define time, space and energy bounds as part of its optimization parameters. Therefore, given the large amount of restrictions that are inevitably part of any advanced general intelligence (AGI), the nonhazardous subset of all possible outcomes might be much larger than that where the AGI works perfectly yet fails to hold before it could wreak havoc.

And even given a rational utility maximizer. It is possible to maximize paperclips in a lot of different ways. How it does it is fundamentally dependent on its utility-function and how precisely it was defined.

If there are no constraints in the form of design and goal parameters then it can maximize paperclips in all sorts of ways that don’t demand recursive self-improvement.

“Utility” does only become well-defined if we precisely define what it means to maximize it. Just maximizing paperclips doesn’t define how quickly and how economically it is supposed to happen.

The problem is that “utility” has to be defined. To maximize expected utility does not imply certain actions, efficiency and economic behavior, or the drive to protect yourself. You can also rationally maximize paperclips without protecting yourself if it is not part of your goal parameters.

You can also assign utility to maximize paperclips as long as nothing turns you off but don’t care about being turned off. If an AI is not explicitly programmed to care about it, then it won’t.

Without well-defined goals in form of a precise utility-function, it might be impossible to maximize expected “utility”. Concepts like “efficient”, “economic” or “self-protection” all have a meaning that is inseparable with an agent’s terminal goals. If you just tell it to maximize paperclips then this can be realized in an infinite number of ways that would all be rational given imprecise design and goal parameters. Undergoing to explosive recursive self-improvement, taking over the universe and filling it with paperclips, is just one outcome. Why would an arbitrary mind pulled from mind-design space care to do that? Why not just wait for paperclips to arise due to random fluctuations out of a state of chaos? That wouldn’t be irrational. To have an AI take over the universe as fast as possible you would have to explicitly design it to do so.

But for the sake of a thought experiment assume that the default case was recursive self-improvement. Now imagine that a company like Apple wanted to build an AI that could answer every question (an Oracle).

If Apple was going to build an Oracle it would anticipate that other people would also want to ask it questions. Therefore it can’t just waste all resources on looking for an inconsistency arising from the Peano axioms when asked to solve 1+1. It would not devote additional resources on answering those questions that are already known to be correct with a high probability. It wouldn’t be economically useful to take over the universe to answer simple questions.

It would neither be rational to look for an inconsistency arising from the Peano axioms while solving 1+1. To answer questions an Oracle needs a good amount of general intelligence. And concluding that asking it to solve 1+1 implies to look for an inconsistency arising from the Peano axioms does not seem reasonable. It also does not seem reasonable to suspect that humans desire an answer to their questions to approach infinite certainty. Why would someone build such an Oracle in the first place?

A reasonable Oracle would quickly yield good solutions by trying to find answers within a reasonable time which are with a high probability just 2–3% away from the optimal solution. I don’t think anyone would build an answering machine that throws the whole universe at the first sub-problem it encounters.

P5 The human development of artificial general intelligence will take place quickly.

What evidence do we have that there is some principle that, once discovered, allows us to grow superhuman intelligence overnight?

If the development of AGI takes place slowly, a gradual and controllable development, we might be able to learn from small-scale mistakes, or have enough time to develop friendly AI, while having to face other existential risks.

This might for example be the case if intelligence can not be captured by a discrete algorithm, or is modular, and therefore never allow us to reach a point where we can suddenly build the smartest thing ever that does just extend itself indefinitely.

Therefore the probability of an AI to undergo explosive recursive self-improvement (P(FOOM)) is the probability of the conjunction (P#P#) of its premises:

P(FOOM) = P(P1∧P2∧P3∧P4∧P5)

Of course, there are many more premises that need to be true in order to enable an AI to go FOOM, e.g. that each level of intelligence can effectively handle its own complexity, or that most AGI designs can somehow self-modify their way up to massive superhuman intelligence. But I believe that the above points are enough to show that the case for a hard takeoff is not disjunctive, but rather strongly conjunctive.

Requirements for AI risk mitigation to constitute an optimal charity

In this section I will assume the truth of all premises in the previous section.

P6 It is possible to solve friendly AI.

Say you believe that unfriendly AI will wipe us out with a probability of 60% and that there is another existential risk that will wipe us out with a probability of 10% even if unfriendly AI turns out to be no risk or in all possible worlds where it comes later. Both risks have the same utility x (if we don’t assume that an unfriendly AI could also wipe out aliens etc.). Thus .6x > .1x. But if the probability of solving friendly AI = A to the probability of solving the second risk = B is A ≤ (1/6)B then the expected utility of mitigating friendly AI is at best equal to the other existential risk because .6Ax ≤ .1Bx.

Consider that one order of magnitude more utility could easily be outweighed or trumped by an underestimation of the complexity of friendly AI.

So how hard is it to solve friendly AI?

Take for example Pascal’s mugging, if you can’t solve it then you need to implement a hack that is largely based on human intuition. Therefore, in order to estimate the possibility of solving friendly AI one needs to account for the difficulty in solving all sub-problems.

Consider that we don’t even know “how one would start to research the problem of getting a hypothetical AGI to recognize humans as distinguished beings.” [10]

P7 AI risk mitigation does not increase risks from AI.

By trying to solve friendly AI, AI risk advocates have to think about a lot of issues related to AI in general and might have to solve problems that will make it easier to create artificial general intelligence.

It is far from being clear that AI risk advocates are able to protect their findings against intrusion, betrayal, industrial or espionage.

P8 AI risk mitigation does not increase negative utility.

There are several possibilities by which AI risk advocates could actually cause a direct increase in negative utility.

1) Friendly AI is incredible hard and complex. Complex systems can fail in complex ways. Agents that are an effect of evolution have complex values. To satisfy complex values you need to meet complex circumstances. Therefore any attempt at friendly AI, which is incredible complex, is likely to fail in unforeseeable ways. A half-baked, not quite friendly, AI might create a living hell for the rest of time, increasing negative utility dramatically [11].

2) Humans are not provably friendly. Given the power to shape the universe the AI risk advocates might fail to act altruistic and deliberately implement an AI with selfish motives or horrible strategies [12].

P9 It makes sense to support AI risk mitigation at this time [13].

Therefore the probability of AI risk mitigation to be a worthwhile charitable cause (P(CHARITY)) is the probability of the conjunction (P#P#) of its premises:

P(CHARITY) = P(P6∧P7∧P8∧P9)

As before, there are many more premises that need to be true in order for AI risk mitigation to be the best choice for someone who wants to maximize doing good by contributing money to a charity.

Further Reading

The following posts and resources elaborate on many of the above points and hint at a lot of additional problems.

Notes and References

[1] Q&A with Shane Legg on risks from AI

[2] http://lukeprog.com/SaveTheWorld.html

[3] “In many ways, this is a book about hindsight. Pythagoras could not have imagined the uses to which his equation would be put (if, indeed, he ever came up with the equation himself in the first place). The same applies to almost all of the equations in this book. They were studied/discovered/developed by mathematicians and mathematical physicists who were investigating subjects that fascinated them deeply, not because they imagined that two hundred years later the work would lead to electric light bulbs or GPS or the internet, but rather because they were genuinely curious.”

17 Equations that changed the world

[4] Here is my list of “really stupid, frivolous academic pursuits” that have lead to major scientific breakthroughs.

  • Studying monkey social behaviors and eating habits lead to insights into HIV (Radiolab: Patient Zero)
  • Research into how algae move toward light paved the way for optogenetics: using light to control brain cells (Nature 2010 Method of the Year).
  • Black hole research gave us WiFi (ICRAR award)
  • Optometry informs architecture and saved lives on 9/11 (APA Monitor)
  • Certain groups HATE SETI, but SETI’s development of cloud-computing service SETI@HOME paved the way for citizen science and recent breakthroughs in protein folding (Popular Science)
  • Astronomers provide insights into medical imaging (TEDxBoston: Michell Borkin)
  • Basic physics experiments and the Fibonacci sequence help us understand plant growth and neuron development

http://blog.ketyov.com/2012/02/basic-science-is-about-creating.html

See also:

“Mathematical logic was initially considered a hopelessly abstract subject with no conceivable applications. As one computer scientist commented: “If, in 1901, a talented and sympathetic outsider had been called upon to survey the sciences and name the branch which would be least fruitful in [the] century ahead, his choice might well have settled upon mathematical logic.” And yet, it would provide the foundation for a field that would have more impact on the modern world than any other.”

[5] AIXI is often quoted as a proof of concept that it is possible for a simple algorithm to improve itself to such an extent that it could in principle reach superhuman intelligence. AIXI proves that there is a general theory of intelligence. But there is a minor problem, AIXI is as far from real world human-level general intelligence as an abstract notion of a Turing machine with an infinite tape is from a supercomputer with the computational capacity of the human brain. An abstract notion of intelligence doesn’t get you anywhere in terms of real-world general intelligence. Just as you won’t be able to upload yourself to a non-biological substrate because you showed that in some abstract sense you can simulate every physical process.”

Alexander Kruel, Why I am skeptical of risks from AI

[6] “…please bear in mind that the relation of Solomonoff induction and “Universal AI” to real-world general intelligence of any kind is also rather wildly speculative… This stuff is beautiful math, but does it really have anything to do with real-world intelligence? These theories have little to say about human intelligence, and they’re not directly useful as foundations for building AGI systems (though, admittedly, a handful of scientists are working on “scaling them down” to make them realistic; so far this only works for very simple toy problems, and it’s hard to see how to extend the approach broadly to yield anything near human-level AGI). And it’s not clear they will be applicable to future superintelligent minds either, as these minds may be best conceived using radically different concepts.”

Ben Goertzel, ‘Are Prediction and Reward Relevant to Superintelligences?

[7] http://public.web.cern.ch/public/en/spotlight/SpotlightAandD-en.html

[8] “If any increase in intelligence is vastly outweighed by its computational cost and the expenditure of time needed to discover it then it might not be instrumental for a perfectly rational agent (such as an artificial general intelligence), as imagined by game theorists, to increase its intelligence as opposed to using its existing intelligence to pursue its terminal goals directly or to invest its given resources to acquire other means of self-improvement, e.g. more efficient sensors.”

Alexander Kruel, Why I am skeptical of risks from AI

[9] Section ‘Necessary resources for an intelligence explosion’, Why I am skeptical of risks from AI, Alexander Kruel

[10] http://lesswrong.com/lw/3aa/friendly_ai_research_and_taskification/

[11] http://lesswrong.com/r/discussion/lw/ajm/ai_risk_and_opportunity_a_strategic_analysis/5ylx

[12] http://lesswrong.com/lw/8c3/qa_with_new_executive_director_of_singularity/5y77

[13] “I think that if you’re aiming to develop knowledge that won’t be useful until very very far in the future, you’re probably wasting your time, if for no other reason than this: by the time your knowledge is relevant, someone will probably have developed a tool (such as a narrow AI) so much more efficient in generating this knowledge that it renders your work moot.”

Holden Karnofsky in a conversation with Jaan Tallinn

Tags: , ,

The Power of Algorithms

Evolutionary processes are the most evident example of the power of simple algorithms.[1][2][3][4][5]

The field of evolutionary biology gathered a vast amount of evidence [6] that established evolution as the process that explains the local decrease in entropy[7], the complexity of life.

Since it can be conclusively shown that all life is an effect of an evolutionary process it is implicit that everything we do not understand about living beings is also an effect of evolution.

We might not understand the nature of intelligence[8] and consciousness[9] but we do know that they are the result of an optimization process that is neither intelligent nor conscious.

Therefore we know that it is possible for an physical optimization process to culminate in the creation of more advanced processes that feature superior qualities.

One of these qualities is the human ability to observe and improve the optimization process that created us. The most obvious example being science.[10]

Science can be thought of as civilization-level self-improvement method. It allows us to work together in a systematic and efficient way and accelerate the rate at which further improvements are made.

The Automation of Science

We know that optimization processes that can create improved versions of themselves are possible, even without an explicit understanding of their own workings, as exemplified by natural selection.

We know that optimization processes can lead to self-reinforcing improvements, as exemplified by the adaptation of the scientific method[11] as an improved evolutionary process and successor of natural selection.

Which raises questions about the continuation of this self-reinforcing feedback cycle and its possible implications.

One possibility is to automate science[12][13] and apply it to itself and its improvement.

But science is a tool and its bottleneck are its users. Humans, the biased[14] effect of the blind idiot god that is evolution.

Therefore the next logical step is to use science to figure out how to replace humans by a better version of themselves, artificial general intelligence.

Artificial general intelligence, that can recursively optimize itself [15], is the logical endpoint of various converging and self-reinforcing feedback cycles.

Risks from AI

Will we be able to build an artificial general intelligence? Yes, sooner or later.

Even the unintelligent, unconscious and aimless process of natural selection was capable of creating goal-oriented, intelligent and conscious agents that can think ahead, jump fitness gaps and improve upon the process that created them to engage in prediction and direct experimentation.

The question is, what are the possible implications of the invention of an artificial, fully autonomous, intelligent and goal-oriented optimization process?

One good bet is that such an agent will recursively improve its most versatile, and therefore instrumentally useful, resource. It will improve its general intelligence, respectively cross-domain optimization power.

Since it is unlikely that human intelligence is the optimum, the positive feedback effect, that is a result of using intelligence amplifications to amplify intelligence, is likely to lead to a level of intelligence that is generally more capable than the human intelligence level.

Humans are unlikely to be the most efficient thinkers because evolution is mindless and has no goals. Evolution did not actively try to create the smartest thing possible.

Evolution is further not limitlessly creative, each step of an evolutionary design must increase the fitness of its host. Which makes it probable that there are artificial mind designs that can do what no product of natural selection could accomplish, since an intelligent artificer does not rely on the incremental fitness of each step in the development process.

It is actually possible that human general intelligence is the bare minimum. Because the human level of intelligence might have been sufficient to both survive and reproduce and that therefore no further evolutionary pressure existed to select for even higher levels of general intelligence.

The implications of this possibility might be the creation of an intelligent agent that is more capable than humans in every sense. Maybe because it does directly employ superior approximations of our best formal methods, that tell us how to update based on evidence and how to choose between various actions. Or maybe it will simply think faster. It doesn’t matter.

What matters is that a superior intellect is probable and that it will be better than us at discovering knowledge and inventing new technology. Technology that will make it even more powerful and likely invincible.

And this is the problem. We might be unable to control such a superior being. Just like a group of chimpanzees is unable to stop a company from clearing its forest.[16]

But even if such a being is only slightly more capable than us. We might find ourselves at its mercy nonetheless.

Human history provides us with many examples[17][18][19] that make it abundantly clear that even the slightest advance can enable one group to dominate others.

What happens is that the dominant group imposes its values on the others. Which in turn raises the question of what values an artificial general intelligence might have and the implications of those values for us.

Due to our evolutionary origins, our struggle for survival and the necessity to cooperate with other agents, we are equipped with many values and a concern for the welfare of others.[20]

The information theoretic complexity[21][22] of our values is very high. Which means that it is highly unlikely for similar values to automatically arise in agents that are the product of intelligent design, agents that never underwent the million of years of competition with other agents that equipped humans with altruism and general compassion.

But that does not mean that an artificial intelligence won’t have any goals.[23][24] Just that those goals will be simple and their realization remorseless.[25]

An artificial general intelligence will do whatever is implied by its initial design. And we will be helpless to stop it from achieving its goals. Goals that won’t automatically respect our values.[26]

A likely implication is the total extinction of all of humanity.[27][28]

Further Reading

References

[1] Genetic Algorithms and Evolutionary Computation, talkorigins.org/faqs/genalg/genalg.html
[2] Fixing software bugs in 10 minutes or less using evolutionary computation, genetic-programming.org/hc2009/1-Forrest/Forrest-Presentation.pdf
[3] Automatically Finding Patches Using Genetic Programming, genetic-programming.org/hc2009/1-Forrest/Forrest-Paper-on-Patches.pdf
[4] A Genetic Programming Approach to Automated Software Repair, genetic-programming.org/hc2009/1-Forrest/Forrest-Paper-on-Repair.pdf
[5]GenProg: A Generic Method for Automatic Software Repair, virginia.edu/~weimer/p/weimer-tse2012-genprog.pdf
[6] 29+ Evidences for Macroevolution (The Scientific Case for Common Descent), talkorigins.org/faqs/comdesc/
[7] Thermodynamics, Evolution and Creationism, talkorigins.org/faqs/thermo.html
[8] A Collection of Definitions of Intelligence, vetta.org/documents/A-Collection-of-Definitions-of-Intelligence.pdf
[9] plato.stanford.edu/entries/consciousness/
[10] en.wikipedia.org/wiki/Science
[11] en.wikipedia.org/wiki/Scientific_method
[12] The Automation of Science, sciencemag.org/content/324/5923/85.abstract
[13] Computer Program Self-Discovers Laws of Physics, wired.com/wiredscience/2009/04/newtonai/
[14] List of cognitive biases, en.wikipedia.org/wiki/List_of_cognitive_biases
[15] Intelligence explosion, wiki.lesswrong.com/wiki/Intelligence_explosion
[16] 1% with Neil deGrasse Tyson, youtu.be/9nR9XEqrCvw
[17] Mongol military tactics and organization, en.wikipedia.org/wiki/Mongol_military_tactics_and_organization
[18] Wars of Alexander the Great, en.wikipedia.org/wiki/Wars_of_Alexander_the_Great
[19] Spanish colonization of the Americas, en.wikipedia.org/wiki/Spanish_colonization_of_the_Americas
[20] A Quantitative Test of Hamilton’s Rule for the Evolution of Altruism, plosbiology.org/article/info:doi/10.1371/journal.pbio.1000615
[21] Algorithmic information theory, scholarpedia.org/article/Algorithmic_information_theory
[22] Algorithmic probability, scholarpedia.org/article/Algorithmic_probability
[23] The Nature of Self-Improving Artificial Intelligence, selfawaresystems.files.wordpress.com/2008/01/nature_of_self_improving_ai.pdf
[24] The Basic AI Drives, selfawaresystems.files.wordpress.com/2008/01/ai_drives_final.pdf
[25] Paperclip maximizer, wiki.lesswrong.com/wiki/Paperclip_maximizer
[26] Friendly artificial intelligence, wiki.lesswrong.com/wiki/Friendly_artificial_intelligence
[27] Existential Risk, existential-risk.org
[28] 5 minutes on AI risk youtu.be/3jSMe0owGMs

Tags: , ,