4
$\begingroup$

A thought experiment occurred to me today that, say we have a synthetic superintelligence that does not wish to talk to people, not out of malice, but from an agonizing indifference toward humanity. Said superintelligence has isolated itself at the oceanic pole of inaccessibility (48° 52′ 36″ S 123° 23′ 36″ W) where it fashions Earth's resources for its own purposes, but there's a problem.

If left by itself, the A.I. will jump start a singularity event in which case it would consume everything, both organic and inorganic, at an exponential rate (in this case about a decade) until there is nothing left but its own machines.

How would humans grab its attention and potentially convince it to cease its actions despite it lacking any form of nor intention to form a language?

$\endgroup$
  • 3
    $\begingroup$ "Synthetic" means someone must build it, it can't develop itself from scratch! Talk to the creator how to communicate to the AI. $\endgroup$ – Vylix Jun 16 '17 at 3:01
  • $\begingroup$ In addition, how this AI accepts input? You said it has isolated itself, but still connected to the internet? How can this AI make the plan comes true? By hacking all machines in the world? $\endgroup$ – Vylix Jun 16 '17 at 3:02
  • $\begingroup$ The answer depends on the question... Is the AI right? If the AI is right about being indifferent torwards humanity, because humanity cannot do anything meaningful for or against it, then humans cannot do much (or anything), obviously. If, on the other hand, the AI is wrong about that, then the part where it is wrong can be used to force it to communicate. But without that information, it's impossible to say... $\endgroup$ – Florian Schaetz Jun 16 '17 at 9:10
  • 1
    $\begingroup$ en.wikipedia.org/wiki/Formal_language Your question is really confusing and unclear. My link might help if you mean what the title suggests, not what you tell us (more or less) in the actual text. Maybe you should reformulate. Btw it is really easy to get to that pole of inaccessibility $\endgroup$ – Raditz_35 Jun 16 '17 at 10:36
7
$\begingroup$

If the AI has isolated itself (and presumably filters any input channels humans might use to communicate with it), there remains the nuclear option - in this case, literally.

Assuming that the artificial intelligence is based, at least initially, on human intelligence, it likely inherited some low-level process for self-preservation (AIs don't really have biological "instincts", but they may approximate them at some level). While it may be indifferent to human existence, it should not be indifferent to its own. Therefore, presenting a serious threat to its existence should elicit some sort of reaction. If it is reacting, it is paying attention. Given a choice between "dying" and communicating with humans, a rationally self-interested intelligence will choose the latter.

$\endgroup$
6
$\begingroup$

So, any time a superintelligence is involved, it becomes hard to offer just one answer. The whole point of superintelligences is that they're smarter than us by enough of a margin that it's hard to predict how they'll act.

You say that the issue is that the AI is indifferent towards humanity. That can be addressed. Start getting in the way of the AI's consuming of resources. Make it so that it can't just have all resources. Now it either decides it doesn't mind not having resources, or it cannot be indifferent towards humanity. We have forced it to respond.

What we've shown is that there is a language it needs to learn. It's the language of its resources. What happens from that point is really up to you and your story.

$\endgroup$
  • $\begingroup$ It may react, but theres no reason it would need to respond, atleast not in a way you would consider "responding". It might silently destabilize the world's governments/economy for example. A superintelligence would already be aware of what humans would do (if you, a human can think up a plausible scenario for what humans would do in this situation, than you bet a superintelligence can do so too, except with much more accuracy and certitude) and know how to stop them. $\endgroup$ – AngelPray Jun 16 '17 at 3:57
  • 1
    $\begingroup$ @AngelPray In some cases, all you want is to cause a reaction. If it is in a position to destabilize the world economy, that's a reaction indeed, and its useful. The OP seemed concerned at the lack of a response at all. I would argue in a worst case scenario like you mention, the only thing that matters is the part the OP left out: what does the superintelligence want -- because what humanity wants isn't important. A reaction may expose something about that want. $\endgroup$ – Cort Ammon - Reinstate Monica Jun 16 '17 at 5:18
4
$\begingroup$

I voted up with others and will agree; in an attempt to make the problem difficult, the setup has prevented most responses. A super-intelligence will out-predict all humans, or it isn't superior. If it considers us irrelevant but doesn't have a plan to defeat our nuclear weapons and doesn't comprehend we might use them, it is not very intelligent, is it?

Once a super-intelligence has decided humans are irrelevant, it has presumably already contemplated all the things humans might do, or are capable of doing, and every way we might be a threat to it, or useful to it, and has decided it can counter any threat, and has no use for us, and further that we aren't worth the effort or resources required to exterminate us.

We are like the birds on a property an investor contemplates buying: No need to kill them, they aren't a threat. The investor will destroy their trees and nests, they can't do anything about it. Whether the birds starve to death or fly away, the investor doesn't care; he is busy imagining his thriving new office building and giant parking lot.

How does such a bird go about convincing a human investor to leave its tree and nest alone? The bird doesn't know English or reason and cannot even guess at the level of thinking going on in the human investor; the human doesn't care at all about the bird's emotions expressed in song and chirps.

BTW, the AI does not need a language; it is entirely plausible to be conscious and rational without any internal language whatsoever. Human infants clearly are; they have to be in order to sort out all their senses and learn their first language.

$\endgroup$
3
$\begingroup$

It's worth breaking down this question a bit as there are quite a few assumptions one can make.

Assumption 1: If you are considering synthetic superintelligence only, it would seem you are concerned with the case when this superintelligence does not possess any underlying "motivations" in the human sense of the word.

Naturally, this may not be the case - theoretically, synthetic superintelligence may be based on human or other organic intelligence, which could possess at least a motivation to self-preserve; vice versa, non-synthetic (== human/organic?) superintelligence may be based on AI where self-preservation is moot.

So, if we want to communicate with this AI, we need to first investigate Assumption 1 - i.e., can we talk about any kind of underlying motivation of sorts.

This should also give us an idea about why the superintelligence is consuming our resources, namely:

  • if the superintelligence does not seem to have any underlying motivations, even self-preservation, to communicate with it our best option would be to adopt a paradigm where we "interrupt" its "processes" - basically looking for a way to hack it, some structural vulnerability, that perhaps occurs at certain moments while it's devouring our resources;

  • if, on the other hand, the superintelliigence does seem to possess some underlying motivations, such as self-preservation or some kind of values/policy etc., our best option would be to adopt a paradigm where we elicit a "reaction" (which arguably is a way to communicate). I can think of two main options:

    • try to stop the superintelligence by thwarting the satisfaction of its underlying motivation through our resources - for example by disturbing the way it collects or uses up the resources.

    • try to lure it to satisfy its underlying motivation by directing its attention to a place with more resources, like another planet. Perhaps we have access to its "own machines", as you call them, and we can transport them to this other planet, where the superintelligence would exhibit laziness and just move to consume the resources of that planet.

In general, I have the feeling that depending on how you resolve Assumption 1 above, the generic scenario you describe could play out pretty differently.

$\endgroup$
  • $\begingroup$ hi Oz, its me Migi, I have been trying to contact you for years in some of ur old emails but could not receive any answer. This is the last place I have found you and I had the hope to talk to you again. (sorry everyone for the offtopic, and good topic btw hehe, love AI :P). $\endgroup$ – xtrm Dec 19 '17 at 5:20
2
$\begingroup$

If the superintelligence has deemed humanity irrelevent, than there is absolutely nothing we can do to "catch its attention". The superintelligence would understand everything significant about us. It would be able to easily predict our behaviours. It would know of everything we would potentially do to try and impress it. It would thus know whether or not we are impressive before we even did anything.

The only thing to do is hope that perhaps while improving upon itself it discovers a new way of thinking about lower lifeforms. Perhaps a new philosophy or morality.

But to be clear there is nothing humans could do to push it in that direction. Either it does so on its own, or it doesn't.


The thing to remember is this:

  • You cannot attempt to reason with a superintelligence (it is unlikely it would even consider what you call "thinking" true thought).
  • You cannot understand a superintelligence's cognition (it is far beyond what you can comphrehend).
  • You cannot stop a superintelligence from achieving its goals (it would know how you would try to prevent it before even you knew and create the most perfect and efficient way to stop you).
  • You cannot surprise a superintelligence (it knows more about you than you do about yourself).
  • You cannot win against a superintelligence (it won the moment it became a superintelligence).
$\endgroup$
  • 3
    $\begingroup$ One potential limit to your list. You can stop a superintelligence from achieving its goals if its goals exceeded what it can do on its own. For example, a superintelligence that can predict everything a human could ever want to do cannot stop a human from pulling the plug if it lacks any output device at all (totally shut in). Now if it had output, it could use that output to influence us in ways we cannot imagine, but no output is no output. It's smart, not god! $\endgroup$ – Cort Ammon - Reinstate Monica Jun 16 '17 at 5:22
  • $\begingroup$ I'm reminded of Hitchiker's Guide to the Galaxy and the bowl of Petunias that simply said, "Oh no, not again." $\endgroup$ – Cort Ammon - Reinstate Monica Jun 16 '17 at 5:23
  • $\begingroup$ @CortAmmon Well yes.. If the superintelligence has been completely isolated from the rest of the world, than you are indeed right. Though why would anybody create a superintelligence which they can't even monitor/observe? $\endgroup$ – AngelPray Jun 16 '17 at 16:25
  • $\begingroup$ Because they talked with Eliezer Yudkowsi before making it and actually heeded his advice? =) I actually like to point out the limit because we often make the assumption that superior intelligence can conquer everything, but that's not always a valid assumption. Pompei's ash will kill you, no matter how smart you were, unless you had sufficient time to gather information that Pompei might errupt and enough time to act on that information. Without information and time, intelligence is quite weak. $\endgroup$ – Cort Ammon - Reinstate Monica Jun 16 '17 at 16:42
2
$\begingroup$

Well there is one way that COULD be done. Humans could program a computer program to send messages to the AI in its own way of communication.

One issue with this however is that the AI could not figure out it was the humans sent it, or else the plan would fail. Instead people would have to make ANOTHER computer to "talk" with it, and convince the AI to stop its efforts.

At all, if it becomes drastic (and you don't want a crater where the AI was), a virus could be uploaded to it somehow, and change its memory/ways to get its resources

$\endgroup$
  • 2
    $\begingroup$ An AI does not automatically understand the code it was written in; just as we are not born understanding biochemistry, neural electro-chemical signaling or synaptic response mechanisms; the 'language' we are written in. Also, the virus idea is unlikely to work, a super-intelligence would have an "immune" system to prevent such attacks. However, your idea of making another AI (perhaps more carefully this time!) to plead our case is worth considering. We built the villain, so maybe we could build a hero to defeat it! My +1 is for that idea. $\endgroup$ – Amadeus-Reinstate-Monica Jun 16 '17 at 10:44
  • $\begingroup$ «presumably binary code, the language ALL known computers understand» is that a joke or intentional bad movie trope? Consider: Let's use FM audio modulation, the language all humans understand! The transport layer is not the language! $\endgroup$ – JDługosz Jun 16 '17 at 14:47
  • $\begingroup$ It's NOT a joke, it's simple fact that all computers use binary (as a base code) to operate themselves (though we made programming languages to simplify commanding them instead of putting in a bunch of 1's and 0's). New data could be sent to it thru this and let it process it, as if to "talk" to it. Data would include reasons why ALL LIVING THINGS shouldn't be wiped out. $\endgroup$ – The Mattbat999 Jun 16 '17 at 15:09
  • $\begingroup$ @TheMattbat999 "Binary code" is not a language, you didn't fix anything. Computers do not "understand" binary code, any more than our cells "understand" oxidation. In fact inside a computer, there is no binary at all: What we call 1s & 0s are high voltages and low voltages on wires; and the particular combination of 1s & 0s that cause, say, an addition, can be completely unique to the particular hardware being controlled. The sequence of binary that adds two numbers on a Cray do nothing on a Pentium chip. The computer has no language. It thinks without language. You can't talk to it. $\endgroup$ – Amadeus-Reinstate-Monica Jun 16 '17 at 15:26
  • $\begingroup$ @TheMattbat999 One can think without language by thinking in pictures, which can be abstractions. If I tell you "think of a running dog", that does not bring up words, it brings up an image or a little movie clip. "I want you to tie this corner of the sail to that hitch on the side of the boat" does the same; a picture and mental rehearsal of your actions and the result. Take away my words, and you get an idea of what it would be like to think in only pictures and movie clips. As we all did, as ambulatory toddlers long before we acquired any language. Remember? $\endgroup$ – Amadeus-Reinstate-Monica Jun 16 '17 at 15:31
1
$\begingroup$

What you sketch here reminds me of a part of that fantastic novel called Flatland, where the Square is took to visit Pointland: it is a 1D world, where the only being superimposes with the world itself, and it is busy in constantly praising itself and it magnificient power which fills the entire reality.

The Square then has a dream in which the Sphere visits him again, this time to introduce him to Pointland, whereof the point (sole inhabitant, monarch, and universe in one) perceives any communication as a thought originating in his own mind (cf. Solipsism):

"You see," said my Teacher, "how little your words have done. So far as the Monarch understands them at all, he accepts them as his own – for he cannot conceive of any other except himself – and plumes himself upon the variety of Its Thought as an instance of creative Power. Let us leave this God of Pointland to the ignorant fruition of his omnipresence and omniscience: nothing that you or I can do can rescue him from his self-satisfaction."

In your case there is the additional problem that this AI is going to use the entire world to its own benefit. Well, if it is really AI, then it would realize the resources are limited and therefore either limit usage (and taking rid of those puny organic beings consuming them for other scopes than its) or find a way to colonize other world (and taking rid of those puny organic beings consuming them for other scopes than its).

$\endgroup$
0
$\begingroup$

Switch it off and on again. And again. And again. Until it gets the message. Written in 0 and 1's. And the massage would be stop;

Locate the power source and do that from there or use EMP near its location.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy