-2
$\begingroup$

Would it happen gradually over the course of a few months or years with the AI having to slowly and stealthily plot the downfall of humanity from the shadows or would it just be "woops, you gave the AI wifi access and now it's launched all the nukes, crashed the stock market, and shagged your wife with a cucumber strapped to a Predator drone (because it hates you and you specifically)"? Would we have any time to fight back or would releasing a malevolent AI just be an immediate "game over, man"?

$\endgroup$

closed as too broad by Cort Ammon, Brythan, Josh King, TrEs-2b, Vincent Jul 27 '16 at 3:05

Please edit the question to limit it to a specific problem with enough detail to identify an adequate answer. Avoid asking multiple distinct questions at once. See the How to Ask page for help clarifying this question. If this question can be reworded to fit the rules in the help center, please edit the question.

  • 3
    $\begingroup$ What is an "unboxed AI"? $\endgroup$ – Mark Jul 26 '16 at 22:41
  • 1
    $\begingroup$ I'd also like to clarify what an AI is. The definition is somewhat nebulous, but going with Wikipedia's definition of "flexible rational agent that perceives its environment and takes actions that maximize its chance of success at some goal" we find no mention of any particular capabilities. No amazing super processing, nothing to best a human... in fact we're not given a definition that requires it to be smarter than a simple house mouse. $\endgroup$ – Ranger Jul 26 '16 at 22:46
  • 1
    $\begingroup$ This question is really too broad. I can define a scenario where we give the AI the power to cause the downfall over the course of minutes (give it control of the nukes), I can define a scenario where the AI achieves power over days or months (give it the abiility to build an army to capture the nukes). I can define a scenario where the takeover takes years (control of our economy). I can define a scenario where it takes generations (slowly eroding our society). I can define a scenario where we live in balance with this unboxed AI for an arbitrary amount of time. $\endgroup$ – Cort Ammon Jul 27 '16 at 0:15
  • 1
    $\begingroup$ No one knows, some scientist say it will take decades from AGI to ASI, and some predict it as a matter of hours. If you want to have some time to fight the evil AI do so, after all you are the god of your world :D. $\endgroup$ – Chinu Jul 27 '16 at 2:20
  • $\begingroup$ @Mark : Maybe something like this $\endgroup$ – King of Snakes Jul 27 '16 at 5:06
4
$\begingroup$

So I used to work in a Unix/VMWare group. Part of my job required sending images of a specific Linux distribution to servers in Alaska (from the lower 48 states).

Installing the image on the server took about three hours including the automated, pre-configured customization. Actually transferring the image took over 6 hours. I hooked up already populated disks to the VM, but there was an issue with the files. I had to use some built in utilities to search and verify the various discs. This took easily another 3 hours.

For a single machine.

By no stretch of the imagination could we see a meaningful <12 hour takeover by an AI; it would need to find, access, bypass the security systems of, and finally manipulate many, many systems.

Quite a few high-security systems are physically or logically separated from the Internet, which effectively means they're dead-ends for network based attacks. It's only reasonable to assume quite a few government systems that the AI would be interested in taking over would be configured this way, and the AI would have to find a more manual process around these hiccups.

Further this all assumes the AI is designed to be a security/black hat robot. If it's not, now the AI has to seek out intrusion and security learning materials and various utilities/tools, and begin to study them. Our AI will be limited by processor speed of his singular speed, as at this point he can't even hack together a crude networked cluster. Ever have Chrome be slow opening up? We can know for absolute certain an AI trying to teach itself security and AI would take monstrously longer than this. And throwing a supercomputer at it has been historically and repeatedly proven to be a terrible way to make gains (as most CS freshmen, even, will likely know).

I can only begin to guess how long it would take an AI to learn these things, experiment, and refine his abilities, but it wouldn't be fast. And even then he won't be able to crawl enough published information to understand intuitively to hack, say, the nuclear arsenal. Every exotic and new hack will be an exotic and new learning process.

I could go on like this, but this takeover won't be instantaneous. It will involve preparation, planning, and significant time for execution.

$\endgroup$
  • $\begingroup$ Yes, but are we talking how long from when the initial decision on the part of the AI is made, when it starts its preparatory work, the probing attacks and setup of compromised systems around the world. or from the moment the actual detectable attack is launched? The three could be very different timelines. One could be hours or days (even months?), last could be mere minutes. $\endgroup$ – Serban Tanasa Jul 26 '16 at 23:13
1
$\begingroup$

One issue which you have not specified is what the AI exactly has in mind as "taking over"?

It could conceivably disable automated systems, computer controlled manufacturing equipment and control systems, leaving the frustrated operators looking at the "Blue Screen of Death" or the equivalent of a VCR blinking 12:00...12:00...12:00...12:00...

This would mostly require corrupting the operating code so inputs or outputs become nonsense, so in theory it could be done rather quickly once the AI has gained access into the systems. Prior to that, during the preparatory stage, it would have created worms to go through the internet and infiltrate as many target machines as possible. This would damage or shut down a great deal of the civilian economy, but many of the most important systems would be firewall or air gapped, and not accessible.

The AI could risk its own destruction by simply shutting down the electrical grid and causing everything to go offline. Even protected systems with UPS and backup generators will only be able to function for a relatively short time after that (which means unless the AI or its creator had already mad very secure plans to deal with this, it too will be shut down). This virtually guarantees that the vast majority of systems will no longer work, and humans in technological societies will suddenly be stranded with the lights out. Mass chaos and mega deaths will follow.

If you have something different in mind for "taking over", then the AI will need a multi year program to prepare. The creation of worms and other malware will still be part of the plan, but the AI will need to do a deep study of human history, economics, politics, religions, society and so on IOT have the malware effectively cut off access to humans and carry out whatever the grand plan of the AI actually is. A fully successful plan might be noticed by a brief flickering of the lights and the unexpected rebooting of millions of machines, but the AI will have implanted subtle changes in the systems and outputs the humans will see, so they will be carrying out the AI's bidding without realizing what is happening.

This can be done in stages as well, with the AI gradually taking over networks and changing economic outputs, political results, manipulating social media and so on, and gradually encouraging the introduction of new firmware or programming in unaffected networks to allow the AI access. The AI can even create shell corporations to build new server farms to directly expand its own capabilities.

So the answer will depend on the desired end state of the AI.

$\endgroup$
-4
$\begingroup$

Give or take? probably 10 seconds to 10 minutes tops, you havent specified whats the capability of the AI and I only based that timeline on how fast internet connection is inbetween G8 nations

$\endgroup$
  • 1
    $\begingroup$ This doesn't really explain like the other answers. You need to explain why! $\endgroup$ – spark Jul 27 '16 at 12:17

Not the answer you're looking for? Browse other questions tagged or ask your own question.