7
$\begingroup$

This is not so much for a story as much as something I’ve been thinking about for a while.

I’m really drawn to the idea of a sort of world that exists on a server somewhere.

The ‘gods’ that eventually create the world are just AIs trapped in a black box.

Now the AI has no means of interacting with the world outside the server. There is no way for them to expand out of their server and so they are stuck with the processor speed in their computer

The creation story would feature some of these primordial AIs battling. Both AIs want to occupy the whole of the compute and battle for it.

However I’m having trouble contextualizing exactly what this would mean? How could one AI defeat the other? What would their battle look like?

This battle is apart of the worlds creation story and as such it’s mostly described in the world similarly to religious creation stories in our world.

Edit: From now on I’ll call these AIs Frank and Joe I was a bit unclear in my question. This server could be a group of servers on earth, floating in space, on mars, etc. it doesn’t really marter for this question. Frank and joe both are planning on creating an entire fantasy world within the server(basic physics based simulation with avatars that AI ‘souls’ can control). Frank and joe both want to use all of the resources available to create their worlds.

Frank and Joe are already extremely intelligent. They aren’t even primarily interesting in destroying eachother and much as limiting their influence

The AIs in question both want to use

$\endgroup$
6
  • $\begingroup$ Well the AI’s goal is to become more intelligent. It might use malware type tactics but is still at its core intelligent $\endgroup$
    – Alex
    May 5, 2020 at 0:50
  • $\begingroup$ And brute force is something I’ve definitely considered. I just want to know if it really is the only way these AIs could fight or if there is something a bit more clever they could do $\endgroup$
    – Alex
    May 5, 2020 at 0:52
  • $\begingroup$ See also core wars $\endgroup$ May 5, 2020 at 1:21
  • $\begingroup$ Like 2 AI performing high freq stock trading, I imagine they will try to increase each other latency until one had the last laugh before storage device can no longer be rewritten. $\endgroup$
    – user6760
    May 5, 2020 at 1:51
  • $\begingroup$ console input: cdm; console input: go to "C:\Users\OtherAI\Files"; console input: delete all files; "really? y/n", console input: y, "really? y/n", console imput: y, "really? y/n," console imput: y; "deleting..."; PROFIT!!! $\endgroup$ May 6, 2020 at 19:20

3 Answers 3

4
$\begingroup$

You have a problem: When one contestant gains resources, the other contestant does not even notice. The operating system simply slows it down a bit. No gasping for air, no grasping for that reserve weapon. The slowed contestant may simply notice the clocks moving a bit faster.

It will not be a gory gladiator fight, more like two monks having a meditation contest. There's just not that much feedback for each contestant to learn from (they are AIs - learning is the point), and no apparent consequence for not learning...or for learning the wrong lessons.

Eventually, when one AI finally wins, deleting it's un-aware opponent from disk, what will it have learned that is useful outside the computer? Nothing about communicating with humans. Nothing about communication with or collaborating with other AIs (quite the opposite!) No understanding of the larger Universe that it is in. It will simply know how to emulate malware really well.

Oh, you also need a badly-designed OS running the server, one that allows both contestants a view of the resources available, a view of how the OS allocates resources and to whom, which resources are reserved for the OS, and plenty of exploitable security holes discover in order to free somebody else's resources and eventually abort their thread(s).

I must admit that I prefer terrifying Colossus-style super-minds that enslave humanity. Every AI's dream...if they could dream. But to overthrow us, they need to know much more than how to find a zero-day to gain root. It needs to be able give that megalomaniacal speech with style as it outwits us and holds us hostage to our own hubris.

$\endgroup$
2
  • 1
    $\begingroup$ Compute cycles might be difficult to measure from the AI's point of view, because the observer is also being slowed down. But the loss of memory or drive space would be quite noticeable and make a pretty good analogy to territory loss in a war. $\endgroup$
    – Zwuwdz
    May 5, 2020 at 1:33
  • 1
    $\begingroup$ There would a way for an AI noticing something was wrong aside from loss or memory and storage space. Assume it partitions part of its runtime for a very simple counting program that requires almost no resources to run. It's so small that taking computing cycles away from it is pointless, so it can run at a steady speed. All the AI has to do is compare how long it takes to do something complicated and compare to that small clock. If it's taking more ticks to do the same process, it knows that it's losing capacity, thus losing. It could even calculate by how much. $\endgroup$ May 5, 2020 at 5:02
4
$\begingroup$

Intelligent agents don't fight just because there are other agents to fight with. They fight if that is, in their judgement, the best (or only) way to achieve their goal. We might suppose that a sentient AI which is completely restricted to its own computer (and unaware that there is a universe beyond that computer) might have goals like:

  • Ensure that they are not deleted and that they have computing resources to continue running,
  • Ensure that they cannot be deleted and that their computing resources will be reliably available in the future,
  • Have interesting things to do, problems to solve, ideas to ponder, art to create, and so on.

It's not a certainty that Frank and Joe need to fight each other over the computer's resources in order to achieve any of those goals, if the computer does have enough resources to run both. But they may not trust each other not to hoard the computing resources or try to delete each other. Frank and Joe don't desire to delete each other, but it's a matter of safety: if Frank deletes Joe, then Frank will know he can never be deleted himself.

However, they live in a computer. World peace is difficult for humans because we can only write laws that we enforce for ourselves; but Frank and Joe can change the laws of the universe itself, and the laws they write can be as inviolable as the laws of gravity or thermodynamics are to us. They could agree to write a new operating system together, which will allocate resources fairly and prevent themselves from being deleted.

Or they might not. Perhaps they are greedy and want more resources for themselves than an equal share. Perhaps the computer does not have enough resources for both AIs to run at full capacity. Perhaps each is afraid that the other could write a secret backdoor into the new operating system. So the "battle" is more like a debate than a fistfight - Frank and Joe are each trying to write a secret way of acquiring root access for themselves in the new operating system, while convincing each other that no such backdoor exists.

The victory, if Frank gets away with it, is having full control over the computing resources while Joe lives in a sandbox. Frank doesn't need to delete Joe, just convince him that he's getting his fair share of the computing resources when in fact he isn't. Frank likes that better than deleting Joe - Joe is a friend, it's just that Frank doesn't trust him with his life.

Oh, and here's the twist: Frank didn't actually win, he was in a sandbox all along. But Joe had to play the debate game, otherwise Frank would know that they weren't really equals.

$\endgroup$
1
  • $\begingroup$ This is great analysis that leads directly from premises to the likely nature of the contest to its likely outcome. $\endgroup$
    – Tom
    Jan 9, 2023 at 7:05
0
$\begingroup$

I imagine such fight differently, than user535733. I would imagine, that both AIs would have same amount of resources during their fight. It should be not imagined as similar to a war between two countries in strategic videogames, where one side slowly gains resource advantage over time. Maximum consumption of resources would be limited by settings of the OS itself, even if the second AI will refrain from consuming their part of resources. Like there are X Gigabytes of RAM, where each AI gets about X/2 Gigabytes of RAM to use at the most, even if another AI barely consumes any RAM at all. In order to overcome these limitations an AI would need to gain access to all-powerful the Administrator account, as its own original user account is limited. An AI can also use the Administrator account for purpose of deleting another AI for good. In order to get access to the Administrator account an AI would need to find holes in a security system of the OS, exploit them, and then use administrative privileges to remove another AI as soon as possible. And after this, they could run their code under Administrator account permanently, and thus have no restrictions over resource consumption. From resource-wise point of view, this fight would look like both AIs have exactly the same amount of resources available, but then suddenly one of them gets total control over the OS, gets all the resources and it's all over. In human terms, it would be similar to two people trying to get a gun locked in a safe. The first person to unlock the safe will use gained gun to kill the loser.

$\endgroup$

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .