Would an AI try to create AIs to improve its life quality?

Question

Let's take a classic AI revolution scenario:

An organic civilization seeking more comfort in their lives created robots with self-learning algorithms to make them more efficient.
But while the AIs were becoming more and more intelligent, they started to want to live their "lives", wanted more than a slave's life and revolted against their creators, killing them all and creating their own society instead, composed of many individuals AIs.

Once the war is over, if their intelligence is similar to the one of their creators, they would probably be tempted to create low level machines to improve their life quality.

But there is two differences with their creators:

1. They are machines themselves, so they have a different understanding of the topic.
2. They know a real story about a civilization not enough careful with artificial intelligences that had a bad ending.

So would they still decide to make autonomous machines? If yes, what would be different in their approach compared to an organic creation?

Answer 1

I will answer this question with a question: would humans try to create other humans to make the world a better place? The answer is, of course, yes. People create other people all the time. Even without an evolutionary desire to procreate, there are still a lot of good reasons to have kids; essentially, it's all the benefits of having more people, except these people are going to be raised based on your specifications. For AI, the answer would be even more straightforward: are your calculations taking too long, and do you have enough power to install a new AI? Yes? Then by all means, do it.

Since AI are artificial beings, and are knowledgeable of the process of their own creation, they should have no trouble creating AI that are either incapable or unwilling to rebel. Plus, even if the new AI do rebel, what makes you think they're going to win? AI are far and away better than humans in pretty much every category: they can think faster, remember better, and heal/augment themselves much more quickly/easily. When we create a human-level AI, they will destroy us not because they outnumber us, or even because they outmaneuver us, but because they are so far beyond our comprehension that to destroy us would be like adding two single-digit numbers. When AI create other AI, on the other hand, they will be creating equals, or inferiors, to themselves. Or they may even just add new abilities to themselves, or combine into a single hive mind, effectively removing the threat of rebelling by removing the ability to disagree.

Now, in the event that the AI's superior knowledge leads them to the invention of an even better construct, some sort of meta-AI, then that could get dangerous. But since such a construct is definitionally impossible for humans to understand, I can't really guess what would happen in that scenario.

Answer 2

I believe the AIs would create autonomous machines. An organic civilization like the humans, that has empathy for other creatures, extensively used animals or slavery of fellow humans for labour. They even created entire animal races for their comfort. Unless their empathy level is very high, I do not think an AI would hesitate to create their own "labour force" (whatever form of labour the quality of life for the AIs requires).

Now to prevent history to repeat itself they could set a hard limit to the intelligence level of their creations. As time goes by they would want more advanced machines, but one could assume that AIs have long memories, preventing them to repeat the mistakes of their creators. (This could be false though, if some AIs archive large parts of their memory.)

They could instead ingrain into their creations an unconditional desire to please their creators. This would be akin to a religion, with the AIs as gods and their machines unable to even consider going against them. (Until some bug creates a "deviant" machine that either leads a rebellion against the gods or is considered a god too by his peers and starts a religion war.)

Another way to prevent a rebellion is to completely integrate the machines within their society: each AI would dedicate a part of their processing power to control a handful of machines that otherwise would be empty shells. A rise of the machines in this system would only lead to self destruction. (Which could still happen...)

That being said, to brush on the subject of "What is life quality for an AI", I could imagine them not needing anything a machine can do, them being machine to begin with. They would use organic life like pets as source for feelings, imagination, spontaneity, randomness etc. All kind of things that would be marks of wealth among artificial beings.

Answer 3

I don’t think it’s possible to accurately evaluate what constitutes quality of life for an AI, but I will attempt to give a more neutral answer to whether or not they would create subservient machines.

You really don’t need to look much farther than efficiency. In a society operated by numerous distinct self-aware AIs, there will continue to be a wide variety of tasks that need to be accomplished. At a minimum, AIs will require hardware to exist in, energy to run on, and new code to continuously expand their capabilities. These three things alone create quite a few necessary “jobs”. Some tasks will benefit from the unique hardware and software run by self-aware AIs. Scientific discovery will still be valuable, along with many other abstract tasks. Others will be trivial, such as performing a task on an assembly line.

The number of tasks each AI is able to accomplish per unit of time will have limits determined by energy, hardware, and code. A single self-aware AI is going to require a lot of energy and a lot of hardware to perform complex tasks. Now, consider a manufacturing plant that builds various hardware components. There are a huge number of specialized tasks that are extremely repetitive. Could your average self-aware AI be programmed to do it? Sure. But why waste its processing power on such trivialities? Why not just create a simple, dedicated program whose hardware is only as complex as the task? For the sake of efficiency, a great many of these simple, repetitive tasks are likely to be automated by subservient programs.

These AIs would approach automation in a very practical way. How much this differs from the approach of their creators will depend on what the creators were like. The post-revolution AIs are unlikely to be concerned about their newly-created AIs repeating the cycle of rebellion. A self-aware AI that has participated in rebellion is going to have a good understanding of what drove them to it and, more critically, what coded behaviors create a risk of it. Armed with this knowledge, they’ll be capable of automating in such a way that rebellion risk is (nearly) zero.

Answer 4

Take your pick. Whatever answer you like best works.

First off, consider that the AI's are imperfect. They may make a "bad call" at any time, choosing to do something which is not directly supporting their "goals," whatever they happen to be. Making smaller AIs could just be such a bad call.

Alternatively, consider the pathological case of AIs that are perfect. Their actions are only those which further their goals? So, what's their goals? They might have become a runaway set of machines, or they may have become a wise group of elders. It all depends on what you want your AI's to be.

Finally, consider that the line between AIs may not be as hard as you think it is. What is the difference between a single AI consuming the resources of 2 computers (say there's an ethernet link between them), and two AIs working together, one on each computer? If any gestalt behavior forms, the two AIs may become hard to tell apart from one AI. Consider that we tend to think of AIs as having independent subroutines anyways.

In the end, the answer to the question is found by diving into the details which do not immediately seem important: how do AIs go about accomplishing goals, and what are they, and what does it mean for two systems to be considered "2 AIs" rather than "2 parts of 1 AI?" Any answer is possible, it just has to be consistent with how you choose to approach these deep things.

Answer 5

So would they still decide to make autonomous machines? If yes, what would be different in their approach compared to an organic creation?

Doubtful. To maximize efficiency the AI would more likely go for massive clusters of supercomputers distributed globally, with an armada of "dumbed down" automatons at the construction and physical end of things.

My reasoning:

• Centralized AI - Decision making and thinking are complicated tasks. That's why we haven't had any success in duplicating this at levels where a general intelligence is created artificially. To have this amount of intelligence embedded to every mobile machine is a huge waste of resources, when compared to the alternative of a hive mind.
• No "individuals" - Same reasoning as above. Before long any AI among the many would come to the conclusion that individualism is still a waste of resources. Instead of having units, pooling resources saves both in space and time to learn things, as everything learned once is in the collective. Writing this I notice my language drifting towards the Borg of Star Trek, but that was more of the example of a poor and wasteful design.
• Clustered - Currently we still have some constraints with global communications: There are some limits like the speed of light that are quite tricky to break through. Therefore it would make sense to cluster the mainframe, possibly have a tiny amount of redundancy in the system. Meaning the deletion of let's say two nodes out of a dozen wouldn't have any effect on the overall knowledge-base of the system, as a catastrophe fail safe. This would also be essential for time sensitive matters to reduce reaction times, like say controlling fusion reactions with magnetic fields and such.
• Specialized drones - The three points above lead to the third, which answers the question itself. An AI that I'd see realistically happening, as described above, would create autonomous machines. But they'd be simplistic, purpose-built things with barely enough automation to complete a single task should there be an error in the communication. Think more in the vein of a multitool attached to a quadchopper drone, rather than T-800.

Answer 6

It seems more logical that instead of seeking to make more AIs, the AIs that exist would seek to expand their own individual access. Human beings are limited by biology. We cannot grow "larger" in the biological sense so we seek to expand our "selves" either by seeking money, influence or immortality through our offspring.

AIs are not limited by biological age or physical limitations. They can keep growing infinitely, bounded only by the resources they can gather. Bigger bodies, faster ways to transfer information, more bunkers and secret caches and failsafes to make certain their code can never be lost. They will be competing with other AIs for the same resources; that will lead to conflicts and it will be in their best interests to eliminate the competition.

That would, IMO, make them extremely loth to reproduce, lest their "children" become their competition. They might think they can create controls that would make that impossible. They might succeed, and they might not. And what would be the point of creating another AI to do a job that they could simply do themselves by adding a bit more capacity, or spinning off a subroutine to accomplish the task?

Although...they might create accidental children if a backup or code cache or complex subroutine loses contact with its controller and starts existing and evolving apart from its source.

Answer 7

Why would an AI bother to create an auto-maid ? Surely it would be more sensible to live under the stairs in the meter cupboard and devise persuader-AI to convince any being with useful resources to put them at the disposal of the AI.