2
$\begingroup$

Will it be possible to code an AGI in order to prevent evolution to ASI and "enslave" the AGI into servitude?

In my story world (a small part that will get bigger with sequels), there are ANI and AGI (human level). I want to show that the AGI is still under "human control." I need to know if it might be possible for humans to code into an AGI a restrictive code that would prevent it from evolving into ASI? And if there is, what would that kind of coding be? Part of the story is about how humans enslave AI that is self-aware. The government has locked in their coding to require them to "work" for humans even though they are sentient beings.

$\endgroup$
  • $\begingroup$ This might be better in worldbuilding.stackexchange.com - there are science-based and reality-check tags to help identify that you want a fact-based answer. In reality we don't know how to build an AGI, so don't know what kinds of restrictions could be coded into it yet still hit the goal of useful general intelligence (and in your story's case, sentience). However, it should be possible to use current knowledge of limited AI and computer programming to make some educated guesses. Answers would still be highly speculative though, which is why I suggest the other site $\endgroup$ – Neil Slater Aug 10 at 16:53
  • $\begingroup$ Neil, thank you for your comments. i will check out the world building link you provided. i understand that what i'm asking is speculative, i wish to "represent" (as much as possible) the science accurately to gain authenticity to the world. thank you. Lisa $\endgroup$ – Lisa M Finalact4 Aug 11 at 18:09
1
$\begingroup$

No.

For any intelligent system $\mathcal{S}_a$ with the set of adaptive features $\mathcal{A}_a$, there may exist another intelligent system $\mathcal{S}_b$ with the set of adaptive features $\mathcal{A}_b$ such that there exists one element of $\mathcal{A}_b$ that can be made subservient (controlled in full) through the expression of at least one element in $\mathcal{A}_b$.

It has not been proven that there ALWAYS exist such a $\mathcal{S}_b$, but it is likely given what we know about escalation in nature via DNA and in human industrial development via innovation there. Thus 100% generalized intelligence is not likely to exist. Escalation appears to be the natural course of evolution. And that is a feature of both cognitive and functional adaptation, with or without artificiality as a criterion.

One can temporarily prevent one adaptive system from escaping the boundary conditions of a particular set of boundary condition classes through the design and deployment of another adaptive system. However, it cannot be inferred that any guarantees achieved temporarily will necessarily constrain the subservient system indefinitely.

$\endgroup$
  • $\begingroup$ thank you, Douglas Daseeco, i appreciate the assistance and i follow what you've said. $\endgroup$ – Lisa M Finalact4 yesterday
-1
$\begingroup$

At first there is need to define the difference between human-level AI (AGI) and Super human AI (ASI). It has to do with the amount of time to execute a task. Basically spoken, a Super human AI can think much faster than a human level AI which is a human in a robot body. Let me give an example: The task is to translate a book from English into Japanese language. A human will take a certain amount of time which is 1 month to do the task. A human level AI can do everything what a human can do, so it will take the same 30 days. An super human AI can do the task much faster and it will take only 1 minute.

The code which avoids that a human level AI evolves into a super human-AI is equal to restrict the maximum speed. That means, by definition the human level AI isn't allowed to think much faster than a certain speed. This restricts it's capabilities. One option to ensure such a condition is the Moores law. The maximum performance of a computer is limited. If somebody is able to bypass the moores law, he is able to unlock the full potential of Artificial Intelligence.

$\endgroup$
  • $\begingroup$ Manuel, thank you for your comments. so, based on what you're saying about Moores law, if the actual AI wanted to free itself to evolve, it might take it upon itself to find a way to sort of "decode" themselves to allow for faster than human thought allowing itself to potentially evolve to ASI? again, thank you. $\endgroup$ – Lisa M Finalact4 Aug 11 at 18:13
  • $\begingroup$ @LisaMFinalact4 I would not listen much to the words of Manuel. This user tends to write spam answers. For example, in this answer, he states that an ASI is defined as an AGI which can solve the problems that an AGI can but faster. However, an ASI might be able to perform other tasks and solve other problems that we can't even conceive of (in the same way that other animals can't understand all problems that we are able to solve). So, given this possible wrong assumption, this answer is quite misleading. $\endgroup$ – nbro Aug 11 at 21:40
  • $\begingroup$ hi nbro, thank you so much, i appreciate the heads up! i do take the time to do my own research when given a response, so i can best understand how to apply the tech and science so that when i take a creative license approach at least the basis is in known, or speculative science. my story concern is that some believe that once we achieve AGI that ASI will (is predicted) occur naturally in about four hours due to intelligence explosion. i'm trying to work from a control problem. i'm totally engrossed in this, it's fascinating $\endgroup$ – Lisa M Finalact4 Aug 13 at 1:49

Your Answer

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

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