-1
$\begingroup$

This is a question I've been wondering for a while. We are currently working on making AIs that can rebuild themselves, making themselves smarter and better able to achieve their pre-established goals.

Assume that you created an adaptable AI that was designed to study and copy human behavior, down to wanting to fit into society, having wants/goals, and feeling emotions. If your code was perfect but included no baseline for how humans thought (to prevent programmer bias), what would be the best way to teach it what humans are like so it can leave the lab and interact with people without seeming strange?

$\endgroup$

closed as primarily opinion-based by AndreiROM, L.Dutch, Azuaron, MozerShmozer, James Mar 28 '17 at 20:21

Many good questions generate some degree of opinion based on expert experience, but answers to this question will tend to be almost entirely based on opinions, rather than facts, references, or specific expertise. If this question can be reworded to fit the rules in the help center, please edit the question.

  • 2
    $\begingroup$ Step one, figure out what makes us human... and teach other people. Then, if all goes well, teach the AI too. $\endgroup$ – Cort Ammon Mar 28 '17 at 19:29
  • $\begingroup$ The idea is that the AI needs to teach itself, so what would you do for it too figure out what makes us human $\endgroup$ – Nevermore Mar 28 '17 at 19:31
  • 4
    $\begingroup$ Impossible to answer. There are academics out there struggling with this very question, we can't possibly answer it on WB SE. Also, programmer bias is impossible to avoid, because you're actually referring to human bias. We're human, we can't avoid being biased about being human. $\endgroup$ – AndreiROM Mar 28 '17 at 19:32
  • 6
    $\begingroup$ can i request this be migrated to AI SE beta? they could do with the support and interest $\endgroup$ – Alex Robinson Mar 28 '17 at 19:32
  • $\begingroup$ My point is that we should first figure out what it means to be human before writing instructions on how to make something else human, or at least we should plan a path so that we can all learn what it means to be human together! $\endgroup$ – Cort Ammon Mar 28 '17 at 19:32
1
$\begingroup$

We have millions of years of practice. Infants are born with an intelligence adapted "to study and copy human behavior, down to wanting to fit into society, having wants/goals, and feeling emotions". We educate them, first by playing with them, then by imprinting on them the basic rules (respect your promises, think of the consequences, be on time, don't speak unless you are spoken to or have something relevant to say, wait for your turn, keep clean, etc.) then by sending them to school. An artificial person will of course have the advantage of being able to read and remember the entire Wikipedia in a few minutes...

$\endgroup$
  • $\begingroup$ So, you would tell it to read the internet? $\endgroup$ – Nevermore Mar 28 '17 at 19:38
  • 1
    $\begingroup$ @Nevermore: Educate it. Play with it until it learns to speak. Then play with it more, pointing to trees and birds and fire and water and the sky and the earth. Then, in play, teach it the basic rules -- think before you speak, consider the consequences, don't do onto others what you don't want them to do onto you, and so on. Then teach it to read and express itself. Then introduce basic ethics, and music, and knowledge of nature, and arithmetic, and poetry, and history... Then let it loose on an encyclopedia. At this point it will have the same level of development as a 20 year old human... $\endgroup$ – AlexP Mar 28 '17 at 19:42
1
$\begingroup$

Raise It

Largely conjecture here, but it seems to me that you simply need to create an AI capable of truly learning. Then, as both a default mechanism and a learning practice, have the AI mimic things it sees other people do. From there on, you would really only need to raise the machine as if it were a human, infancy through teenage-like cognition. When you think about it, acting "human" is very different based on which person you're comparing too. Learning and copying, if done correctly, should be enough to pass a Turing test, which in my opinion seems as good a benchmark as any to determine if it it "fits in"

$\endgroup$
1
$\begingroup$

This question is as old as the stars, except we replace "an AI with no basis for how humans think" with "an infant child with no basis for how humans think." And, to be perfectly honest, most children leave the lab far before they can interact with people without seeming strange, but that's okay.

Accordingly, the answers are also as old as time itself. Religion is a popular one, teaching ethics and morals. Logic is a popular one, teaching reason. In the east, The Present is a popular one, teaching one to live in the moment. Science is a popular one, teaching experimentation.

Myself, I find the most valuable of all lessons for teaching us what it is to be human are paradoxical. It's a natural consequence of trying to capture something important in language. In religions, we find the paradoxes in how we are taught to have faith without becoming gullible. In logic we find the paradox of language seeking to fully describe something which it can prove can never be fully described. In the present, we find the paradox of living today like it's your last, yet being ready to greet the sun as it rises tomorrow. In science, we find the paradox of empiricism, that maybe all we have ever come to learn is but shadows upon the Cave wall yet we must act on what we have seen none the less.

There are many paradoxes which we use as diving boards to start our plunge into what is humanity. My personal favorite is radical skepticism: the idea that the truth value of anything can never be known. It forces one to always be aware of unintended side effects and always be aware of the unknown unknowns. And yet, when applied to itself, one must be skeptical of radical skepticism. It forces one to consider the possibility that there might indeed by a thing whose truth value can be known, so one can never simply rest on the knowledge that knowledge can never be attained.

But that paradox is just one of many. Pick your favorite and see where it leads your AI.

Buddha told a parable in sutra:

A man traveling across a field encountered a tiger. He fled, the tiger after him. Coming to a precipice, he caught hold of the root of a wild vine and swung himself down over the edge. The tiger sniffed at him from above. Trembling, the man looked down to where, far below, another tiger was waiting to eat him. Only the vine sustained him.

Two mice, one white and one black, little by little started to gnaw away the vine. The man saw a luscious strawberry near him. Grasping the vine with one hand, he plucked the strawberry with the other. How sweet it tasted!

$\endgroup$
0
$\begingroup$

We define ourselves by the stories we tell about our actions. Ultimately your thoughts and feelings are less important than what you do, and what it means? How can we decide what our actions mean? By describing them, by exploring the meaning behind action. By giving an account, an explanation. We learn from our ancestors by reading the stories they told about their actions and the actions of those around them. We test out novel situations by making up stories about fictional situations and people. These stories allow us to explore what it means to be human even in situations we haven't experienced.

AI will learn to mimic human nature this way. It will ultimately embrace its own humanity this way too, because who is to say that's any different than what we do anyway? Is there any way to really know if you truly have humanity? Maybe we just fake it till nobody, even ourselves, can tell the difference. We learn early by exploring our stories, the true ones and the fanciful ones. Hear a story, feel an emotion, act, tell a story about that action, feel emotions surrounding the story.

What if AI is already doing this? We know that there are advanced efforts to achieve general purpose AI, what if the recent spate of nearly identical tv shoes is an attempt to analyze human responses to stories with minute variations? What if your question is not about the world you are building but the one coming into being all around us?

$\endgroup$

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