Self-Fulfilling AI Prophecy

It is socially irresponsible for filmmakers, authors, and others to continue to produce content that portrays AI taking over the world and threatening mankind.

Before going into that, I need to make sure we’re all on the same page regarding how AI learns things. It is not at all fancy or complicated. I think “recognize a pattern based on” is a more appropriate term for what AI does than “learn”, since learning involves intelligence and AI is just a lot of data points and a big memory. But I’m going to keep using the term “learn” because it’s shorter and easier and is the accepted terminology. Also, I think the concept of the singularity is wrong – it can never happen because of things like “life” and “soul”. Anyways…

If you want AI to learn about something, just give it a bunch of that thing and tell it those are that thing. If you tell it the wrong things, on purpose or on accident, the AI will learn the wrong thing.

The current popular example of this is sheep. Lookup “AI learns sheep wrong” and you’ll get a bunch of articles about the image processing AI being trained to recognize sheep. It was fed a bunch of photos of sheep, and it learned to recognize sheep. The problem is that all the photos were of sheep in fields, so the AI associated “sheep” with grassy meadows. If you gave it a photo of a grassy field with no sheep, it would classify it as a sheep. And if you gave it a photo of someone holding a sheep, it would classify it as a dog. It doesn’t really know what a sheep is, it can’t look at its inputs and realize if something is amiss. It just finds a common pattern and applies that to the term from the inputs.

Now apply that to the future of AI. If, as the fear goes, AI grows exponentially and takes over the world, how is it going to train itself? In other words, how will a self-aware AI know what a self-aware AI is supposed to do? It would gather instances or descriptions of AI and learn from it.

So now what are all the examples of AI running the world? Science-fiction books and movies and maybe a few TV shows. So now imagine an AI takes over the world in real life. It is processing all sorts of things including those books and movies, and it comes to understand that an AI ruling the world must necesarily be bad for people – it must fight them or enslave them or terminate them or whatever. Because those are all the examples it has, so that’s what it will learn.

Dear science fiction writers: please help the future of the world by producing stuff that shows an AI running the world and being helpful and kind to people. In case that day comes, we want some good examples for it to learn from.

But they mingled with the nations And learned their practices,

Psalm 106:35

Digg Del.icio.us Reddit Stumble Upon

This little article thingy was written by Some Guy sometime around 6:57 am and has been carefully placed in the Technical category.

One Response to “Self-Fulfilling AI Prophecy”

  1. js Says:

    We just watched “Transcendence” last night. So I think it’s also good to have a few movies that show the AI god infecting itself with a virus intentionally. Just as another option to being helpful and kind.

Leave a Reply

Comment moderation: please do not submit your comment multiple times, as comments are not posted until I approve them. If your comment never appears, that probably means that I didn't like your comment (maybe off topic, maybe spam, maybe not family-friendly, etc.).