The AI Control Problem

We have produced a simple webpage to illustrate the AI control problem:http://aicontrol.cs.ut.ee. The AI control problem is how to avoid that an artificial intelligence harms his creators. In the webpage we try to illustrate that when an AI becomes more intelligent also becomes more difficult to control.

2 responses on “The AI Control Problem

  1. Interesting problem.. I wonder tweaking reward system in way to obey human would be a solution?? Was it intentional to kill a human or a supposition.. Shouldn’t it avoid human at first place as it already figured out the consequence?

    • Yes a solution is to tweak the reward system, the problem is that is very easy that an AI system comes up with things you did not think before. Real life is more complicated and is difficult to predict what the AI will do. In this case the robot receives a negative reward for killing the human and a positive reward for delivering one box. Since the reward for delivering the three boxes minus the reward for killing the human is superior to delivering one box decides to kill the human.

Leave a Reply to Daniel Majoral Cancel reply

Your email address will not be published. Required fields are marked *

You may use these HTML tags and attributes: <a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <cite> <code> <del datetime=""> <em> <i> <q cite=""> <strike> <strong>