The Alignment Problem A Feature, Not a Bug!

The Alignment Problem: Why Conflict is Built into Life -- Lessons from "Greater than Ourselves" by Amnon Katz. The Alignment Problem is not unique to AI. Humans have never been aligned! People don’t think about the alignment problem very much, but it occurs in every single relationship. For instance, when I was getting ready to adopt Bow and to teach him language and how to read and to write, it never would have occurred to me that he could do it, but he wouldn’t want to display or show that he could do it. Now that is an alignment problem because what motivates me and what motivates Bow is different. I wanted to prove some things scientifically. I wanted to make a name for myself as a researcher. I wanted to show the world that chimpanzees had the capacity for language and could read and write. But what did Bow want? None of that really appeals to him. And when Bow heard that it was possible that if people found out he could do what he could do, another researcher stated, ‘you know, if they knew he could do that, people would try to kill him.’ Well, Bow was little and he heard that and he thought that wasn’t a good idea. So, ever since then, his motive has been to keep that a secret and to keep me working with him, taking care of him, and being committed to him without any regard to the mission of my research. Is misalignment a problem because we all want to have collective, centralized goals that everyone agrees upon? Then why does centralized control always lead to unintended consequences? And why do natural systems use built in conflicts as quality control? My father, Amnon Katz, wrote about this in "Greater than Ourselves."