Working Up

by Dwayne Phillips

Are we spending too much time and effort on making AI systems “safe?” Perhaps we should allow adults to be adults and move on.

Consider an LLM or whatever that creates text promoting:

• fill-in-the-blank with your least favorite group of people ever
• fill-in-the-blank with your least favorite philosophy ever

Those people, those philosophies are horrible!

Okay, are you an adult? If so, you can easily reject those horrible people and philosophies. If those people and philosophies are horrible, other adults will reject them as well. If some adults accept those horrible people and philosophies, they will suffer and who wants to be like them?

There is some merit to preventing some adults from seeing, hearing, reading, etc. some philosophies. For everyone else, however, adults can be adults and decide as adults.

I am writing about adults, not children. That is another topic for another day of writing.
All the efforts to make “AI safe,” i.e., not ever ever show horrible people and philosophies, are they worth the effort? And “horrible” is a subjective term with a definition that changes.

I recommend adulthood for adults. And I am sure I could find examples where that recommendation didn’t work well.