AI isn’t the mirror of humanity
There are articles stating that AI is the mirror of humanity, but this isn’t quite accurate. An AI model just reflects whatever its creator fed it. At the same time, the data used for (general) AI is often what’s available on the internet. In practice it means that AI is the mirror of humanity’s internet.
The interesting question from there is whether the data available on the internet reflects humanity. I’m by no means an expert in psychology or sociology (or any related field) but I think there are two fundamental differences between online data and “real” data:
- There are less behavioral/social cues online and these are an important part to keep discussions polite and thoughtful
- There is a lot of content online to steer our negative feelings since this is an easy way for companies to keep people engaged on their product
The net result is that data on the internet reflects a less civilized, a darker society – as such AI without explicit alignment efforts would reflect a dark humanity.
From there the interesting questions are who gets to decide how to align the models? How do we make sure they are fairly aligned? How do we even measure this in the first place?
Thoughts on this topic and how to better move forward?