From: Jack Richardson (jrichard@empire.net)
Date: Tue Jun 26 2001 - 19:27:35 MDT
It seems to me that the process of monitoring a controlled ascent of an AI might be totalling fascinating in and of itself--particularly if, in addition to the friendly aspects, the AI was given various kinds of subjective skills that would allow it to comment on each improvement it was experiencing. For example, just as a critic would evaluate a new poem and compare it with similar styles, the AI could rate each improvement and comment on what it thought might be a likely next step. The AI could be given a sense of humor and take on various humorous styles. It could even have a whimsical or playful side that could reflect those aspects as it came across them in its explorations.
In this way, every single improvement, aside from capturing any code changes, would have a brief and an elaborate AI commentary on the improvement. There might be billions or even trillions of these, but what a fascinating record it would be.
There could even be questions and/or comments from the programmers back to the AI about its initial commentary on an improvement with a subsequent response from the AI. One could even observe whether the AI acted on specific suggestions or went its own way for following improvements.
This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:36 MDT