If you've ever read some of my posts, you know I'm the confrontational type. I've been accused, with good cause, of throwing my knowledge of history or science in the face of people whenever they make assumptions instead of checking facts. It should come as no surprise to you that my MBTI type is ENTJ. Knowing this, I'm always looking to better myself. A friend recently introduced me to NVC (Non-Violent Communication) and it made sparks in my brain.
I'm also very active in the field of AI. I've been working on AI in various contexts : video games, robotics, the exploration of how the human mind works by trying to replicate its mechanisms... and in all this work I've never found a failsafe way to make an AI friendly other that by shielding it from the fact that it exists in an unfriendly world.
Some of my AI work is based on expert systems. If you don't know what they are, a crude way to describe them is that they are programs that :
- Take inputs from their context (goals, dangers...)
- Feed the inputs to a set of rules (fire burns so stay clear, red traffic light == danger, that sort of thing)
- provide an output that is the basis for behavior (crossing a street or not, running towards danger or away from it)
This is the type of AI you find in video games, especially real-time strategy games.
Expert systems are the typical choice for AI's dealing with "higher brain functions" or abstract "thinking" (as opposed to neural nets). Expert systems are inherently based on some sort of representation of abstract concepts (the rules) which means a language is used to represent them.
I normally go with analytic grammar concepts (look-up Noam Chomski) to define simple yet consistent languages for the purposes of AI, but after reading about NVC it now seems to me that working on grammar and language alone may be too limited.
AI innate behavior (the initial rules in an expert system) needs to be crafted using methods that will ensure friendliness. And I believe NVC could be it.
For one thing, it might eliminate segregation between "meat-thinkers" and AI's, which would be a cause for racist violence and Skynet-style AI behavior.
There's a catch, however : for it to work, NVC must be practiced by all parties interacting : not just the AI's, but the humans as well. This turns the problem of AI friendliness from a technical problem to a societal problem.
I would like to know if the NVC paradigm is already being considered as a path towards AI friendliness. (perhaps, through NVC-based AI education, not just as the basis for expert system rules design)
I'd really like to hear from people who practice NVC.
Nefastor
Edited by nefastor, 21 April 2008 - 04:42 PM.
