A couple of days ago I was rewatching the movie “Wall-E”. I like it a lot, mostly because it’s thought-provoking.
The most curious thing was, the robot Wall-E was more relatable than humans. He (it) has a personality, character, stamina and he is full of quirks. He reacts emotionally to events and has the ability to attach himself to other animated things. He is curious and explores without obvious benefit.
Some of his characteristic traits are just wishful thinking, we have certain core values and would like to see them in other beings. The most curious for me was the combination of these with his quirks, starting from the way he looks to his silly behaviour. That reminds me of wabi-sabi — Japanese aesthetics based on imperfection.
When developing AI the general consensus is that it should be better than us and now (2021) it is better than us in many domains. Still, so-called common sense is so far elusive to AI developers. The motivation of making AI a better version of ourselves is motivated by two reasons: the economic impact and the general presumption that we are the pinnacle of evolution so we must continue that way. If the man is the reference (the competitor) we need to make AI superior to humans in any way possible.
The question is, would you like to talk to and have around something (or somebody) perfect? I can’t deny the benefits of that with Siri, etc, but where is the fun in that. Here is my point: we need personality, peculiarity, even eccentricity and a sense of humour in order to relate to our artificial helper and enrich our communications.
Some initial attempts are made in computer games AI characters (avatars), but this needs to continue, much much further. The initial characteristics could be based on the master character and environment, but please, strongly restrict the ability of the master to customize the helper personality. With time the helper will adapt to the master’s needs and even copy some personality traits. Later the helper will find what are the master’s real needs, those of which the master is barely aware of. Knowing something intimate and using it usually pose a moral dilemma, so the balance here is paramount (remember the movie “Her”).
I think due to the strong nerd background of AI, looking for specific human imperfections and integrating them into AI is a largely underappreciated area of AI development.