In my work with artificial intelligence, thinking about the borderline between a perfect algorithm to perform a perfect action and the superiority over the human brain, I reached an area that made me think about the concept of perfection.

Hyper smart applets

In 1996 we were trying to design hyper-smart applets (applets was concept in Jave programming language). The idea was that applets travel through the internet autonomously, find receptors in routers or computers and have dialog systems in connected machines. It was a similar idea like smart contracts in a blockchain, but far more advanced in its autonomy. We were philosophizing: how far could we bring this? Would these applets become one day self aware? Would they be able to execute tasks not only based on our direction but also be influenced by incidents that occur while traveling in the net? For instance could they find their best path virtually on their own? Today’s AI discussion are a real dejavue for me. One thing we discovered at that stage: We have to allow the autonomous applets to make mistakes. We cannot predict every move and every situation. We have to add some sort of error correction mechanism and being aware that mistakes will come from probability of success calculations. This was the first time I realized that failure is not something we humans have because we are dumb – we have the behavior to make mistakes by purpose.

AlphaGO Example

20 years fast forward. The AlphaGo project 2016 showed very well that the AI system that learned to play GO, was incapable predicting every possible move in order to make a perfect moves. The variety of moves are so vast, that all computing power on earth could not perform this task in the next 10,000 years. So what AlphaGo did is taking a shortcut and calculates probabilities to win in percent from several possible and more logical moves. Obviously the way it does it was superior to how we humans can do it – but it is far away from being perfect (perfect as we humans use that term).

Permutation

I was reminded that permutation is a strategic concept of nature. We humans see permutation as a “mistake”, because we don’t know much about mistakes. We “instinctively” try to avoid mistakes at almost all cost. On the other hand we are triggered by rewards and any success is rewarded by our brain with “rewards points” also understood as joy.

Perfection

Humans are definitely imperfect. Our imperfection is mainly noticed by the mistakes we make and the joy we have when we succeed. However, the more we think about the reasons why we ‘must make mistakes’ eventually we come to a point where we understand that imperfection is the perfect scheme of development. The fear, that with the rise of AI ‘we are all gonna die’ because we are soooo imperfect, is definitely simply another mistake.

Imperfection Oxymoron

A “perfect system” that still develops makes the ‘perfect’ system being previously ‘imperfect’ or the development is a decay. Development or evolution including permutation can only be considered evolving if it is not perfect. The state of ‘perfection’ is the end of evolution or development.
Imperfection is the only thing that is perfect!

The post Imperfection Oxymoron appeared first on Axel Schultze.