I have spent many nights and hours at work thinking about the implications of artificial intelligence. It has been something which has excited and made people fearful about the bounds in which it has come in the last decade. On twitter, you can’t go very far in your feed without seeing some mention of either fear of a Terminator ending or excitement about how AI will make the world a better place.
Many people are worried about the things in which AI is expected to replace in our world; From low skilled jobs like those in fast food and retail to total automation and AI driven vehicles.
I try not to dive too deeply into the fears of AI but rather I love to look at what it can make better. Looking at the bright side of situations is something I’ve been better about over the last year compared to before which has enabled me to think about concepts more deeply.
I’m not entirely sure this idea has been fleshed out or even discussed much, but I have always been interested in how AIs choose to store memories. We have always known computers store specific data because that is what it was programmed to do, but I wondered how close to human memory and memory optimization could we get with AI.
Recently I worked at a data center and spoke in depth with a coworker about AI systems in games. How would an AI know how to prioritize memory functions if in most cases where it learned something it would just store the whole process of learning instead of analyzing and picking on its own what should be kept temporarily (think short term memory) and what to be keep long term (think long term memory).
We understand through years of research how our brain stores data. Typically it chooses instances to store long term like the memory of burning your finger on the stove, and things to not keep long term like the environment around you as you walk down the street. I have wondered how you could integrate an AI to chose on its own without an intensive base code to tell it what to store and what to not. Perhaps the brain is not understood well enough to translate this into an AI, but I hope that one day we could understand how to do this.
I personally don’t think an implementation of this would cause self awareness in AI, but it could help prioritize and streamline data creation to include only vital information in its long term storage. At the end of the day they are still machines built by humans. But I would be cool to have AI that could learn efficiently without excess useless data.