In my estimation we have two minds which manage to speak to each other in dialectic through a series of interrupts. Thus at any one time one of these systems is controlling master and inhabits our consciousness. The subordinate system controls context which is constantly being "primed" by our senses and our subordinate systems experience of our conscious thought process( see thinking fast and slow by Daniel Kahneman). Thus our thought process is constantly a driven one. Similarly this system works as a node in a community and not as a standalone thing.
I think what we have currently is "artificial thinking" which is abstracted a long way from what is described above. so my question is "are there any artificial intelligence systems with an internal dialectical approach and with drivers and conceived above and which develop within a community of nodes? "
2 Answers
There are a lot of systems that follow the ancient maxim: "Always two there are; no more, no less. A master and an apprentice."
In reinforcement learning a class of such setups is called Actor-Critic-Method. There you have a master, whose duty it is to create feedback for the actions of the apprentice, who acts in a given environment. This would be comparable to how a human learns some physical activity, like playing table tennis. You basically let your body do its thing, but your consciousness evaluates how good the result is.
The setup of AlphaGo might be even closer to Kahnemann's system 1 and system 2. AlphaGo has two neural networks which provide actions and evaluations (system 1, fast, intuitive, etc.) and the Monte Carlo tree search, which uses these actions and evaluations to prune a search tree and make a decision (system 2, deliberate, logical).
In the end, this kind of structure will pop up again and again because it is often necessary to do some kind of classification or preprocessing on the raw data before your algorithm can be run on it. You could frame the whole history of gofai as the story of how scientists thought system 1 should be easy and system 2 should be doable in a few decades, where the reality is that we have no idea how difficult system 2 is because it turned out that system 1 is extremely difficult.
You could argue that some Multi-Agent System approaches do, and some systems based on the blackboard architecture could conceivably fit this regime as well.
