>>11366074>>11366079>>11366074Honestly I think it will be devised anywhere between 5-20 years. The main limitations today are the small number of active researchers in the field. Most academics follow a trending trail of modern research and don't experiment with wild ideas all that often.
Secondly, I think that computation resources are a limitation (believe it or not). Most universities have access to a grid system or distributed computation resources which I think is not enough to simulate a fully functional artificial general intelligence system. Either there will need to be a massive amount of funding put into research toward efficient and correct code, or there has to be considered toward integrating some sort of larger-scale distributed computing platform.
At the moment technology is geared toward classification and regression. Most research now is looking at automatic feature selection and representation learning. The pinnacle of this field is AI capable of generating never-before-seen human faces by learning the "essence" (representation) of what a human face is meant to look like. As it stands the full pipeline still requires human intervention.
Identify Problem and Goal (Human Intervention)
Identify Domain (Human Intervention)
Establish Constraints (Human Intervention)
Feature/Representation Construction (Automated)
Learning based on previous tiers (Automated)
Output (Automated)
Interpretation (Human Intervention)
Abstract Learning about Domain and Problem (Human Intervention)
Go search auto-encoder and encoder technology for GNNs. That is as close to general-intelligence as we are. If you think about the path and trend of AI, it should be apparent that the "intelligence" aspect is emergent of an algorithm that coincidentally captures real-world information and truths by simultaneously solving a problem. To capture general intelligence, research must be done to analyze systems that can encode and associate abstract information from varying sources.
