I am curious to learn IF (and a big if) an artificial general intelligence will be invented, what will be the roadmap for such an invention? e.g. what would need to be created / demo-ed along the way to ascertain that such an invention might be possible? First it was chess, then image-classification, then Go...maybe a natural language demo will blow our minds soon...what would be the "next demo" (or a set of demos) that would show that such a goal is increasingly within reach?<p>Sub-questions:<p>1) There are some groups (e.g. Strong AI group in Berkeley, OpenAI, DeepMind etc) that are working towards the explicit goal of inventing artificial general intelligence. If a group has to start working on it today, what would be the approach that group can/should take? Is it possible to open a new line of inquiry that will add to this research?<p>2) IF (and again, big IF) something resembling AGI would be invented, would it be in large companies (with lot of data and compute power) or can a smaller group (e.g. a startup) approach such a lofty (and nebulous goal)?<p>Does HN have any thoughts on this?
Personally, I'm looking forward to the day when Automated Theorem Provers can compete with top high school students on math competitions. My guess is that the way to go is to combine resolution theorem proving with intuitive pattern matching (via deep learning), similar to the way AlphaGo combined monte carlo tree search with deep learning.