[EliasY]

I believe before claiming AGI is possible or impossible, one would need to define the operational features as well as the properties of what an AGI system is or can do. The primary problem with modern day ML research is that all of the folks that do the research including the major labs think that using one or two primary algorithms is enough to simulate general intelligence. But to think that an algorithm or two can have the ability to solve the hundreds of operational requirements needed to fully emulate intelligent behavior is misguided. What are these requirements, you say? Let's start with language. To substantially solve language understanding, you would need: physical world models, quantitative processing, long term memory, working memory, theory of mind, a discrete situational simulator, plan understanding, detection and generation, language grounding, functional and behavioral models of physical objects, temporal representations of events, affect and emotional processing, reflective understanding models and so many others.