What I mean is we have lists of capabilities that AGI needs to have and know how it differs from narrow AI. For example: online continuous learning, perception, generalization, transferring skills between domains, etc. These aspects of the _what_ (rather than how) of AGI have been well researched.
That seems wildly improbable to me. It seems far more likely that anyone who thinks they understand them are grossly underestimating what's involved.
Yeah, I know that I can't prove my position...