Wow wow, what an insightful paper, thanks for the link! Everyone was obsessed with expert systems back then, and foundations of modern ML backdate that far.
However, speaking on assistant systems, I need to mention the point which was intuitively known before, but Oleksandr Nikitin was the one who pinned it down to me: they are not that universal, and to understand one’s needs we need to study this person’s mental “deficiencies”. A system for OCD-like person would be all different from a system for an ADHD-like one, even if both persons are certified sane by a psychiatrist — there is much variance within the norm. Tasks we have at hand now are related closer to he human behavior rather than data ontologies.
(and regarding the user-agent header in the HTTP, everything went south when Internet Explorer emerged, and we started dealing with browser-specific hacks instead of generic capability sets, and it became only worse since then)