The internet offers vast resources if you can read (and “you” are a machine)
We are assuming the system can read accurately, and in detail. Orion can do that. New vistas of self-directed learning open with this approach.
Knowledge expansion starts simple. An unknown word – is it just a misspelling? No, then is it an infrequently used word of English, or a specialised technical word, or a made-up word? Medicine is full of words made up on the spot out of prefixes and suffixes – other domains less so.
Starting from a spellcheck, Orion ramps up the resources to find about a word or a concept. To do so, it stops what it is currently reading, and starts to read other textual materials -a dictionary, a technical dictionary or encyclopedia, a list of drug names (74,000 at last count). It doesn’t stop there – it also checks on concepts – Brown syndrome (which is definitely not brown syndrome) or is it idiom (what does “stitch in time” mean?).
Maybe it is a sense that doesn’t match the modelling – if justification can be found, extend the modelling (need to be careful here – there are a lot of mistakes in external sources – flag it for review).