Sounds like Okuma has figured it out at the machine level.
many okuma lathes from 80s are still running today, not only rough, but finish ... back then one could have animation, tool load monitoring, conversational programing, hicut ( special function for motion dynamics ), cycle time reduction, etc; it was a marvel, not only back than, but even in comparison with some new machines from 2020s of other brands

they do all stuff in-house, castings, electronics, cnc, parts, assembly, etc, so they can develop and test, without externalizations costs

I understand look ahead for trajectory smoothing, Very good idea
look ahead :
... 1-2 lines : adjust motion accordinlgy with next block ( 1 joint processing )
... 3-4 lines : radius compensation math ( 2-3 joint processing)
... 5 - tens : smoothing ( multiple joint processing )
... hundreds : numerical aproximation + smoothing ( multiple joint processing )

Is the Control system "reactive" or "predictive"
everything that happens until the reaction loop, is predictive ... after is reactive

for example, calculating path is predictive; after that, keeping the machine on that path, is reactive

Can Ai capture the control adjustments and "learn"
( feed back to correlate with each each Gcode block )
of course, this can be done : for example, there are milling robots, thus the spindle+tool is no longer on a rigid machine, but on a robot; being less rigid, will deviate pretty much more than a cnc

for repetitive tasks, the robot is calibrated, in a way that makes it cut a different toolpath, that, affected by deviation, will actually result in a good toolpath

so corelation with each block makes sense as long as the deviations are consistent, predicitible, repetitive

cnc machines have to deal with hazard, like lack of consistency, less predictibility ... and still deliver that accuracy; so functions developed for cnc machines do not intend to map back a program, but be reactive at a high frequency

And then upward into the cam pathing software?
let's consider a linear movement :
... the robot will have to cut a different predefined line
... the cnc will make many infinitesimal small corections while cutting it

thus the robot had no reactive system, it is simple re-programed for a different toolpath

the cnc is reactive

What if we had unlimited compute and software development capability?
hobby sistems, or desktop plotters, etc, use a pc or a laptop a cnc has much more powerful electronics, it must be able to handle those high frequency loops, etc

for example, okuma uses a pc only as interface with the operator, while in the back, there is the electrical cabinet, etc

software and computer power, of 90s, is enough for machining ... nowadays, many have performance relative to the graphic card, and have developed with those; it is a combo that sells, because many wish for the last version and a powerfull computer. Also, some old versions are no longer suported, even if they are still in use, so receiving a file format from a current version will no longer work on an older generation ...well, it could, but they have to sell

i have had the oportuniyy to discuss such things a few years ago, with someone that developer a wordwide top class cam software ... he opened it, showed the basics, then said this is it, from 90s, etc

in the end, is the client the person, the one, that actually puts the price, not the vendor ...the vendor is only happy to sell at that price, and be glad that the client is satisfied; a cam developer only wishes that the client will buy his product over and over again, but they can not sell him twice, so ... when you go down this path, you can discover things

What if we had unlimited compute and software development capability?
so to rephrase, actual computers are ok ... and if you struggle a bit, you can find a software developer to do things how you wish

i will look over that video you mentioned / kindly