Remember how the "experiment" in splitting the judging panels into TES and PCS was kaboshed by the people who had the most to lose power-wise by the split accepted as a failed experiment because the judges were bored?
As I recall, it was the judges who were judging GOEs only who were bored, not so much the ones who were judging components (of which there were 5 at the time).
It's one thing to be talking about using AI for the technical panel's calls, but this, "then overlays the key points of a jump or spin —
As I understand the technology described in the article, at this point it's just designed for use in training, using a single phone or tablet. The coach and skater can plan where to place the camera and where to place the jump on the ice to get the most advantageous angle. That won't be the case in competition where skaters can place their elements anywhere on the ice surface, approaching from and landing at different angles.
For use in competition, we'd want to make sure the recording device has high-resolution capabilities (easy enough) and that data would be available from multiple camera angles (easy in large arenas with potential camera positions on all sides, not so much in some of the smaller venues sometimes used for JGP or senior B events, let alone lesser domestic events).
If there were only one angle, no matter how high the resolution and how unbiased the computer model making the assessments, it would still be common for the view from the camera the AI is using and the view from the broadcast videos to show something that looks significantly different.
I can imagine this issue getting solved for major events, e.g., ISU championships, Olympics, Grand Prix events and Grand Prix/JGP finals, and also nationals of federations with large enough fields and fan bases to hold their nationals in large arenas.
But at what point does the ISU need to demand that a certain level of competition must use this technology and any associated updates in overall scoring process, and where are compromises and reliance on human assessment permitted for lesser events?
To the extent various aspects can be objectively measured, why not incorporate the actual measurements directly into the scoring rather than relying on human estimation and yes/no cutoffs (which may differ from judge to judge) as to what constitutes "very good" as opposed to just "good"?
However, I would ask how these measurements would take the size of the skater into account. If, e.g., Midori Ito and Brian
Boitano get the same height and distance on the same jump, would Ito get more credit because she's over a foot shorter? Or, e.g., Yuma Kagiyama vs. Nikolaj Memola competing in the same event?
the idealized version of a given element — and records the metrics that technical panels typically use." could easily be applied to many, if not most, aspects of GOE: for example, to compare the ideal/one of the ieals a jump/jump combo/jump sequence, like the ratio of speed into a jump, height, distance, flow out speed and distance, and for pairs, distance between skaters and synchronicity to the actual jump, the way the graphics on Japanese broadcasts do to a broader respect. This could easily translate into GOE algorithms.
Some of the jump and spin GOE points certainly would be capable of being measured, given the appropriate technology. "Very good height and very good length" of a jump element or "good speed and/or acceleration during spin" and "maintaining a centered spin" could be determined objectively by machine measurement, in which case the "very good" and "good" designators could be removed and there could just be a raw score. Same with, e.g., average and maximum ice speed or maybe acceleration in the program as a whole or in the step sequence.
This can be true for some of the negative GOE criteria as well, aside from jump rotation and takeoff edge.
How much would body type affect the "ideals" that the AI learns to reward, while penalizing deviations? E.g., if landing positions are to be assessed, would a skater who is naturally bowlegged have a built-in disadvantage compared to skaters who can achieve a nice straight extended free leg, when it comes to the "very good body position from take-off to landing" bullet point. Would Ito have been penalized there?
Many of the current GOE points are qualitative: e.g., "effortless throughout" (maybe an advanced AI would be able to assess this), and bullet points for creativity or matching the music, on all types of elements.
Would we want to rely on the AI's knowledge base of what has been done before to allow it to identify what variations it sees today count as creative? Would we want it to determine how well elements match the music? Even if there is an emotional component to how the skater varies timing and subtle positioning of body parts to reflect subjective aspects of the music? Would AI be able to make meaning out of a particular arm and shoulder position on a jump landing matched with a particular facial expression to express "yearning," for example, as well as a human observer?
Or do we still need humans to weigh in, relying on panels of multiple judges to amplify the limited knowledge base of any single judge?
An AI well trained on the history of the sport and its technical content and variations, and also in the history of music and human movement to music (i.e., dance) of all genres would be a much more sophisticated model than anything that's being proposed here or likely in the near future. Someday in the future, models might exist that possess historical and cross-cultural knowledge beyond what any given human brain contains. But if it's limited to what it's been taught, how well will it be able to recognize innovation?
I.e., should anything that can't be measured objectively be removed from the GOE criteria, and skaters who enhance their elements in this way no longer get rewarded in the element scores for these enhancements?
If that every happened, how quickly would we expect the judges to suddenly not be so bored by judging only/mostly PCS, even when ice coverage, pattern/node depth and complexity, % in each direction, for example, could all be viewed and assessed through algorithms?
Judges never did say that they were bored judging only PCS.
But if we were going to take everything related to TES away from the judges entirely, perhaps it would be appropriate to restore the 5 different components and to build in ways for them to reward creativity, musicality, and intricacy in and out of individual elements more granularly than allowing them no more than three scores to reflect everything they thought about all aspects of the program.
apnews.com
Excerpts:
Please just say "computer analysis".
on those jumps right up until her retirement! She'll be long gone before AI or computer analysis arrives on the scene! 
