Mark S Abeln
Forum Pro
The article stated that the AI was trained via input from human judges, and the AI was trained so that it would match the average and standard deviation of the scores given by the humans. The AI was validated by comparing its judgments to human judgements of another set of images. This system was not programmed with 'rules of thumb' such as the Rule of Thirds, and only had a few preconceived ideas about technical merit, but rather followed the lead of the human judges.I call BS on that... A computer is in no way equipped to judge aesthetics or especially the context and emotional power that a photo can have. Even as far as technical merits I'd say that it doesn't make sense to use this kind of thing as a judge. A lot of great photos are technically "wrong" but have broken the rules for a specific effect and that's what gives them their power...
As people can and do agree to a large extent on things regarding aesthetics, then this agreement is evidence that aesthetic judgment has an objective quality about it, and therefore in principle can be done on a machine as well. But this AI also takes note of disagreements among the human judges: the AI can state that a particular image is probably good, but not to everyone's taste.
Regarding emotions, while these are subjective qualities, a lot can be objectively known about them, as they are basically motions of a person towards the good and away from evil. Context is another issue altogether, and certainly this AI does not and cannot judge images according to context, which would require a full human intellect to do so, and which would vary in time and place and with the individual. Instead, this AI is judging on a highly abstract level, knowing nothing about the objects and places portrayed in the photos.
While we make much ado about the term "artificial intelligence", at its root, AI is mainly statistical analysis of pre-existing, human-curated data. It's what machines do best: analyze lots of data, and calculate statistics quickly.
