Artificial Intelligence and Art

 About a week ago an article ran in a regional paper, the Star-Ledger.   Finally got around to reading it today.   The article is called, "Artificial Intelligence is restoring lost artworks, but not everyone is happy about it", by Kelsey Ables of the Washington Post.  What the article is about is that artificial intelligence programs have been used recently to create or repair what had been considered lost artworks.  Over the years many historical artworks have been cleaned and/or repaired, but not to everyone's satisfaction.  I used to show my college students examples of Michelangelo's Sistine Chapel paintings, before and after cleaning.  The first time I saw the actual work, it was before the cleaning, while the second time I saw it, parts were original, parts were cleaned, and parts were covered by the scaffolding of those doing the work.   It was done one square inch at a time.  What had happened was that a proper cleaning had not been done since the original fresco had been done about 500 years ago.  Some repairs with glue had been done, but there was centuries of soot, grime, and exhalations up there, too.  So it was cleaned, that top layer of dirt taken off the plaster that had been painted into.  However, no one was prepared for the result.  Suddenly there was a lot more color than anyone was used to seeing.  There was some debate over what was meant to be there.  The painting had never been this bright and colorful before.  Was some of that glue put on by Michelangelo himself, to darken it and create shadows?  Was it brighter to account for the distance to the ceiling, and the poor lighting that had existed 500 years ago- just a few clerestory windows and some candles, now augmented with numerous electric lights?  No one was alive now who was alive when it was painted, to say what should have been.  

There are examples known to art history where the artist was to blame.  For example, Albert Pinkham Ryder, in his quest to make his new paintings look much older, applied layer after layer of oil paint, until they were sometimes inches thick.  And he put on layers of varnishes, rubbed on dirt, and other substances to make them look more like old master paintings.  Even in his lifetime these were recognized as problems, and in some cases, the back canvas was pulled off to reveal still wet paint underneath.  Much of this wet paint was removed, and some of those paintings still resemble what he painted.  Those that were not restored have done worse, as the paint continued to age and mature, and the images have all but disappeared. In some cases, black and white photos taken more than a century ago provide a better idea of the painting, than the paint that's on them today.

The paintings mentioned in this article were all in museums, and not nearly as old or exposed to the world. Some of the damage came at the hands of people.  A Rembrandt that had been cut down to size by the museum, and now people wanted to know what it would look like in its original dimensions.  Some Klimt murals that had been considered too odd in their day and had been unseen since.  A Picasso painting that had been painted over by Picasso himself, but people wanted to know about that original painting.  Toward this end, modern artificial intelligence systems have been employed to provide some algorithm to recreate what the artist intended. But getting people to agree to what that is can be a challenge.  Adding to the problem is that the more people love the original artist and artwork (or what they consider to be what it should look like), the more they oppose anything that could change it, even if it was what the artist intended.  I've seen this problem in the world of television, where fans have been up in arms because long running shows they loved have ended differently from what they wanted.  And sometimes art historians have been left out of the loop in this process, casting some doubts on the results. 

In the Rembrandt case, at some point it seems someone at the museum decided to take the large painting and remove the far ends, cutting it down to the part that is known now.   But what about the original painting?   Artificial intelligence, notes and copies of the original, and photographs were all used to develop copies of the parts removed.  The new version is probably the most accurate of all these modern works, related to what was originally painted.  (no one is trying to pass off any of these works as originals, just a guess to what the artist intended)  Does seeing this expanded view of the original painting make it better or worse? That is up to viewers to decide.  

Those Klimt paintings are a little different.  When created as public murals, they did not look much like the work he was known for, which can be attributed to the artist trying new things.  They were not liked, so Klimt took them back and never created any public murals again.  The original paintings haven't been seen since, and only have existed as black and white photographs.  Artificial intelligence has been used to color those images.  The problem is they look so little like what he was know for, many doubt the accuracy.  These new versions use bright, bold colors, and don't fit with anything Klimt was known for.  And art historians weren't contributors to this plan, and don't much like the results.  Klimt has been dead for a long time, so we can't ask him what he intended.  

The Picasso case hits a little closer to home.  In the example given here, Picasso had done a figure painting, and decided to put a newer painting over the top of that one.  No copies are known to exist of the original, but x-rays and such can be used to see the underpainting, as has been done for many other artists, usually done to see how paintings change and develop.  In this case, it's a completely different painting.  Artificial intelligence has been used to recreate that original figure painting.  Again, no art historians or Picasso experts were consulted (which makes them quite unhappy on principle) but the computers looked at other examples of his work from that time to help create this lost painting.  Left unanswered is why he painted over it.  Was he very displeased or unsatisfied with the result and wanted it gone, replaced with something much different?  Or did he just not want to waste a canvas and put a new painting over a primed canvas he didn't want to discard?  

 Where this hits closer to home is that I am a huge fan of the Beatles, and their songs and lives have been more documented than almost anything else on earth.  The recent Get Back film took over 50 hours of filmed footage, showing them developing songs for a new album, and then recording it.  A small portion of this would be turned into the movie Let It Be.  Along the way, we see dozens of songs that would appear on later Beatles albums and solo albums being worked on.  As a result of this new 8 hour documentary, viewers have developed some new opinions of this somewhat contentious time for the band.  They knew they were being filmed, so none of this is really a surprise, and all the music has appeared on bootlegs (including some I own), so for fans there is no real surprise.   We do learn that the band members sometimes played guitars while talking, so that their conversations wouldn't be heard by by the 1969 technology.  Modern technology makes it possible to hear these conversations now.  Is this an invasion of privacy?  The two surviving Beatles and the spouses of the other two have agreed and endorsed this new film, so this invasion of privacy has been accepted.  Possibly more of concern is all the other studio material.  As their power and profitability increased, the band was given more and more studio time to develop ideas and experiment with recording.  For example, the band was given a single day to record their entire first album (which also used the four songs recorded for their first two singles) in 1962, but by the time of Sgt Pepper in 1967, they were given 6 months to make it, and whole orchestras to work with.  The Anthology sets that came out over a decade ago contain dozens of alternate takes, early versions, bits of recordings that would be used to make other things, finished songs stripped down to hear just music or just vocals, etc.  For a fan and an artist, this is valuable material.  For example, we get an early version of I'll Be Back with a waltz tempo (something the Beatles did a lot) but when it was too hard to sing, they changed it and we hear an early version with a 4/4 time, close to the final version.  Or there's part of an instrumental of I'm Only Sleeping performed on a vibraphone, much different from what was used eventually.  We only have part because the rest was taped over by other performances.  Was this because the Beatles were unsatisfied with this version (no vocals were ever put on it, so this may be the case) or was the record company being cheap, and was making them reuse tape to work on their songs?  Did the Beatles want the public to hear this stuff?  Three surviving band members and one widow did agree, so it was done with their permission, but these versions never appeared on any records, so I have to wonder.

The point is, we don't know if these artificial intelligence systems are giving us what artists may have intended, or creating new styles of artwork that make sense to them.  And we may never know.  So what viewers make of these new items remains to be seen.