A book is a tool to help us connect with others' experiences. Colorization and other techniques—done sincerely and as accurately as possible—are additional tools that can accomplish this in other ways.
Altering primary documents to make them more palatable to modern audiences is great for emotional engagement and attracting attention, but it risks creating false impressions of what history actually was like and how we know what it was like. "Hamilton" is a powerful piece of art--today--but it's not accurate history in an academic sense.
(Well, I should say it's not history of Hamilton's time. It will be studied as an important part of the history of our time.)
The folks running these upscaling operations understand the concerns:
> Antic and Kelley aren’t under any illusions that images treated by DeOldify will come out historically accurate, though their reservations are with the practicalities of training a neural network. Making sure colourised films are accurate is “a literally impossible problem,” Antic says. DeOldify uses modern images to train its AI on, he explains, “and we know that's a big weakness, because, amongst other things, it biases people to wearing blue jeans.”
The challenge is, once they create one of these films and post it for the public, the implications and effects are out of their hands.
You might have millions of people watching an AI-changed film and think that what they are seeing is more accurate than the original, since it transmits more information (detail, color, etc). But if that additional information is made up, it's actually not more accurate, and maybe actually less accurate (e.g. wrong color instead of no color). That nuance is going to be hard for a lot of people to understand.
I can confidently say there is a 100% chance that ML/AI-invented content (colours, fill-ins during upscaling, textures, etc.) already outperforms genuine historical content in every possible way in terms of outreach. Just look at any "today in history" youtube or twitter channel and compare it to any actual museum or historical society's website/channel/social media. Orders of magnitude difference.
On the tweets you see replies commenting about how beautiful and vibrant a [false colour, invented content] image is. Hell, those places will post blatantly totally false content (like.. a comedian) and say it's a historical image: https://time.com/5028121/history-twitter-photo/
They are not there for historical accuracy, or accuracy of any kind, those accounts are 100% purely to farm clicks and likes to spam with later.
I think you are missing OPs point.
Better (colours, fill-ins during upscaling, textures, etc.) is not more accurate. Any change is worse from the original because now its modified (however ugly the original is). It suddenly becomes a fake.
It isn't taking a true record and making it fake, it's taking a fake record and making it differently fake.
I would expect tech people to understand the risks here by comparison to using AI to perform image upscaling: you are inventing detail based on what the AI expects to see. For artificial works, this is often great. For, say, a police department trying to upscale grainy CCTV footage, it produces a work of fiction that has the potential to become outright dangerous.
It's fine to present these as works of art, and with the proper context, even as records of history. But pop culture rapidly divorces the art from the context, and I would expect any responsible historian to be extremely clear about the processes use here, even to the point of protest (it is analogous to how astronomers have a hard time expressing to people that false-color photography and artistic illustrations of far-off bodies do not represent reality).
I would assert that even with the errors, people get a more realistic view of the past through these reconstructions than by watching the original footage. The reconstructions drop the barrier of otherness, which greatly diminishes the value of the old footage as a historical lesson for the masses. The reconstructions make it seem like a real place that had real people in it.
Moreover... what terrible, horrible thing are we worried about with these reconstructions anyhow? People will have slightly wrong ideas about what color roofs were in the past? People might have wrong ideas about exactly how slidey people's gaits were in the past? It's not like you feed in footage about a train ride or people riding horses down the street and the AI is adding in cell phones, translating people's speech to modern slang, and changing political posters to features modern politicians. We're fighting videos that increase people's connections to the past because we're afraid they... what, exactly?
By all means keep the original and label the reconstructions, but I see no reason to start complaining about them or running around telling people "STOP! Watching these videos might give you WRONG IDEAS about COLORS! What if that BLUE WINDOW was in actuality MAUVE?" This seems more like Victorians having conspicuous cases of the vapors than a real concern.
I think you make a very important point. I watched lots of silent movies as well as historical footage growing up, and the jerky motion and slow frame rates gave them a cartoonish quality that I still find artistically endearing but which makes history less real to the viewer. For historical figures one admires, it confers a sort of mythic aura, and for ones you don't it makes them seem clownish or misshapen. This impacts written history too, where it's already hard to separate our posterior knowledge of how things turned out from the historical figures of whom we have a visually distorted mental image.
I watched a historical series on WW2 where the imagery was colorized and somewhat stabilized/cleaned up a year or so ago and was struck by the different perspective it offered without all the 'emotional blurring' that occurs due to the technical limitations of the time. It was much easier to relate to events, both good and bad, through a 'happening now' frame of reference that would have been closer to what people experienced at the time, when newsreel footage offered immediacy and accuracy that was new and modern.
No doubt in the future there will be debates over whether HD and 4k footage of today and the recent past should be given the full VR treatment, allowing people to experience current events from the point of view of the participants and so forth.
Linking derivatives makes sense. Replacing original works does not.
And the little things matter.
We will learn how much over time, and ideally we keep originals to benefit from the lesson.
Really, what we do by augmenting originals is lock in one best guess interpretation.
The point of history is for people to go back and see what can be learned, not settle on what everyone should have learned.
On the other hand, unless there was a 4K full color film those millions of people wouldn't even watch the original.
The debate then becomes -- is no knowledge better than minorly-distorted knowledge?
Yes. No knowledge is much better than distorted knowledge.
No. Academic historians are concerned with maintaining the dominant historical narrative that they themselves have created.
> "Hamilton" is a powerful piece of art--today--but it's not accurate history in an academic sense.
Sure, but academic history is no more "true" or "accurate" than the Hamilton play. The historical concept of Hamilton is fiction created by historians affected by their own biases. History is interpretation. It isn't fact. It isn't science.
> That nuance is going to be hard for a lot of people to understand.
That's true for any "history". Nevermind that almost no historical document/artefact/etc is the "original", but the "history" we know is ultimately manufactured fiction. There are facts and then there is history ( which is what is colored in between the facts by historians ).
There is Hamilton the actual person, then there is the historical adaptation of Hamilton, the play adaptation, movie adaptation, documentary adaptation, etc which are all fiction.
"History is interpretation. It isn't fact. It isn't science."
Science is interpretation. Scientists are concerned about maintaining their dominant narrative. "Facts" are very scarce on the ground. (I'm willing to act as if Mumbai exists, but I have no personal experience of it and therefore no positive reason to believe it does.)
In more serious terms, a good academic historian will point to evidence as a way to explain why they want you to believe some statement, and---because they recognize the perils of what they're doing---many of them will complain about potentially misleading modifications to that evidence.
I doubt those youtube historians could possibly do more damage to false history lessons, than all those cheaply made hollywood "historical" movies.
And the youtube historians I have seen, were very good. So mostly on the right side in battle to truth (dramatically speaking) and mostly not on the side of cheap effects for drama, or even intentional misleading for political reasons.
(seriously, there is nothing wrong with historic fiction - but it disturbs me deeply that a) most movies/books are not labeld that way, even if they should and b) even if it is labeled, most people probably do not notice)
There's a British cartoon from 2003 called Monkey Dust, where a recurring feature is completely historically inaccurate Hollywood movies ("Dedicated to all the Americans who died in the early Middle Ages").
No one is altering the original documents. They still exist, right?
"actually less accurate (e.g. wrong color instead of no color)."
That's debatable. Why is "no color" more accurate? The world wasn't monochrome.
If that person is wearing an air force uniform and you colourise it to look more like an army uniform, or a different country's uniform, you've made it less accurate as it looks like they're in a different branch of the military.
If you've got a photo of someone whose hair was white from age, and you colourise it as blonde, you've made them look younger and maybe given them a hair colour they never had.
If you've got a photo of a famous scientist or war hero who had a Mediterranean complexion and you colourise them with more of a pale complexion, you've made it less accurate as it looks like they're a different race.
(Of course, these sorts of accessibility-accuracy trade-offs aren't unique to colourising photos - they also apply to everything from translations of historical documents, through illustrations and selecting what content makes it into books, to questionably accurate history facts shared on social media)
But what about the opposite? What if a B&W photo shows someone with white hair, but his hair was blonde. Now the colorized photo is more accurate than the B&W photo. I just don't think you can claim B&W is more accurate absolutely.
> If that person is wearing an air force uniform and you colourise it to look more like an army uniform
Do you understand that the typical consumer of this media cannot tell the difference between air force, and army uniforms from 100 years ago? Get people interested in history, then worry about the little details like the color of someone's hair or uniform.
Get out of your ivory tower and into the real world. Most people's knowledge of history is abysmal. History is boring, as it is taught in schools today.
Your ad hominem attack devalued your whole argument. Argue the merits of his points, not your preconceptions of his mental state.
That argument being that the academic approach hasn't lead to people having a more accurate image of history. It has scared them off and resulted in most of them having almost no knowledge of history. The approach doesn't seem to lead to it's intended goal.
The comment was perhaps made a bit crassly, but I think it's something worth considering.
That's such an arrière-garde cause. It doesn't take too much time nor too much effort for any self-examining historian to understand that you can't create a "right impression" of history at the level of the general public. It will always be partial, truncated, falsified. Your cherry-picked period is misunderstood by the general public ? Too bad, that's part of the job. The best you can hope is making somebody interested enough that someday they'll have to reflex on historiography, philology and epistemology.
So they want to ban textbooks and force kids to learn dozens of languages to read primary sources and to travel to places of their origins?
Is it wrong to extrapolate the missing pieces to make a complete skeleton? Some experts are able to make really good academic guesses on if the dinosaur species in question had feathers or scales, the size of its lips and toungue, and sometimes even its color. Is it wrong to create a drawing or a 3D rendering of these guesses?
I suppose, to an extent, it is really important to understand how we can know these things from only a few fossilized bone fragments. But to a layperson it is certainly more interesting to see our best guess of how it looked like in reality. Just a picture or description of the original pieces of fossils in that layer of sediment certainly does not do justice to an amazing animal that once walked the earth.
I suppose you could also make the case that we can get it wrong and a false picture of dinosaurs would enter the popular discourse. But I would argue that it would probably be even worse if experts wouldn’t paint their best guess at the time, and left it to the imagination of fiction media to recreate it for the population.
But a high definition video is what we've naturally been used to "believing in" as being real.
Of course the next generation of people will grow up with deep fakes, so they will not trust anything they watch unless it's digitally signed or something.u
I feel that is the difference.
I’m not saying that’s a bad thing. They were able to capture the fascination of a kid by making the dinosaur look as real as a lion, and that’s really cool. It turned out their recreations were wrong, e.g. the T-Rex didn’t have any lips and no feathers in these pictures, but now we think they had both. That’s OK. Growing up I learned that they don’t know everything, and there are still things left to discover.
I bet a future teenager will be astonished to find out they didn’t have high quality color film in the early 20th century, that the films they’ve seen from that period have been colored in using modern technology. Just as I was astonished to find out that they didn’t actually know how dinosaurs looked like, but were able to guess using varying techniques.
It's possible to see recreations of extinct animals that humans have overlapped with enough to write and draw how they looked. It's even possible to see real specimens of extinct creations, such as the Oxford Dodo https://www.oumnh.ox.ac.uk/the-oxford-dodo
You don't need much more knowledge, but definitely more than "they don't exist today". But equally with colourised photos/video: you don't need more knowledge than "this was filmed before it was possible to film in colour, so all colours are guesses".
https://www.youtube.com/watch?v=a-UZXBF63z4
They also have a good one on the history of dinosaur illustration:
The original films were not altered in any way. They still exist and were not destroyed or compromised by the process. They can still be studied by historians while others are free to enjoy the colorized versions.
--
I can kind of understand this if it was made very obvious that it was "AI generated"; that is, the colours are totally invented and made out of thin air, that spaces might be filled in with random computer generated content to support upscaling and make it look better.
I don't want a machine-generated random filler content to be repurposed for "oh, here is a source video for the skin colour of specific people back in the day" or "you see those marks in the corner (that didn't exist in the original source), those are hieroglyphs"
AI does not mean you magically get colour restored to an image that did not have any in the first place, it is invented, machine created content that potentially has absolutely zero bearing on the real world.
Additionally, stuff like this ends up ranking higher than the original source content; preference is given to HD+ videos in search, "viral cool colourised version of the past" on instagram will rank higher than "actual history"; it is potentially a large problem when you have people that don't actually know how to separate genuine primary source vs computer generated content.
Or, alternatively, back in the day there were these shows called "situation comedies" or sitcoms. One of them was "Three's Company." Many of the stories of the episodes of Three's Company were taken more or less directly from Shakespeare's comedies. Is it elitism to suggest that seeing a production of Shakespeare's plays is a better way of connecting with what he was trying to say than watching a '70s sitcom?
The play is pretending to be a Shakespeare production when it isn't, and will give people unrealistic views on the history, whereas the sitcom won't do that
I visited the Vasa museum in Stockholm a few years back. The Vasa is an amazingly preserved shipwreck from the 1600s. There ship is mostly original and looks like bare brown wood like everyone imagines a shipwreck to look like.
However, they have a model and reproductions of parts of the ship showing how they were brilliantly painted. The tour guide made clear that the aesthetic beauty and conspicuous consumption on display was as much a part of the kings arsenal as the cannons. It really changed the way i looked on the past.
The article unfortunately doesn't point out why this is a naive take.
If you upscale an image for display on a 4k display, you can:
1. Leave an aliasing artifact intact.
2. Remove an aliasing artifact.
Now, suppose you have two clients-- one making a documentary about the moon landing, and another making a documentary about moon landing conspiracies.
Pick your favorite "sincere and accurate" ML algorithm-- it cannot generate a single image that will be suitable for use by both clients.
Perhaps "straining and toiling" is hyperbole. But I don't see an alternative to careful, deliberate study of the primary sources of history.
Also-- keep in mind that was a didactic example. When you talk about colorizing films or cartoons, things get quite muddy.
If you give people black-and-white film to watch, people are guess the colors that are there based on what they are familiar with anyway, so this has a chance to be more historically accurate than that.
I think there's a very similar pattern in tech with developers who hack on apps in 'proper languages' like Rust and C++ dunking on developers who write JavaScript or PHP because those aren't seen as 'true development' because they make things 'too easy'.
With respect, historians who are involved with archiving aren't grinding these historical points fine because they're 'rooted in elitism'. Even if some are elitist (some people are), that's not what the profession of archiving is all about. Snowwrestler before me aptly sums the matter up. I'd also offer the following comments but I tackle some of the reasons why archivists do what they do from a more technical perspective.
Archiving is a very complex business, especially so if historical context is to be preserved with fidelity. Moreover, as we've seen from both the article and these posts, that historical 'fidelity' is defined or set by the different contexts in which the archive material is used. Simply, people will use archived material in many different ways according to their individual requirements; furthermore, one's requirement will also likely vary depending on one's circumstances or surroundings. (For instance, a curator will likely require the highest quality of reproduction for a museum exhibit and a lower one on his/her smartphone.)
Here, I'll mainly confine myself to archived images as per the article, however these issues could also involve paintings and other works of art, and even objects (have you noticed how the value of a rare coin dramatically decreases in value if you are silly enough to remove its patina?)
Back to images: Ideally, for preservation, we'd like to replicate an image perfectly and end up with an identical clone—and the only way to do that is copy the whole physical object molecule-for-molecule. Clearly, this is impossible, so what’s next?
The limits of current technology means that essentially we are limited to capturing a planar reproduction of the image by photographing or scanning it from directly above or by projecting light through it onto some recording medium. So what are the criteria for doing this, that is if we wish to copy every trace or single bit of information from said image?
1. First we have to acknowledge that even if we were able to record every bit of information that's available from this planar view/projection (which is very unlikely), that it is not all the information that comes with the image. Being a physical object, it has thickness; it has a photographic emulsion; its emulsion is chemical and we may wish to know how it was made. Is the emulsion only blue sensitive, say a collodion emulsion, or is it new enough to be either orthochromatic or even panchromatic? What is its physical condition, has it been damaged, or scratched, or has it been repaired, and how was it stored, .etc (as that extra (metadata) information tells us a great deal more about the image and its history than just the image itself)? Clearly, those who require this ancillary physical information are likely to be different people to say casual YouTube viewers—this is an important difference.
2. Incidentally, archivists are also concerned not just the intrinsic nature of the image itself but also aspects about the way it was viewed, its environment, etc. For example, did the original film stock have a tint or was it completely neutral? Has the film changed in color since the print was made, does it now have sepia tint? When film was first viewed, what was the brightness and color temperature of typical projector lamps of the period? This information is important as it will influence how we end up viewing the film today,
3. To capture every detail of the image requires some very stringent requirements and often it's not possible to meet them. If we attempt to capture every detail to the point that the law of diminishing returns has determined it's not worthwhile going further, we'll almost certainly never get to that point. Even today, it's almost impossible to scan even 'ancient' 35mm film to the limit of its resolution and density because scanners just don't exist that are capable of doing it in any practical way. Sure, for still images, you could use a $100k+ drum scanner that uses photomultipliers but it's a very expensive and extremely tedious and messy process. For moving images, you could use a telecine that also uses photomultipliers but I've never come across one that's capable of exceeding the limiting spatial resolution of 35mm film made circa 1900 (some of this film was remarkably sharp)! (And these days, every Telecine I've seen uses lossy compression to store images.) Sure, one with suitable specifications could be made to order I suppose but I hate to think of the cost (the fact is most people just don't need to reproduce every bit of info that's on film)! That said, the film may not even be the limiting factor, the image itself could be out of focus or otherwise deteriorated (but you have to allow for best case).
4. With still images, when scanning, you need to consider the limiting spatial resolution of the film as well as consider the Nyquist limit of the film/scanning combination. For images, this can be a complex matter (especially movies) but theoretically an ideal scanner should have around double the resolution of the image (it's essentially the same reason why 44.1kHz is used in CDs to record audio that only goes to 20kHz). Again, no scanner that I know of outside drum scanners comes even close to these figures).
5. Therefore, from the outset we have to compromise. Whether current-day best performance is satisfactory for most archivists is moot. I'd suggest, given the chance, that almost every archivist would keep the original film after scanning so that at a later time a better facsimile can be made. Of course, with disintegrating nitrate film this is not likely.
6. OK, we have the best photographic reproduction we can muster, so what comes next? First this, true-as-is-possible image is the master copy and it must be distinguished as such from any later reproduction that 'enhances' the image in any way—irrespective of reason. There should be no argument whatsoever about this (I doubt you'd ever find a respectable archivist who would disagree with this logic.)
6. I see no objection to image enhancement so long as it is made clear that the image is enhanced and that it not the original or master copy. It's a big subject so I'll try to keep matters short here. Essentially, the future of image enhancement has hardly begun. In future expect to see images such as those horrible 16mm grainy WWI battlefield scenes come to life with enormously enhanced resolution as well as having accurate 'true' color—color that's calibrated to real-world object.
7. AI will make this possible by 'learning' the properties of the objects in these old movies then applying new data to the existing images. Moreover, it's already beginning to do so now. What about the accuracy of the color you ask when we had none to start with. In short, AI will scan millions upon millions of objects including thousands of items from WWI and it will do so in color and in high resolution and even in 3D. To reduce noise and film grain, it will also scan multiple instances of the same footage and compare side-by-side frames both within and across films (and also from other source material such as related hi-res B&W photos—for example, Frank Hurley's famous, wonderfully-sharp plate camera photos of WWI). It will then match and analyze all this information to provide both stunning and realistic images. But as I said, these are not the originals, and they serve a completely different purpose altogether!
There's much more to this story (such as making almost flawless copies of famous paintings including using identical paint pigments, paint textures and brushstrokes as well as keeping existing damage and crazing marks for reproduction authenticity). This technology will act as a means of insurance against damage, loss etc. but I've not time to cover more about it here.