When AI Restoration Becomes Recreation: The Problem with Filling in History's Blanks
The internet’s been buzzing about an AI “restoration” of the world’s first photograph - Niépce’s “View from the Window at Le Gras” from the 1820s. What started as excitement about bringing history to life quickly turned into a fascinating debate about what constitutes restoration versus recreation, and honestly, it’s got me thinking about how we’re approaching our relationship with the past in the age of AI.
The original photograph is barely more than shadows and light on a pewter plate, the result of an eight-hour exposure that captured a moment in history we can barely make out. Then along comes modern AI, promising to “restore” it into something crystal clear, complete with detailed buildings, sharp shadows, and what appears to be a fully realised 19th-century streetscape. The problem? Much of what the AI added simply couldn’t have existed when Niépce took that photograph.
Sharp-eyed viewers quickly spotted the anachronisms. Cast iron gutters that weren’t invented until decades later. Shadows far too crisp for an eight-hour exposure (the sun moves, people!). Architecture that doesn’t match historical records of the actual location. What we’re looking at isn’t restoration - it’s educated guesswork dressed up as historical fact.
This hits a nerve for me because it represents something I’ve been increasingly worried about with AI technology. We’re so impressed by what these systems can do that we sometimes forget to ask whether they should do it. There’s a difference between enhancing what exists and creating what never was, and that distinction matters enormously when we’re talking about preserving historical accuracy.
I’ve worked in IT long enough to know that data doesn’t magically appear from nowhere. If the original photograph doesn’t contain information about the texture of roof tiles or the exact architectural details of neighbouring buildings, then any AI “restoration” is making it up. That’s not restoration - that’s historical fiction.
The whole thing reminds me of those documentaries where they use CGI to show what ancient Rome “really looked like.” It’s impressive technology, sure, but it’s presenting speculation as fact. Future generations might look at these AI-enhanced images and assume they represent historical reality, when they’re actually showing us more about the AI’s training data than about the past itself.
What particularly frustrates me is that this kind of work can actually damage our understanding of history. Someone in the discussion mentioned that this is exactly what history educators have been warning about - the way flashy technology can obscure rather than illuminate the past. When we smooth over the gaps in our knowledge with AI-generated content, we lose sight of what we actually know versus what we think we know.
The irony is that the original, grainy, barely-visible photograph tells us far more about the realities of early photography than any polished AI version ever could. That eight-hour exposure time, the way the light falls from both sides because the sun moved across the sky, the incredibly primitive nature of the technology - these limitations are part of the story, not flaws to be corrected.
There’s also an environmental angle here that keeps nagging at me. The computational power required to generate these detailed “restorations” is enormous. We’re burning electricity and contributing to carbon emissions to create historical fantasies when we could be using that same technology to solve actual problems or genuinely preserve the historical records we do have.
Don’t get me wrong - I’m not anti-AI or anti-technology. The possibilities are genuinely exciting, and there are legitimate uses for this kind of image processing. Cleaning up damaged photographs, removing scratches and stains, enhancing legibility of faded documents - these are all valuable applications. But there’s a world of difference between revealing what’s actually there and inventing what isn’t.
Maybe what we need is clearer labelling and more honest discussion about what these tools are actually doing. Instead of “restoration,” call it “speculative recreation” or “AI interpretation.” Make it clear that we’re looking at an educated guess, not recovered historical data. Give people the tools to understand the difference between enhancement and invention.
The technology itself isn’t the problem - it’s how we’re choosing to use it and, more importantly, how we’re presenting the results. History is messy, incomplete, and full of gaps. That’s not a bug to be fixed with AI - that’s a feature that reminds us how much we still don’t know and how precious the fragments we do have really are.
Rather than using AI to paper over the gaps in our historical knowledge, perhaps we should be using it to help us better understand and preserve what we actually have. The blurry, barely-visible original of the world’s first photograph is far more valuable as a historical document than any polished recreation could ever be.