The Truth Is Blurred?

…

Recently, I spent a whole weekend “enhancing” old photos with the help of an AI application. The offer had popped up on the Internet, a trial subscription for 48 hours, just 99 cents! I couldn’t miss such a bargain.

Me, the older lady who dismisses AI so far successfully as stupid nonsense couldn’t help but dip a toe in the A1 waters.

Especially since I was preparing a lecture on old upholstery techniques and wood restoration. Some of the photos from the 19th century were of very poor quality. Annoying cracks, creases, and stains, some of which came from newspapers and correspondingly consisted of nothing but pressure points. So why not let an “artificial intelligence” do the ‘job’ in the hope of better quality?

I have to admit. I was thrilled by the first attempts. The scenes were suddenly razor-sharp and face and furniture details were easily recognizable, especially since the program not only “improved” photos but also scaled them -or upscaled them to a higher resolution. So everything was good?

Strangers

On the second day, I was done with the historical photos, but because I still had one day left for a trial subscription, I rummaged through my computer for more pictures that I could improve. I found a folder with pictures that I had from my younger years. I dressed up as a “hippie” at carnival as a ten-year-old, or on a school trip at the age of twelve. Pictures of my grandma and friends. Most of them were taken in the 1960s and early 70s at the time of the first color films, and the quality was correspondingly miserable.

Clearly a case for AI improvement! And indeed, at first glance, the results were good here as well. But something was strange, almost horrible. The faces, now sharp and easily recognizable, showed complete strangers! The program had turned my grandma, my friends, teachers and neighbors and myself into people I don’t know!

This is me according to AI

…

With a little research, I could have predicted that. Lifeless things – furniture, flowers, clothing – can be extrapolated from incomplete or blurred source data but AI has to fail when it comes to faces. After all, software built according to the principle of Chat-GPT works in such a way that missing information is gradually supplemented with the most likely information that follows but logically, this can only lead to acceptable results if it is a generic item. In other words, something for which there is a pattern from which one can “copy”, so to speak. The uniqueness of a person cannot be calculated from incomplete data, precisely because he or she is special. It is not the pattern that is decisive, but the detail.

Not even the dog is right in the picture. AI just gave me a different dog. What’s up with that? However, at least it got the hair right…kinda. Please, disregard the chaos in my workroom, it’s always like this lately…sigh.

…

…

Notch in the cabinet

Let’s take a cabinet that can be seen in an old blurry photo. It may have a notch on the lower right side that somehow got there and distinguishes it from other cabinets. If this spot in particular was not clearly visible in the photo, the AI would not portray it, because there is no such notch on any other cabinet but that wouldn’t be bad for the result. In the AI-enhanced photo, the cabinet would still be the cabinet, the missing notch would mean a small deviation from reality, but it would be negligible if it didn’t cause irritation. With people, on the other hand, this is exactly what cannot work. If you change your face (and especially your eyes) even marginally, people become someone else.

…

…

It would be tempting to postulate a superiority of humans over “artificial intelligence” at this point but be careful. Are we humans really fundamentally better at recognizing people or do we not actually proceed in a very similar way when we meet another person?

…

…

The image we have of others is inevitably incomplete, contains gaps or is blurry, and the resolution is often low. Therefore, we have no choice but to fill in the gaps by making assumptions and extrapolating from existing information. We may be taking a more differentiated and complex approach than AI of today’s generation, and above all we can also pay attention to the specific circumstances and individual aspects. I know what a person from my family looks like, whether they are ten, thirty or fifty years old in the photo, whether their hair is short or long. I like the biblical wording, according to which having sex with a person is a synonym for “knowing” him.

Recognizing limits

But even this recognition is not complete. We are better than an AI, but not fundamentally different. In the future, there will probably be computer programs that can assign people in old photos to certain individuals, and then make the additions accordingly based on more precise data. Then the person in an AI-enhanced picture may look so similar to my real sister that I no longer recognize the fake.

However, there is one important difference between us humans and AI, namely that we can reflect on the limits of our cognitive ability. We know that we can never look one hundred percent into another person (that’s why religious people say that only God can do that or non religious people use the universe). We also know, if we are honest that we never fully do justice to others, that our judgments about them must actually be subject to reservations, because they may be wrong.

So with a heavy heart, I took all AI-generated photos that depicted recognizable people and deleted them. It is better to spread a blurry and incomplete picture about other people than a false one.

Am I the only one who is not into AI? I find it scare how reality can be altered. I don’t think the truth should ever be blurry. I follow blog with AI pictures and while they seem attractive and interesting at first, then the pictures quickly all look the same.

Maybe I am too old or overly critical?