Note: This article was not written using Artificial Intelligence

I ask all those who know the author to kindly refrain from asking themselves the question whether or not the following was written with any form of intelligence at all. And for those that can refrain from such a folly, the following article is meant for you….

It has become very apparent that a lot of us are asking the question how far will the IT technology craze of artificial intelligence impact our hobby. I don’t want to get into philosophical debates on the theme of how close these things approach actual intelligence with self-replicating capability and self reliance and to what extent they are ‘just’ sophisticated algorithms of data neural networks with vast amounts of data sources to draw from.  Rather than that, I want to reflect upon the outputs of AI and its potential applications to us. In the current fashion boom, many systems are called artificial intelligence, which, although very advanced, are not actually self-learning or self-improving systems.

The boundary, when it comes to ‘artificial intelligence’ (or ‘artificial’ intelligence... or artificial ‘intelligence’ - the attentive reader will surely recognize the nuances of these formulations) can probably best be demonstrated by OCR, i.e. systems with the ability to recognize writing from a source in the form of an image. These systems have been around for some time now, and most of the more readily accessible ones are not, in reality, actual AI. The more common, simpler ones lack a substantial self-learning element. In order for a system to be able to self improve and learn, it needs feedback, where it is told ‘... but the letter you read as Z is actually a slightly skewed 2’, and the system takes that into account when making decisions, and adapts when a similar situation is subsequently encountered. Neural data networks have been able to do such things for many years, and many articles can be found on the subject. And that can already be called artificial intelligence, although ‘self-learning system’ seems more adequate to me.

Still, I don't want to give the impression that I am underestimating artificial intelligence. Development is very rapid and I believe that we will see things within its scope the likes of which we can’t even fathom right now. As with everything, however, I am somewhat skeptical when it comes to hyped up headlines and various impetuous statements. Nevertheless, I can certainly imagine that AI systems will gradually find their applications in a natural way in production technologies that are used in the design and production of models and accessories.

Due to AI being in vogue, we often come across attempts to create published articles with the help of AI. Know that as far as I am concerned, our articles will be written by real authors with real knowledge, and historical facts or technical data will come from our own sources, books, archives, knowledge. This is the case with most of what is, particularly in our historical articles, printed in our monthly newsletter. They are based on the author’s own research, often the result of many years of work, information gathered from many experts from all over the world, and is not information that can simply be extracted from Internet sources and compiled using AI.

Of course, I can't rule out that one of our authors will help themselves by having AI advise them on better sentence formulation, a more graceful word flow - which, after all, these systems can do quite well. However, as far as this particular article is concerned, if you find the style acceptable and with a relatively graceful flow to it, praise should not be showered on AI, but rather on our proofreaders... :-)

I currently see three areas of AI (or advanced IT) that can be used in plastic modeling that I would like to outline with this article and I will discuss them in the following paragraphs (although surely some of you may come up with other possible uses and for your feedback, we will be grateful).

OCR (Optical Character Recognition)

The first area is the study and further processing of historical documents. This is where OCR is very useful, so I won't go into more detail about it as mentioned above, but I will show an example instead.

Excerpt from the combat report of Lt. Francis M. Matney of the 55th FG, on 11/2/1944:

Here is the unaltered OCR rewrite:

I was flying Hellcat White 4 on Lt. Jowell's wing at an altitude of 32,000 feet when 15 plus Me109's and FW190's came in on the bombers below us at 27,000 foot. We immediately dro'ped our belly tanks and dove on them from r o'clock. Lt. Jewell nicked out one Me109 and I stayed on his wing. The E/A started a steep diving turn to the left and ended in a vortical dive.

A few misinterpreted letters are obvious, but after minor editing the text is very usable for further processing. Advanced OCR systems can deal with written text to a certain extent, but the already mentioned self-learning element with feedback from the user must be applied... and you simply have to ‘get used to’ such a system taking into account any given handwriting.

OpenAI/ChatGPT Type Text Compilers

 I don't want to knock these systems or underestimate them. I am playing around with ChatGPT myself, and I was really amazed recently when I asked it for advice on, among other things, an issue with radio navigation, while preparing for theoretical PPL exams. However, it should be noted that although the sources from which it compiled and formulated its outputs are very extensive, they do have their limitations. Sometimes we are given some pretty absurd answers, which this AI is able to offer with staggering confidence.

I tentatively touched on one of my fundamental topics – the air battle over the Ore Mountains of September 11, 1944, with the question of why this battle started. I wouldn't be surprised if the system was not able to provide an answer. Surprisingly (and misleading for the unaware), the answer is wrong, but presented with a high level of confidence:

‘The air battle over the Ore Mountains on September 11, 1944, was a part of a larger operation known as the ‘Schweinfurt-Regensburg mission.’ The mission was an Allied bombing campaign against German industrial targets, specifically the factories producing ball bearings, which were essential for the German war effort.’

Of course, the Schweinfurt/Regensburg mission took place on August 17, 1943…

On other examples, though, the system should be congratulated, and I found it impossible to mislead it.

Question: What was role of Adolf Galland on liberation of Prague?

Answer: Adolf Galland did not have a role in the liberation of Prague on May 9, 1945.

Galland was a German fighter pilot during World War II and a high-ranking officer in the German Luftwaffe. He was involved in many significant battles and operations throughout the war, including the Battle of Britain and the defense of Germany against the Allied bombing campaigns.

However, by the time of the liberation of Prague, Galland had been relieved of his command and was under house arrest by the Nazi authorities for his outspoken criticism of the leadership of the Luftwaffe. Therefore, he was not involved in any military operations during the final days of the war in Europe, including the liberation of Prague.

The insight to be gained is clear - similar systems, whether we agree that they are artificial intelligence or not, can be very helpful. However, we must be very careful about what they present to us and not leave out our own sense of reason.

Historical Photo Analysis and Color Recognition

Systems for defining the colors of objects in historical photographs will certainly find considerable applications in our field. Not that I think the end is near for debates over beer between modelers about  the actual shade of RLM 66 or the fact that technology could so easily determine the exact shades of camouflage colors, but it is already clear that the potential for its help is there, and it is significant

Of course, limitations will apply here too, and we must not lose our ability to be reasonably skeptical or critical. The quality of the input material also plays a very important role here. In a completely different way, this system will be able to deal with a high-quality scan of an original historical photograph and in a different way with an image downloaded from the Internet and resampled and color-converted or corrected many times.

The outputs we are offered - and I stress again - we need to be careful with them and apply due scrutiny, interesting insights. And even if the color solution is wrong, the colorization itself in the sense of what the AI sees can draw our attention to interesting details that we would have missed when studying these photos in the original black and white. I can certainly imagine such an aid utilized by our experts dealing with camouflage schemes for kits. But everything in moderation….

As in the previous examples, there is a wide qualitative range of systems in this area. From web applications to professional systems used, such as, for example, color films. You can now incorporate feedback from users and really learn and advance your approach when analyzing shades of black and white photos. One more example:

This example is a portion of a photo of a Bf 109G flown by Fhj.Fw. Hans-Theo Ableiter of III./JG 4. At the far left is a scan of the original photograph, in the middle and at the right are two color reproductions made using two different AI systems. Although the quality is different (one system is a professional film system, the other is a commercially available converter), the color reproductions in both cases match fairly well. The colors of RLM 75 and RLM 76 can be quite clearly distinguished against a light background, which logically should probably be the rather gray variant of RLM 76. Note how each system reproduces the individual spots. The simpler one (on the right) made 74 even from some stains, which the first system identified as RLM 75.

So that it all isn’t too easy, I'm attaching one more picture showing how AI coped with color resolution in another case. Many of you know that this airplane is supposed to be all red with a black spinner…

It is obvious that the tools at our disposal today, whether correctly or otherwise considered artificial intelligence, can be of tremendous help and will increasingly be so as time goes on. However, we should not turn off our own judgment and ability to draw conclusions. To end in style, I asked ChatGPT to say hello to the readers of our newsletter:

Dear readers of Info Eduard magazine,

I hope this message finds you well and that you continue to find the content of this publication informative and engaging. Thank you for your continued support and readership, and I wish you all the best in your personal and academic pursuits.

Best regards, ChatGPT

Jan Zdiarský