What did the “Miracle of the Sun” look like to eyewitnesses? Generating A.I. images of their accounts produces fascinating results, but we need to be cautious.

With the Feast of Our Lady of Fatima upon us, my mind turned to her apparitions at Fatima and particularly to the last one on October 13, 1917. It is only natural to think about what it would have been like to be among the tens of thousands gathered that day who saw the “Miracle of the Sun.”

Would it be possible, I wondered, to get some idea of what they saw using generative A.I.?

“Revelatory possibilities”

Not surprisingly, others have already pointed out this possibility. In their fascinating paper “Choreographing Shadows,” the scholars Mark Burchick and Diana Pasulka express enthusiasm for the “revelatory possibilities” that A.I. imaging tools present, particularly in regard to religious phenomena. It now becomes possible to create “images for which we have credible eyewitness testimony, but which we do not have photographic evidence for.”

Burchick and Pasulka actually use the Miracle of the Sun as one example. Using a 1917 photograph of some of the crowd as a starting point, they uploaded the image into the Midjouney A.I. program to produce “an additional imaginative layer capable of expressing the psychology of the day.” You can view the results in their paper, which I highly recommend reading.

One does not have to be an academic, of course, to engage in similar experiments. For instance, what if we used ChatGPT, a tool available to almost anyone, to generate images not of the crowd and their psychological experience, but of the spinning sun itself, a phenomenon described by numerous eyewitnesses?

A simple experiment

It is an exercise that anyone can perform. One simply types a prompt into ChatGPT or another A.I. program, plus the description of one of the miracle’s witnesses. In less than a minute, an image is produced. In my case, I first entered a prompt that would ensure that the generated image was only of the sun. I then entered one of the testimonies, removing any references to crowd reactions, emotions, etc. since I did not want the image to contain those.

You can see the results in the PHOTO GALLERY below. I further describe my process there. During the exercise, I remembered that Pope Pius XII also saw the “miracle of the sun” four times in 1950 and left us a description, so I had an image generated based upon that testimony, too. The results of my amateur “experiment” were fascinating — or at least I found them to be.

Reasons for caution

The first thing that I realized was that we need to exercise caution when using A.I. imaging tools to recreate real life events. There is a strong temptation to believe that “this must be what the event really looked like,” when nothing could be further from the truth.

As Burchick and Pasulka point out in their paper, generative A.I. only works because it draws from billions of preexisting images and texts. In other words, instead of generating original content, A.I. regurgitates what already exists on the web. This raises all sorts of ethical questions, of course, but strictly from the point of view of “realness,” it should be evident that artificial intelligence can only produce results that are, in a word, artificial.

You have only to look at the last slide in the gallery to understand the problem. There you will see a photograph of Lúcia, Francisco, and Jacinta, the three shepherd children to whom Our Lady appeared. Next to it you will see an image of the same children generated by A.I. The difference between the two images is quite stark — and revealing.

Potential benefits

Does this mean that A.I. is utterly useless when it comes to visualizing the miraculous? I don’t think so. In fact, the exercise convinced me (though others may disagree) that A.I. can be a beneficial if used carefully and wisely.

When I initially read the descriptions of the eyewitnesses, I found them intriguing, but to be honest, they left me unmoved. That is partly due to the fact that all the descriptions are very short. They were also extremely difficult to visualize in my head, mostly because the event that they describe lies completely outside my personal experience. I have never seen the sun swirl wildly, shoot off beams of rainbow light, or rush towards the Earth. It’s hard for me to imagine what that might even be like.

That immediately changed when I was confronted by the A.I. generated images, however. They surprised me. Seeing them made me understand that the witnesses were struggling to describe something absolutely impossible that they had nevertheless seen with their own eyes. Even if the A.I. images are not accurate depictions of the miracle (and there is no way to know whether they are or not), they nonetheless manage to convey something of the strangeness and grandeur of the event.

In the end, I still don’t know what the Miracle of the Sun really looked like. However, I now have a least a little idea of what an unforgettable, terrifying, and life-changing experience it must have been. My sense of indifference has been shattered.

As the A.I. revolution continues, there will no doubt be growing numbers of people typing or dictating Bible verses and saint’s stories into A.I. programs. They will do so hoping for a glimpse of what those events were “really like,” just as I wondered about the Miracle of the Sun. As for me, I am finished with such experiments, and will happily leave them to professionals like Burchick, Pasulka, and their colleagues.