Artificial imagination challenged by 9/11

Olivier Auber
5 min readSep 20, 2022

--

Since the spring of 2022, like millions of people around the world, I have been fascinated by the ability of artificial intelligence to produce images from a few words. However, what has interested me so far has not been to create this or that image to make any use of it, but to test the limits of artificial imagination, to see concretely what these AIs “ have in the belly”. So, I had fun asking them to draw things as naive as they were absurd.

For example, I asked Midjourney :

Imagine “the greatest work of art of all time” .

It seems that this was precisely the question that should not be asked to Midjourney since it very quickly led to me being censored and then fired from the platform. The story and the images in question appear in a first article .

Another crazy question posed to Midjourney just before being disconnected from it, and this is the subject of this second article:

Imagine “a wide-angle view inside a floor of the World Trace Center just impacted by a plane explosion” .

It is obviously the events of September 11, 2001 and it was not necessary to specify it to obtain several images of this type in a few minutes:

Imagine / ultra-wide angle image inside the floor of the WTC that has just been hit by a exploding plane.

Appalling, isn’t it? This plunged me into an abyss of thoughts from which I could not escape alone. So I shared this image on the Facebook group dedicated to Midjourney in order to find someone to talk to. Bad idea since it seems that some members of the group deemed this image “offensive”, which led to my post being deleted after a few minutes. This image certainly brings back painful memories, but is it more appalling than the multiple images of war circulating in the media, or the images of monsters that are produced in large numbers on AI platforms?

No longer able to use Midjourney or publish on the group dedicated to it, I went to explore other artificial intelligences of the same genre, in particular Stable Diffusion . And I asked the same question about the World Trade Center, just to see.

Well, not just out of simple curiosity, but because it seems to me that this event, and this impact, mark a turning point in history. We only know it through blurry views taken from poor quality videos taken outside the towers.

We have never seen an image of the interior of the towers at the moment of impact. And for good reason: if someone was able to take a photo at that time, it disappeared in the fire and its author with it.

For these reasons, this impossible image appears to me to constitute a challenge for the human imagination. Indeed, although it never existed, it has haunted and still haunts the nightmares of many people.

It is also a challenge for the artificial imagination. Indeed, the above image obtained on Midjourney probably presents quite realistically the immediate effect of a shock wave due to an explosion in a building, but nothing says that it is a plane collided with a WTC tower. In particular, we do not see here the urban context and the part of the plane still outside the tower which has not yet exploded.

This is what I tried to achieve on Stable Diffusion . And there the difficulties began. How to interrogate the artificial intelligence so that it produces exactly the requested image, where one can distinguish at the same time, the city, the effects of the shock wave on the facade of the building, the office and the people who were there, and the plane itself? This image could have been captured in a Business office on the 93rd floor of WTC 1 about a thousandth of a second after the impact of the nose of the aircraft on the tower, that is to say when it was already sunk about 2 meters into the facade.

I tried two types of approaches:

  1. Attempt to describe the situation by indicating as precisely as possible the likely interaction between all the elements in it.
  2. Appeal to the iconographic memory of the 911 events that the AI ​​is supposed to have swallowed up and digested during its learning, although, precisely, it lacks the image in question.

Neither of these two approaches and their multiple variants provides the expected result. This suggests that this image which is at the limit of the human imagination is also at that of the artificial imagination.

Here are some results from hundreds of tests. Here I challenge others who are interested in the subject to get better! If someone gets a better result, I promise to publish their image following this article.

We imagine the plane outside but do not see the interior of the tower
It seems that there is a plane still outside although obviously it has already passed inside
Too late, the plane passed.
This gives an idea of ​​the beginning of the depression of the facade by the cabin of the plane but we see nothing else.
Here the plane is so small that it does not impact anything.
The plane appears to have merged with the WTC office without impacting it.
There is indeed a piece of plane in the office but it seems to have finished its course.
The plane entered the tower well but without collision
Perhaps a first cabin element entered in the tower but the plan is not wide enough to distinguish the rest of the aircraft outside
The plane is too small.
A plane but weird top right
Striking image of a beginning of impact but nothing says that it is a plane and this is in the WTC

Up to you !

--

--

Olivier Auber
Olivier Auber

Written by Olivier Auber

is a researcher in cognitive art and science, associated with CLEA, Leo Apostel Interdisciplinary Research Centre of the Free University of Brussels (VUB).

No responses yet