AI image, left, real award-winning image, right.
The Royal Meteorological Society, which runs the Weather condition Photographer of the Year competition, has postured an appealing question: Can artificial intelligence (AI) win a photography competitors?
To address this, the Society prepared a Turing test in which the audience is welcomed to think which is an AI image and which is a real award-winning picture.
The Turing test, created by Alan Turing in 1950, is a test of a device’s ability to display smart habits indistinguishable from a human.
5 of the listed below images are real images that were shortlisted finalists in the 2021 Weather condition Professional Photographer of the Year competitors. The other 11 are generated from either DALL-E, Midjourney, or Steady Diffusion. The answers are at the bottom of the post.
Recreating Images With AI
The Royal Meteorological Society attempted to recreate winning photos from its weather condition photography competitors with AI. It took 3 winners from the 2021 competition and attempted to artificially copy it utilizing the 3 leading AI image generators.
The above images reveal the AI attempting to recreate 2021’s total winner, an image of a foggy autumn early morning in northern Italy taken from a hill church.
“The timely mentions only that a church is included rather than explicitly specifying it as the viewpoint, and that ambiguity leads Midjourney and Steady Diffusion to produce pictures of an Italian-style rural church surrounded by early morning mist. DALLE-2 produces an image much more detailed to the winner, catching a foggy valley (albeit including a church tower) lit by a low elevation morning sun,” composes the Society.
Prompts refer to text composed by human operators that notify the AI what image to create.
“These results highlight the importance of utilizing the best timely and right style with the ideal AI artist, and reveal that this type of direct comparison is probably not a fair competitors,” adds the Society.
To get the AI to produce competition-worthy images that might deceive people into thinking they are genuine pictures, the Society modified the triggers to include phrases like “acclaimed,” “4K,” and describing real places and events.
Separated into four styles, audiences have the ability to get a concept of how the various AI image generators carry out. The Society describes DALL-E as “outstanding at recording physics and structural features,” Midjourney as “the most creative with the prompts and has perhaps the best lighting,” and Stable Diffusion as excelling at “structure.”
The Society concludes: “There is still an affordable degree of human creativity needed– in thinking up the text prompts, in refining them to make the produced images preferable, and in picking the best images from those produced to take forward.
“Lots of produced images still consist of artifacts that immediately provide away as fakes– and they are still far much better at developing artistic stylizations than photorealistic images. However, these spaces are decreasing with every new hardware and software development.”
Turing Test Outcomes
When it comes to the Turing test, the images are little and low-resolution making them hard to determine, however the 5 real images are (b), (d), (i), (l), and (n). An anagram of the world blind.
Image credits: All images courtesy of the Royal Meteorological Society.