The technology now exists that will allow AI (artificial Intelligence) to produce about any image that you desire. AI can make an image without using a camera by utilizing millions or billions of real photographs to make virtually any picture. Want a picture of a moose next to a pond? No problem, AI can do that. How about a jack rabbit sitting on a rock in the desert? Done. A snow-capped peak with a refection lake at its base? Sure, easy. No more getting hot, cold, wet, or chased by mosquitoes in trying to get that beautiful landscape. No more worries about f-stop, shutter speed, tripods, and all the other equipment we tote around. And think of the travel costs that you can save by just remaining at home working on your computer.
The basic question is, will AI make fundamental changes in how we produce our images? Is this the future? If so, how will we know which is a real photograph and which is a synthetic image? (BTW, I am reluctant to refer to a synthetic image as a photograph; others may choose to differ). A foolproof identity system is needed and is being developed by Google DeepMind/ Google Research called SynthID. A beta version is being tested at this time to see if it will be effective. This technology embeds an AI identifier in the image which is invisible to human eyes and is very difficult to remove by cropping or resizing the image. To be effective, this system will require an easy and widely available system to insert and “read” the “hidden” AI identifier. Can we trust that all synthetic images will be generated with the appropriate identifier? Would you consider an image generated by AI more or less valuable than one created by a human?
As photographers, we make alterations to every photograph by improving density, color, contrast, and sometimes removing unwanted elements by cropping or using various methods including the use of AI. B&W photographs are accepted as interpretations of reality since the modifications made by the photographer are inherently obvious. But changes to color photographs are not so obvious yet we have come to trust that they are “real”. At what point, if any, should the use of AI in image post processing be identified?
A recent example comparing real photography with similar synthetically made images can be seen in the Summer 2023 issue of Audubon Magazine here along with a discussion of the identify problem in the article by Allen Murabayashi, These Birds are Fake, in which he concludes that:
“AI’s rate of evolution is outpacing legal, ethical, and technological frameworks that might constrain its use and protect society from harm. We don’t want a system that is reliant on experts to detect hallucinations, or what’s real from fake, nor to have to fix a broken technology after it has inflicted harm.”
The synthetic images in the Audubon magazine are surprisingly good but not (yet) the quality of the real photographs but might be good enough for many people who won’t know (or care?) that they are “fake”. Think about how easily(?) and quickly such synthetic images can be produced and treated as believable photographs. There are probably many markets available for the use of synthetically generated images.
On a global scale, there are many advantages to AI and how it can solve difficult problems and make huge improvements in our society, yet there are serious inherent risks that remain. Geoffery Hinton, the “father” of AI, has stated that “AI must be built with strong ethical principles”. Eric Schimdt, ex-CEO of Google recently stated that “AI must stay within acceptable human behavior” and be given all the safeguards needed to keep it from controlling humans.
Finally, many technological and ethical questions remain including the issue of whether AI produced art can be copyrighted. I was reminded that one court has ruled that AI generated art cannot be copyrighted, however this ruling may face challenges. Your perspectives on AI?
-By John R. DeLapp (not a robot)