by Brent Eades
Some readers may know that I do an occasional feature called ‘Backyard Birds,’ which is just what it sounds like — photos of birds in my backyard. Birding is a hobby that I quite enjoy.
Another of my interests is artificial intelligence (AI) tools, which are proliferating online.
Some of these can write very human-sounding text, or provide convincing voice narrations for videos, complete with your choice of accent. Others can crunch data in ways that no humans ever could. But the most interesting — and, frankly, scary — category is AI image generation. Online tools now available to anyone can produce images that are hard or impossible to distinguish from photos taken by a human with a camera.
I recently got access to the most powerful of those, called DALLE-2. I asked it to generate the kind of photos I might include in a Backyard Birds column. With some tinkering, I was able to create ‘photos’ that are pretty darn convincing. But I used only words to do that, with prompts like “realistic photo of a chickadee on a branch, sharp focus.”
They aren’t perfect; if you look very carefully some small details of the birds or the backgrounds look a little off at times. But it’s still astonishing how fast this technology is advancing.
(I’ll add that the creators of DALLE-2 have strict rules on what kinds of images you can produce, restricting anything with a potentially political slant as well as violent or sexual themes. But still, this particular genie is out of the bottle. We’ll need to be ever more cautious about accepting what we see online as real.)
Here are a few more samples:
