On Thursday, 12 September we played around with AI and images.
We took a look at this website which generates faces of people who do not exist. They look very real! Just refresh/reload the web page to generate a new face. https://thispersondoesnotexist.com
How can we be sure that images like these – or any photos for that matter, are real or fake? This website gives us the tools to find out: https://sightengine.com/detect-ai-generated-images Simply click on: “Check Image” to upload the photo you want to check. Here’s what it reported when we uploaded a fake face from thispersondoesnotexist.
Sightengine shows you the AI model which was used to create the image. In this case, it is GAN, Generative Adversarial Networks. Most AI images are created via a technique called “Diffusion” The GANs technique is less frequently used. We uploaded a real photo of an orchid and then a composite of two real photos created using Pixlr express both of which Sightengine recognised as being “not likely to be AI-generated or deep fake“. (See: https://javeacomputerclub.com/2024/03/16/click-and-tap-instant-photo-editing/ to find out how to use Pixlr Express.) The photo is a fake – the original background has been changed – but it’s not AI-generated.
The Sightengine website also includes a “take the AI or not test“, so you can find out how good you are at differentiating between images of real things and AI created images. Some are obvious – others are not.
How can anyone be sure that the photo on an ID card is not an impersonator? OnFido https://onfido.com/solutions/biometric-verification/ uses AI to detect fake ID’s. AI generates a score based on the similarity of the face and the photo ID, and runs liveness checks to identify photos of screens, 2D and 3D masks, and image upload manipulation.
Some people use AI face recognition to unlock their phones – is this really safe ? perhaps not. They can spoof your selfie! https://turnto10.com/i-team/consumer-advocate/selfie-spoofing-how-hackers-are-duping-facial-recognition-software-scammers-authenticate-private-biometric-scans-july-15-2024
Gillian mentioned that ChatGPT had been able to extract the French from a photo of hand-written text, so Chris challenged it with images of stylised hand-brushed Chinese characters on the lid of a soup tureen with this prompt: This is some Chinese script on the side of a lid of a porcelain soup tureen. What does it mean?
ChatGPT came back with the unlikely answer: “This translates to “Eating this will be poisonous” or “This is poisonous if eaten.” It appears to be a warning label, likely indicating that the item is decorative and not safe for food use. Be cautious when using such items!
When asked to have another go it said:
“Apologies for the confusion earlier! Upon a closer look, the Chinese characters are:”食得其寿” This can be translated as “Eat and you will have longevity” or “Eating (from this) brings long life.”
…. that’s a very different kettle of fish! (or tureen of soup). It sounds more feasible, though two of the characters ChatGPT wrote are different from those in the photo.
As a point of interest, Google Translate said: “Longevity means heart food” and Deepl (which only handles simplified Chinese) couldn’t do it at all. Microsoft Co-Pilot came out with “The Chinese inscription on the lid of the soup tureen reads “毒則火” (dú zé huǒ), which translates to “Poison is fire”. This phrase could be a warning or a symbolic statement, possibly indicating that the contents are very hot or potentially harmful if not handled properly.
… So we are none the wiser. I guess I’d better consult a human being who can read Chinese !
Finally, we tested ChatGPT on two more image tasks – which it completed perfectly:
Christine Betterton-Jones – Knowledge junkie