How to distinguish photos created by Artificial Intelligence
Last month, an image of Pope Francis wearing a white Balenciaga sweatshirt went viral on the Internet. She managed to convince thousands of people that the Pope was getting ready to spend a fun evening on the town. A few days later, another series of photos circulated on social media, purporting to show the street arrest of former President Donald Trump by New York police officers.
These fake photos were created by Artificial Intelligence (AI) systems such as "Midjourney", "DALL-E" and "Stable Diffusion". They have been in operation for several years, but only recently have the images they produce become convincing enough to fool less astute observers.
And some experts think we're speeding towards a future where it's virtually impossible to tell the difference between real images and AI-generated fakes. "The systems are already very good. Even without using specialized tools for
analyzed the images, even experts can have no trouble telling the difference. AND
these systems are only going to get better," says James O'Brien, professor of computer science at the University of California. But how can you spot fake images created by Artificial Intelligence?
1. Pay attention to the fingers and teeth of the protagonists in such a photo
Since the datasets that train AI systems tend to capture only parts of hands, this technology often fails to create real hands. This can lead to images with deformed hands, a typical telltale sign that an AI-generated image is fake.
Thus in the viral image of Pope Francis, his right hand (and the coffee cup he is holding) appears crushed and deformed. But the teeth can also present problems.
"AI models have problems with that kind of structure, although newer ones are getting better," O'Bryan points out.
2. Beware of overly smooth textures
Disa gjeneratorë imazhi të AI, prodhojnë tekstura që janë tepër të lëmuara apo lëkurë me pamje plastike dhe me një shkëlqim shumë të fortë. Kjo do të thotë që një xhaketë, mund të duket shumë e bukur nga sa është në realitet, thotë eksperti.
3. Kërkoni për detaje që nuk përshtaten
Ndoshta gjërat kryesore që duhen analizuar janë mospërputhjet në logjikën e vetë imazhit. “Mënyra më e lehtë për të kuptuar se një imazh është i rremë, është të shohësh me kujdes objektin kryesor. Në çdo foto, do të vini re se modeli rreth syrit është i ndryshëm”- thotë O’Brajan, duke iu referuar një fotoje tjetër të bërë virale. Mospërputhje të tjera, përfshijnë detaje si pëlhura e veshjeve që përzihen së bashku në tema të ndryshme apo modele sfondi që përsëriten në mënyrë të përsosur.
4. Bëni kërkimin tuaj në faqe të tjera
When in doubt, don't be afraid to cross-reference what you're seeing with other reliable sources. “The things we see now, we have to immediately Google them to find more information about them. "It is true that there are small data that you can see in the image, but just reflect and do your research" - declares the expert.
O'Brien doesn't think the day is far off when we can no longer tell the difference between real photos and AI-generated images. He suspects we may cross that threshold within the next few years. "Therefore, as a society, we must become more capable of accepting the fact that the things we see on the web may not always reflect reality," he concludes.
Originally published on bota.al