In recent weeks, fake images created by artificial intelligence (AI) have been spreading massively on social networks. These systems are capable of creating works that look like realistic photos, featuring personalities in unlikely or controversial situations.
Despite their potential, image-generating AIs also raise ethical challenges and issues. For example, they can be used to create false images for the purpose of disinformation or manipulation of public opinion. In addition, they raise questions about intellectual property and copyright attribution, since they can generate images that are very similar to existing works.
The best known, Midjourney, announced on March 30th that it was suspending the free trial offered to its users because of the extent of these fake documents. But how do these AIs work and how not to get fooled? Here are some tips to recognize an image generated by artificial intelligence.
How does image-generating artificial intelligence work?
Image-generating AIs such as Midjourney and Dall-E work with deep neural networks, capable of learning from thousands of existing examples. By analyzing this data, the AI is then able to produce new images by combining different elements present in the examples it has studied. The result is often stunningly realistic, making it difficult to distinguish between a real photo and an AI creation.
Limitations and flaws of images created by AI
However, the images created by these AIs are not free of defects. Certain elements can betray the presence of an artificial intelligence behind the creation of a photo:
- Inconsistent or impossible details in the image: for example, a reflection that doesn’t match the scene depicted, or objects floating for no apparent reason.
- Repetitive textures: the AI tends to reproduce certain identical textures in several places in the same image. This is particularly noticeable on complex patterns such as clothing or hair.
- Perspective or proportion errors: sometimes the AI makes mistakes in respecting the rules of perspective or in the proportions between different elements in the image.
Tips for spotting an image generated by artificial intelligence
To avoid being fooled, here are some tips for identifying an AI-generated image:
- Carefully observe the details of the image: look for inconsistencies and errors mentioned earlier.
- Use online tools to check whether an image comes from a reliable source or not: some sites offer services to verify images by comparing their content with existing databases.
- Verify the source of the image: if it is shared massively on social networks without a clearly identifiable source, there is a good chance that it is an AI creation.
- Use common sense: if an image seems too startling or provocative to be real, it was probably generated by the AI to generate reactions and be shared.
User responsibility for AI-generated images
Finally, faced with this proliferation of misleading images generated by artificial intelligence, it is essential that social network users be vigilant and critical. Before sharing an image that appeals to us, it is important to verify its veracity and to ask ourselves the question of the impact that its diffusion could have. By doing so, everyone contributes to limiting the spread of false information and to preserving a healthy and constructive exchange space on social networks.