As the use of AI continues to grow exponentially, it is becoming increasingly challenging to distinguish what is AI generated and what is created by humans, especially when it comes to imagery.
This creates a plethora of reputational challenges, especially for us in the public relations and news journalism space.
Misinformation, defamation, deepfakes, copyright issues – you get the gist. Viral tweets about an explosion at the Pentagon, Donald Trump’s arrest and the Pope wearing a puffy jacket are some examples of easily believable information. Artists and photographers are the other largely affected profession.
Leaders like Open AI’s Sam Altman and Google’s Sundar Pichai have repeatedly called for the regulation of AI. However, regulating it is so complex, I’m guessing any sort of legislation is going to take time to come into place.
The obvious flaws of AI generated images
If you have tried using AI tools to generate imagery, you will have a decent idea about the flaws of AI generated images.
Faces and objects are distorted or they clearly look like drawings, fingers are weird, the person has too many teeth, you see unnatural patterns, and there is a general creepiness in the imagery – your spidey sense will feel that something is off.
However, as the tools get better, especially the latest version of Midjourney (v5), the AI generated pictures that are coming out are pretty amazing.
What’s being done to help us know the difference between AI generated images and real ones?
This is a pretty neat feature where you can upload a picture or the URL of the picture and it will show you where else the image has appeared and in what context, so you can better judge its authenticity.
2. Google’s release of ‘About this image’ (US only at the moment):
I’m not sure if this is an extension of the above point, but this feature will give us more information on the image:
- When the image and similar images were first indexed by Google
- Where it may have first appeared
- Where else it’s been seen online (like on news, social, or fact checking sites)
3. Watermarking visual content: SynthID and FACET
Google has just launched SynthID, a tool that embeds a digital watermark directly into the pixels of an image, making it imperceptible to the human eye, but detectable for identification by a tool that is trained to do so.
Traditional watermarks are not sufficient for identifying AI-generated images because they are often applied like a stamp on an image and can be easily edited out.
SynthID uses two deep learning models — for watermarking and identifying — that have been trained together on a diverse set of images.
According to the launch information, SynthID is being released only to a limited number of Vertex AI customers using Imagen, one of Google’s latest text-to-image models that uses input text to create photorealistic images.
It seems like other image generators like Midjourney and Stable Diffusion will start this feature at some point too.
FACET by Meta is another tool. From their blog:
“Made up of 32,000 images containing 50,000 people labeled by human annotators, FACET — a tortured acronym for “FAirness in Computer Vision EvaluaTion” — accounts for classes related to occupations and activities like “basketball player,” “disc jockey” and “doctor” in addition to demographic and physical attributes, allowing for what Meta describes as “deep” evaluations of biases against those classes.
By releasing FACET, our goal is to enable researchers and practitioners to perform similar benchmarking to better understand the disparities present in their own models and monitor the impact of mitigations put in place to address fairness concerns.”
I’m not quite sure I understand FACET exactly.
4. Cryptographic signatures:
The Coalition for Content Provenance and Authenticity, also known as the C2PA, is an open technical standard providing publishers, creators, and consumers the ability to trace the origin of different types of media.
This group is lobbying for verification tools, such as digital certificates, controlled capture technology, and cryptography. These tools can help establish provenance.
5. Innovative tech startups:
There are many companies working on a tech solution to this problem. For example, Truepic offers technology that claims to authenticate media at the point of creation through its Truepic Lens; and Truepic sign aims to increase transparency at scale about how, where, when, and by whom content is created.
Deepfake detection and prevention companies like Reality Defender and Hive Moderation (detection by moderation for social platforms) allow users to upload existing images and then they receive an instant breakdown with a likelihood percentage for whether the image is real or AI-generated.
The optimist in me would like to believe that AI detection tools will catch up in AI arms race and protect us all from the potential disasters that manipulated imagery could cause.