75º

Image generation: What you need to know about AI images and their risks

UTSA professor explains what to look for when deciphering AI-generated images

SAN ANTONIO – Trying to create a logo in seconds? Programs like OpenAI’s Dall-E 3 and Stable Diffusion can do it for you at a very low cost with just a simple prompt.

“The prompt is just a piece of text... So it could be something like ‘Generate an image, you know, with me near the White House,’” said Dr. Nishant Vishwamitra, an assistant professor of information systems and cybersecurity at UTSA.

The programs that generate images are trained on vast amounts of image and text data sets to create straight from your imagination.

“It’s cheap, and it actually can make very attractive-looking thumbnails. So there are many applications that people are sort of connecting,” Vishwamitra said.

Risks when using AI generation tools

The professor said there’s also some concern about risks with having this tech tool being so readily available.

“AI-generated pornography and AI-generated, you know, child sexual abuse materials and AI-generated phishing emails,” Vishwamitra said, referring to what some bad actors may try doing with AI tools.

With the November election right around the corner, there’s a concern people may start seeing more AI-generated images painting candidates in a bad light.

So, what can you do to determine if something is artificially enhanced?

“There are two kinds of efforts here then. So the traditional, you know, efforts where we do fact-finding. Right? And the second effort is automatic means, where we use some algorithms or some detectors,” Vishwamitra said.

He recommended using several detectors to be more confident that an image was made using AI or not. But when in doubt, look at the finer details, like extra fingers or toes in an image.

Government regulation

AI models can be tuned to align with certain values, and industry regulation talks remain in the early stages.

More than a year ago, President Joe Biden said he would work with tech executives on federal regulation. So far, the Massachusetts Institute of Technology (MIT) said the United States is still relying on the tech sector’s goodwill to protect people from harm. These companies can still decide to make any changes without real consequences.


About the Author

Ivan Herrera, MSB, has worked as a journalist in San Antonio since 2016. His work for KSAT 12 and KSAT.com includes covering consumer and money content, news of the day and trending stories.

Recommended Videos