Gina blinked her eyes hard before staring at the screen again. The headline remained the same: 673 Dead in Catastrophic Double Plane Crash. But was it real?

Since acquiring her new job as part of the “Generative AI Hallucination Detection” team, she couldn’t look at anything in the news or on social media without questioning its origins. Ironically, her co-workers and friends called her the “hallucinator”, even though her job was to detect and dispel them. She had a knack for identifying Gen-AI hallucinations, the real damaging ones. However, sometimes she did feel like she was hallucinating herself. It got tough to manage the ever blurring lines between real, factual, experienced, and just plain made up.

Even though apps and programs abounded to detect hallucinations, deepfakes, and other forms of generated deceit, sometimes the work required a human. Other times it came down to hardcore research, perusing non-digital formats in archives. The thing about generative AI is that it created based on what it was “fed”. Sometimes the AI generated new, and often fictitious results, labeled as “hallucinations.” Once created, hallucinations could also become part of the generative AI “diet”, and could be regurgitated later with previous references to back it up. It could be a perpetuating cycle, hard to detect and even harder to extinguish.

Her attention focused on the headline before scrolling quickly through the article. She jotted down key information points. Airline names. Locations. Plane models. Number of passengers. After, she zoomed in on the photos looking for telltale signs of fabrication. In the 90 seconds it took her to do this, her feed had been tingling madly with frantic work messages about the dramatic headline. In the 120 seconds it took her to listen to the frantic work messages, the headline was viewed over 100,000 times and counting.

Gina cringed hearing her boss’s orders barked through the feed. Due to some recent upgrades, the sender’s real voice read the text-based messages. Another feat of AI-magic. In essence, her boss needed answers and fast. The challenge, for Gina and others on her team, is that finding the facts needed to dispel hallucinations often came from the same source. Unless she could verify details, like the ones she noted, in alternative ways, it could be tricky to disprove.

She started where she always did, by reaching out to human experts. This time, she started with her airline contacts.

Leave a Reply

Your email address will not be published. Required fields are marked *