Earlier as we speak, the FBI shared two blurry photos on X of an individual of curiosity within the capturing of right-wing activist Charlie Kirk. Quite a few customers replied with AI-upscaled, “enhanced” variations of the photographs virtually instantly, turning the pixelated surveillance pictures into sharp, high-resolution pictures. However AI instruments aren’t uncovering secret particulars in a fuzzy image, they’re inferring what may be there — they usually have a monitor report of displaying issues that don’t exist.
Many AI-generated picture variations had been posted underneath the unique pictures, some apparently created with X’s personal Grok bot, others with instruments like ChatGPT. They range in plausibility, although some are clearly off, like an “AI-based textual rendering” displaying a clearly totally different shirt and Gigachad-level chin. The pictures are ostensibly supposed to assist folks discover the individual of curiosity, though they’re additionally eye-grabbing methods to get likes and reposts.
But it surely’s unlikely any of them are extra useful than the FBI’s images. In previous incidents, AI upscaling has finished issues like “depixelating” a low-resolution image of President Barack Obama into a white man and including a nonexistent lump to President Donald Trump’s head. It extrapolates from an current picture to fill in gaps, and whereas that may be helpful underneath sure circumstances, you undoubtedly shouldn’t deal with it as arduous proof in a manhunt.
Right here is the unique publish from the FBI, for reference:
And under are some examples of tried “enhancements.”