As US literacy declines, AI images are reshaping how people decide what’s true

The Petroglyph National Monument sits along Albuquerque’s western edge, where volcanic rock carries centuries-old carvings from Pueblo people. The images are simple, sometimes faint, but they have endured — visual storytelling as a durable record.

Today our tools for making pictures have never been more powerful. But their relationship to truth may be getting weaker. 

“Images don’t reflect reality, they train it,” said Denise Zubizarreta, director of development and communications at the New Mexico Local News Fund, speaking last month at the organization’s local news summit. I was there speaking about AI ethics, and Zubizarreta’s remarks caught my attention. 

A mixed media artist and culture writer in addition to her comms role at NMLNF, Zubizarreta has been studying how audiences process visual information as part of her PhD work in applied social justice. Her concern: the rapid improvement in AI-generated imagery and the way it hijacks our brains, aided by declining trust in institutions and declining ability to sort the truth out of the information flood.

The share of US adults scoring at the lowest literacy levels jumped from 19% in 2017 to 28% in 2023, according to federal data released in December. The National Center for Education Statistics called that 9-point increase “substantial.” Meanwhile, reading scores for 4th and 8th graders hit new lows, according to 2024 NAEP results, with about 40% of 4th graders below basic proficiency — the worst since 2002. The decline is steepest among lower-performing students. Fewer kids read for fun, which leads to fewer adults who can read for any reason. In the 1970s, we reached effectively full adult literacy. We’ve been sliding backward ever since. 

So we’re losing critical analyzing skills just as we’re under attack by a swarm of industrial-scale fakery. 

“Feeling is faster than thinking,” Zubizarreta said, noting that the brain can process images in milliseconds, many times faster than we process written language. The long-feared “deepfake” future is now — and it’s more pernicious than we first understood it. (Related: A still-useful glossary of AI terms we put together a few years back.)

In March 2023, I was scrolling social media and saw the now-infamous AI-generated image of Pope Francis in an ostentatiously stylish white “puffer” coat. Barely paying attention and not particularly interested in the vagaries of the Catholic Church, I only registered this as “people online are amused that the pope was seen in an uncharacteristic outfit,” and I scrolled on past. Only later did I understand that the hubbub was because it was an AI-generated fake. 

Silly as that example sounds, Zubizarreta argues that AI scale and social media speed mean deepfakes don’t even have to hold up to close scrutiny. In most cases, there won’t ever be close scrutiny.

“Truth is not competing with falsehood,” she said. “It’s competing with velocity.”

AI-generated images of President Obama DJ-ing or President Trump surrounded by smiling Black voters are often created by supporters to reinforce a preexisting belief. Detractors can do the same. Some look photorealistic, but even those that plainly are not can create a “belief pathway,” Zubizarreta said: “We remember 80% of the images we see, and just 20% of what we read.”

She frames the dynamic through what she calls “DRIP”: disinformation, repetition, internalization and propaganda. Images don’t need to be fully accurate to be effective. Just “real enough,” repeated often enough, and easy to process.

In a feed-driven environment, repetition comes cheaply. As philosopher Hannah Arendt famously wrote: “If everybody always lies to you, the consequence is not that you believe the lies, but rather that nobody believes anything any longer.”

That has implications not just for journalists, but for all of us, especially those operating in a civic square. Carefully edited words are still effective for many, but they might increasingly be used as a foundation for images and videos for others. 

After speaking at the New Mexico conference, I toured Petroglyph National Monument. I couldn’t entirely understand the significance of many of the faded carvings. Leaning on the black rock of an escarpment above the desert bowl in which Albuquerque sits, I leafed through descriptions that the National Park Service printed on a map. Only with both the words and the physical rock did I begin to fully appreciate the site’s importance.

“We teach people to read,” Zubizarreta said earlier that day. “But we don’t teach them to see.”