Have We Already Crossed the Uncanny Valley of AI-Generated Content? (1/3)

This is a three-part series discussing how AI-generated content is plunging us into a psychological “uncanny valley.” We trace the phenomenon from its robotics roots to viral AI media and examine the future prospects of increased regulation, improving digital literacy, and developing conscious AI-human collaboration.

Part 1: Understanding the Uncanny

The uncanny valley is a psychological phenomenon first described by roboticist Masahiro Mori in 1970. Mori observed that as a robot’s appearance becomes almost, but not quite, human-like, people’s affinity for it suddenly plunges into a feeling of eeriness or revulsion. In other words, we find very-realistic robots or animations creepy when they’re nearly human but still “off.” Mori’s classic graph shows a peak of comfort that drops sharply in a 'valley' just before lifelike perfection, then rises again once the replica is truly indistinguishable from a real person.

The same concept has now moved beyond robots into the world of AI-generated media. Today, we often feel that exact chill when viewing an AI-generated image, listening to a synthetic voice, or even reading text that is almost, but not quite, natural. As National Geographic explains, uncanny valleys aren’t just about faces, they show up in voices and other domains too.

 

For example, generative models can create human-sounding voices that seem lifelike on first listen, yet something about the timing or inflection tips us off that “this isn’t quite right.” In experiments at UC Berkeley, volunteers could only tell real human voices apart from AI-cloned ones about 65% of the time, which is barely better than chance.

 

In the words of researcher Hany Farid, “AI-generated voices are passing through the uncanny valley.” In short, as AI advances it is filling in the valley’s depths: synthetic voices are becoming harder to distinguish from the real thing, and visual or textual outputs that were once obvious fakes now often look strikingly real on first glance.

How do we know uncanny AI content when we see (or hear) it?

 

For many people it’s an instinct. Take the case of Justice Potter Stewart’s quip from 1964 “I know it when I see it” about defining obscenity feels just as fitting for identifying “creepy AI” today.

 

 

We might not articulate why an AI image feels wrong, but there’s a gut reaction: something is off in the eyes or a stretch of text that makes us recoil. As one recent review puts it, “we can detect if something’s off even if the stimulus is more conceptual, like a conversation with AI.” Just like spotting em dashes on online and social content, it screams like it gone through ChatGPT and then copy-pasted to post it online. In other words, that visceral “I-know-it” moment is alive and well.

Some recent examples went viral precisely because they triggered this uncanny reaction. For instance, Coca-Cola’s 2022 holiday ad generated entirely by AI drew widespread criticism for looking deeply uncanny. Viewers complained that the family scenes looked real, but “something about them is off,” sending the ad straight into the valley. Media outlets reported social-media users saying Coke’s synthetic scenes “go too far into the uncanny valley,” undermining the warm Christmas magic the brand usually aims for.

Just a few days ago, an AI-generated ad has already hit prime time in the 2025 NBA Finals coverage with a 30-second spot showing a shirtless elderly gentleman draped in an American flag, a farmer floating in an inflatable pool filled with eggs, an alien chugging a pitcher of beer, and a lady in a sparkly pink tracksuit driving a Zamboni. That would surely trigger alarm bells on some people hating AI.

 

Similarly, weird internet images and videos powered by generative models often circulate for their strangeness: faces with too perfect skin but wrong eyes, landscapes that slide between realism and fantasy, or AI-speaker announcements with a flat tone in the middle of an otherwise normal sentence. These uncanny clips prompt a mix of fascination and discomfort and we’re drawn in and puzzled.

All this underlines that the uncanny valley concept now applies to AI content broadly. We once talked about the valley in terms of “robots and zombies,” but now it includes digital media. Each shaky AI-generated face, each slightly alien smile or repetitive phrase in text, reminds us that we know it when we see (or hear) it. As Farid’s research suggests, this intuition is backed up by data: when people nearly mistake fake for real, it signals that the technology is just peeking out of the valley.

In conclusion, the uncanny valley may no longer be a mysterious spring in the Tokyo Institute of Technology; it’s a lived experience for anyone scrolling social feeds full of AI media.

Looking to update your website or company blog? Sign up for an affordable content subscription service with JP Creative Solutions now!