Stuck in time: Why AI can’t stop drawing clocks at 10:10

Let’s be honest, AI is amazingly cool, even amazingly predictable.

By now, you’ve probably seen some headline-stealing examples of generative AI conjuring up surreal art, dazzling visuals, or impossibly creative designs. Ask him to imagine exotic cities bathed in neon light or forests where trees grow bioluminescent flowers, and – what a blast! -You are presented with images that go beyond what humans usually imagine.

But then, you ask the AI ​​to draw a clock. And all the magical screaming stopped. What do you get? A watch stuck stubbornly 10:10.

It’s almost funny: no matter how the AI ​​prompts “Draw an old clock!” “Future hour!” Or even “Dali’s Melting Clock!” -Those hour hands somehow find their way into this strangely pleasing position at 10:10. If AI is supposed to understand nuance, randomness, and creativity, why should that be the case? related On this?

The answer isn’t just an amusing artifact of training models, but a microcosmic look at the larger challenges AI faces when it comes to understanding creativity, bias, and breaking free from outdated conventions. So strap on your wristband, and let’s delve deeper into this fascinating philosophical and deeply technical mystery.

The 10:10 phenomenon: a human legacy

Before we start wagging our fingers at AI, let’s talk about it we. The reason the AI ​​leans toward 10:10 doesn’t come from the algorithm deciding, “Yes, this is where the time looks perfect.” No, it’s just a regurgitation of a behavior that we humans have built into the design of watches Contracts.

Almost every ad you’ve ever seen uses the same distinct 10:10 timestamp. No, it’s not because every product photographer in the world has en masse joined the “10:10 cult.” This is why choosing this time is so dominant:

1. The symmetry looks good: At 10:10, the clock hands create a nice sense of visual harmony. It is symmetrical, but not overly rigid. It also perfectly surrounds the brand logo, which is often visible at the 12 o’clock position on most watches.

2. The “smiling clock” effect.: Look closely: At 10:10, the upward-curved hands mimic the shape of a smile. Whether consciously or subconsciously, brands realize that happy, welcoming design cues sell more products.

3. Marketing overload: Once this convention became dominant, it increased. From advertisements to stock photos to catalog images, everywhere A clock appeared, and 10:10 was the norm. It has become a self-sustaining design rule.

For decades, we’ve constantly fed the world these visual elements, making them so ubiquitous that our brains default to them when imagining a clock face. We don’t even think about it, we just expect it.

And now, artificial intelligence does too.

The AI ​​mirror problem

To understand why AI, sometimes called the “Great Imitator,” can’t break free from 10:10, let’s quickly unpack how these models learn.

Every generative AI model — including powerhouses like Stable Diffusion, DALL-E 2, and MidJourney — relies on massive datasets for its training. These datasets are massive collections of images (often billions) extracted from the Internet: stock photographs, online repositories, user-generated content, you name it.

When the AI ​​learns the concept of a “watch” from these images, it is not just analyzing the aesthetics or function of the watch. She’s looking for Repetition patterns.

Guess what dominates the images of watches on the internet? Yes, 10:10.

For the cashless AI “brain,” the most statistically significant fact about clocks is no They tell time. It almost always looks like this:

• Symmetrical hands indicate 10 and 2.
• Logo sits proudly at the 12 o’clock mark.
• Sometimes, additional complications such as chronograph dials are present as window decorations.

If 95% of the “watch” images an algorithm sees are essentially identical, guess what happens when you ask it to create a watch? Artificial intelligence doesn’t know any better. Presumably you want any familiar version of the clock that is 10:10.

But wait, AI is not that only Follow the data… right?

You’re probably thinking: “Wait, AI is supposed to be creative! Why doesn’t it rebel?”

This is where things get difficult. AI may seem creative, as if it is pulling ideas out of thin air, but it is not. Instead, it works probabilistically, drawing from patterns it learned during training. Let me demystify that.

Think of the AI ​​mind as a giant game of “autocomplete.” Imagine typing “dog breeds” into Google – autocomplete suggestions like “Labrador” or “German Shepherd” appear because they are the most common. Likewise, when the AI ​​creates an image of a “wristwatch,” it tests what it thinks the average wristwatch looks like based on its existing patterns. I already saw.

Here are the main technical details:

Generative models create images by exploring their “latent space,” a high-dimensional mathematical representation of everything they have learned. Imagine this latent space as a dense galaxy made up of patterns, ideas, and shapes. Objects like “clock faces” form clusters in this galaxy, and in the case of clocks… the densest and most accessible part of that cluster is – you guessed it – 10:10.

When the model starts generating an image, these dense areas act like gravity wells. You’re more likely to choose something close rather than wandering around in “creative randomness.”

Situation Collapse: The AI ​​cannot escape

There’s also something else at play here: Situation collapse.

Mode collapse is a common pitfall in machine learning where the AI ​​model begins to favor only a narrow subset of possibilities, ignoring less popular options. It’s like highlighting only the most common examples while the rest fades into darkness. Because the clocks are at 10:10 Hugely overrated In AI training datasets, they become the “default”. Every time you prompt the AI, it resorts to this safe and familiar choice.

Here’s the thing: it’s not just about the watches. The same bias creeps into all kinds of generative outputs. Ask an AI to generate, for example, a generic image of a “businessman,” and you will often get a stereotypical image of a Western male wearing a suit and tie — because that is what dominates stock images. AI is only as unbiased as its data, and data sets, as we know, are laden with decades, even centuries, of human bias.

Wait… can’t we just fix it?

In theory, yes. Technically? It is a much harder nut to crack.

For AI to break out of the 10:10 rut — or any other inherent cultural bias — it needs data and Algorithms that effectively resist the safety net of mediocrity. Here’s what it might look like:

1. Diversify data sets: First, check the feature of the training datasets Underrepresented alternatives. If the AI ​​training data shows clocks at random times of 10:10, we can mitigate this bias. But extending this to large datasets is no small feat – and cleaning the datasets requires significant computational and human resources.

2. Possibilities of re-weighting: Engineers can tweak AI reward algorithms to effectively promote extraordinary deliverables. For example, they can add penalties for gravitating too aggressively toward default outputs such as 10:10.

3. Injecting noise into claims: Advanced systems can introduce “fast noise,” explicitly forcing the AI ​​to randomly select fine aspects of its output, such as the position of hands on a clock — or, more broadly, to explore unexplored regions of latent space.

4. Custom fine tuning: Models can also be fine-tuned to push creations further. By training smaller, specialized models on more diverse or specialized data (such as a dataset of hours at 7:13 or 4:47), creators can target specific outputs toward breaking the mold.

However, there is a slippery slope here. Encouraging too much randomness means that AI can lose its grounding entirely, creating outputs that seem disjointed or illogical rather than “creative.” Finding the sweet spot between virtual patterns and real innovation remains one of the biggest dilemmas in AI development today.

So, what’s the big takeaway?

The reason AI keeps drawing clocks at 10:10 isn’t just about training data or programming quirks, it’s a microcosm of how generative AI reflects the limits of our creativity, biases, and data. When we expect AI to “think outside the box,” we forget that it was built inside our box to begin with.

What strikes me about this isn’t the technical monotony of how latent spaces or training distributions work (although I’ll admit that’s cool in itself). What’s striking here is how AI forces us to deal with it Our own styles. We have made 10:10 the universal symbol for hours. Until we change our traditions — or teach AI to value diversity more than familiarity — it will continue to repeat those choices back to us.

So, next time you ask AI to create a watch stuck in the past, consider it a gentle reminder: creativity isn’t always about algorithms. It’s about intention.

And right now, the AI ​​watch face is still smiling at you, forever frozen at 10:10.