Humanlike – not human. Why UneeQ digital humans aren't photorealistic

Three key considerations have led us to use CGI over deep fake technologies to create UneeQ's digital humans. Let's discuss them.

March 25, 2024
Danny Tomsett
Humanlike – not human. Why UneeQ digital humans aren't photorealistic
Loading the Elevenlabs Text to Speech AudioNative Player...

If you’ve even a passive knowledge of sci-fi, you’ll have observed a curious trope in how these stories are visually told. Have you ever noticed how many robots, droids, or other AIs look somewhat human?

In film alone, there’s C-3P0, those oddly smooth androids in I, Robot; even the eponymous automaton in the 1897 French short Gugusse et l'automate was thought to have a very human look. 

For 120+ years on the silver screen, we’ve envisioned a type of machine that looks somewhat human. Why? Well, for one, it’s easier to find some sort of emotional attachment to something that has human qualities. 

In cinema, those emotions are often fear, mistrust, and dread – emotions that exist for narrative purposes. Because no film writer is ever having their film commissioned where the AI simply helps the protagonist find insurance.

But the same principle applies in reverse – and in real life. Just look at Martin the Geico gecko. He has humanlike qualities so we can engage with him and therefore find him charming, fun, and full of life – even as he’s trying to grab 15 minutes of your time so you can save 15% on car insurance.

And from digital geckos, we arrive at digital humans. In a world of synthetic media, GAN image creation, and AI-generated avatars, we thought it was time to tell you why UneeQ doesn’t use these technologies to create digital humans that are virtually indistinguishable from real people.

It’s a topic I see myself talking about more often recently, not least of all this year at SXSW. Three elements have led us to use CGI over deep fake technologies to create our digital humans.

1. It’s a matter of trust

Photorealistic digital avatars are often made using deep-fake technology. Think of those AI-generated videos, or avatars with superimposed moving mouths. They no doubt have their place in the world, particularly by solving the problem of video creation at scale. However, we intentionally steer away from this design style for our real-time AI avatars because we believe it doesn’t promote healthy, trusting relationships with brands.

A study in 2023 found that between 27–50% of people can’t tell the difference between an AI-generated deep-fake video and a real one. With every year the technology progresses, we’d wager that number will grow significantly.

Even if you’re confident in your ability to spot a deep fake, you might be wrong. Research shows that people generally cannot differentiate between deep fakes and real people, even when they are financially incentivized to do so. And they largely overestimate their ability to pick a fake out from the crowd.

According to Stanford University: “Not only has this technology created confusion, skepticism, and the spread of misinformation, deep fakes also pose a threat to privacy and security.”

To be clear, it’s not brands that employ deep-fake AI avatars who cause these threats. But the more deep fake technology is used by those with criminal intent, manipulating people’s trust and making them doubt whether what they see is real, the less successful photorealistic digital humans will be at building trust with customers.

It seems redundant to point out how important trust is, but consider this. Whether the digital human is helping you to shop or providing recommendations around your skincare routine, we as customers need to trust them to take their AI recommendations.

Not to mention when they’re tasked with helping people in healthcare scenarios. Patients need to know that they can talk about their most sensitive issues; things that need to be disclosed so they can receive treatment. Much like in the case of Ellie, the breakthrough avatar from USC, who helped to identify signs of PTSD in veterans.

So for these reasons, we don’t believe digital humans that are indistinguishable from real people are the right technology for enterprise AI applications and conversations.

It’s why we rely on CGI to create our 3D avatars at UneeQ. They’re not human – you know at first glance they’re not human – so people aren't deceived into thinking they are.

2. A strong basis for fun 

When it comes to using CGI characters to create an emotional bond, gaming has shown us the way. Any gamer out there will be able to think of a character that resonated with them on an emotional level.

Ellie or Abby in The Last of Us 2; Cloud and Aerith from Final Fantasy 7; the eldest sibling in Brothers: A Tale of Two Sons.

OK, so they’re all sad emotions. But there are fun moments too! In my era, it was the cast of The Secret of Monkey Island on the Amiga 500 that made us want to play right up until bedtime. 

The point is, video games CGI has set a benchmark for creating characters that are humanlike enough in their appearance, actions, and personalities. We connect with them, we talk to them, and we root for them.

That’s transferable to digital humans – in a way, a type of AI-powered NPC (or non-playable character). There’s already a pathway that links these characters to entertainment. Our digital humans are even created in Unreal Engine, so naturally resemble many of our heroes in the gaming world.

In many ways, it also sets enjoyment as an expectation. This is one reason why a digital human’s personality matters so much. They should be able to engage their lighthearted nature when the conversation calls for it, or even play a game. Labeeb from Masdr did this perfectly at Leap 2024, analyzing and guessing what attendees had drawn via computer vision.

That’s not to say photorealistic AI avatars couldn’t play games. It’s the expectation that’s different. We see them and we have a basis of understanding what they are from the world of video games.

3. Carefully navigating the uncanny valley

Everyone’s sensitivity to the uncanny valley is slightly different. I know some people in our early days back in 2017 said our digital humans set off their uncanny alarms.

If something doesn’t look human enough, it rings those internal alarm bells.

It’s why we’ve focused on upping the realism in our digital human models, improving the fidelity of our digital humans, and creating Synanim™ for hyper-realistic humanlike animation.

Still, the opposite is true. When something looks incredibly human, even the slightest inhuman flaw is incredibly jarring. There really is no room for error.

It means expressions that are meant to be comforting can fall headfirst into the uncanny valley. To return to cinema, just look at the 2022 horror movie Smile, where possessed people, despite being human, were only identifiable by a grin that stretched a little to broadly across the face. So unsettling was the infamous smile, when people did it at baseball games in promotion of the film, it was ranked among the scariest horror movie marketing campaigns of all time.

It’s true, the simplest way AI avatars fall into the uncanny valley is mouth movement, because it’s incredibly hard to do. When a digital human’s mouth moves but it doesn’t trigger subtle muscle movements around the rest of the face, we all notice – even subconsciously. Everything from blinking rate to jerky body movements and slightly off postures can send people’s creepometer into “I need to leave” mode.

Perhaps it’s the years of video game history we have, but CGI digital humans are much more forgiving. Lapses in animation look like a quirk, not a full-blown horror movie. And when it comes to visually representing some of the biggest brands in the world, we don’t find it right to gamble on their reputation when the chance of falling into the uncanny valley of photorealism is so high.

CGI or GAN: So what’s the answer?

Digital humans are tipped to be one of the fastest-growing markets in technology. Gartner bullishly believes that around half of marketing teams will use digital humans by 2027, and that the “digital human economy” will be worth $125 billion by 2035. Others go further, predicting an annual growth rate averaging 47% (CAGR), which outstrips the growth expectations of cloud computing and electric vehicles.

So as we quickly build this new era of AI interactions, it’s important to keep discussing such powerful (albeit foundational) decisions like ‘what does the digital human look like?’. 

For us, the answer is clear. 

For our technology to have the greatest positive impact, for it to be an enabler of trust and support, and for it to inspire fun and entertainment, we need to focus on what creates the greatest user experiences. And for that, we need to create AI that doesn’t look identically human but intentionally digital human.