Most AI avatars look female, young and attractive. Are they a passing trend or here to stay?

What do German tabloid reporter Klara, Korean newsreader Zaein and Mexican entertainment news presenter NAT have in common? ‘NAT’ stands for Neural AI Technology; Klara Indernach’s initials spell out ‘KI’ for Künstliche Intelligenz, AI in German; and Zaein’s body is played by a team of actors. All three young women in journalism are not women at all, but AI-generated ‘avatars’, three-dimensional human likenesses whose features and voices have been selected and honed, and whose ‘personalities’ and careers are designed for them. 

These avatars, often newsreaders or presenters, predominantly for the entertainment beat, typically resemble attractive young women with side gigs as Instagram influencers, models or (in Zaein’s case) K-pop stars. Multiple similar experiments have been emerging from news and entertainment organisations around the world, from large Spanish broadcaster Mediaset to local Indian outlet News18 Punjab/Haryana. 

 

 

 

 

Sieh dir diesen Beitrag auf Instagram an

 

 

 

 

 

 

 

 

 

 

 

Ein Beitrag geteilt von Alba Renai (@albarenai)

As experimentation ramps up and AI creations become both more complex and more accessible, the appeal of these AI-generated characters has grown. On social media, these characters can be developed even further, given hobbies and interests, personalities and identities. One of the first AI influencers to gain a significant following was Miquela, created to portray a Brazilian-American singer and model in 2016. Zaein, the Korean newsreader and K-pop star is a member of the all-AI 11-member girl band IITERNITI, which released its debut song in 2021. At the time of this writing, the band had more than 107,000 subscribers on YouTube.

 

 

 

 

Sieh dir diesen Beitrag auf Instagram an

 

 

 

 

 

 

 

 

 

 

 

Ein Beitrag geteilt von Miquela (@lilmiquela)

Some existing multi-purpose avatars are now adding news presenting to their list of talents, while others are directly developed by news organisations keen to be at the forefront of the AI revolution. For Oswaldo Alejandro Aguilar from the Mexican radio network Radio Fórmula, the idea to create an AI avatar came as a result of his own experimentation with generative AI.

“I was playing with the early versions of ChatGPT and my partner and I discussed the possibility of creating a news presenter using artificial intelligence, so I started working on this. l did a little research and I created the first avatar,” he says. This was the first iteration of NAT, the Mexican entertainment news presenter and influencer. 

Why are there so many ‘female’ avatars?

“We decided that this had to be a female character because we believe people feel more comfortable when they speak with an algorithm or a machine when it has some female characteristics. For example, Alexa, Siri, and Cortana are female characters so we decided to follow this trend,” Aguilar says, referring to the big tech digital assistants who animate mobile phones or devices in our homes. 

While most digital assistants now have a selection of multiple voices of various genders to choose from, the default seems to remain a “gentle, well-modulated woman's voice,” agrees April Newton, Assistant Teaching Professor in Multimedia at Loyola University Maryland. 

Why are people more comfortable with feminine digital assistants? “Because we order those devices to do things for us and we are very comfortable ordering women to do stuff for us. It reinforces these stereotypes that women are meant to be subservient and responsible,” Newton argues. 

For tech companies, the choice of gendering their products is likely to be based on the results of consumer testing than from directly following societal expectations, says Jesse Shemen, co-founder and CEO of AI dubbing startup Papercup. The company offers a service to replace the original audio of a video with a translation in another language by an AI-generated voice designed to resemble the original speaker’s.

“I don't think the likes of Google or Amazon operate blindly or just try to philosophically refer to certain preconceived notions about gender,” Shemen says. “That's probably just the brutal reality of numbers. Now, is it incumbent then on these players to try and change those paradigms? I think there's probably a deeper societal question than just changing the gender of a voice assistant.” 

Shemen’s own company tries to remain as close as possible to the original voices in its dubbing work, including in matching the demographics of the speaker.

Pulse9, the Korean production company behind Zaein and IITERNITY, went through a complex development process to reach the final versions of its avatars. 

Despite the final members of the band all being female, the company initially developed both male and female faces by analysing the features of K-pop stars from the past 20 years, says Khurliman Kozibaeva, a member of Pulse9’s business development team. “This allowed us to analyse and synthesise traits and elements that have captivated fans over the years, ultimately leading to the creation of a new, non-existing face that draws on these influences,” she says. Members of the public were invited to vote on their preferred avatars. 

Male AI avatars do exist: as well as NAT, Radio Fórmula now has two additional currently operational avatars, SOFI and MAX, and the latter, a sports news presenter, resembles a man. Greece’s first AI newsreader, who debuted on the country’s public broadcaster, depicts a young man named Hermes, mimicking the Greek deity for roads, commerce and news. Nonetheless, this first wave of experimental AI characters appears to be overwhelmingly female.

Why do so many look young and attractive?

AI avatars are also typically attractive young women: often thin and with traditionally feminine features. Diversity does exist at the level of the race and ethnicity the avatars aim to represent, which usually matches that of the majority of the intended audience.

Speaking about the choices he made when he first designed NAT, Aguilar says: “I decided to create her with very Mexican looks. She had to have tan skin and dark hair.” He acknowledges that he intended to make her conventionally attractive, but says he also tried to design her looks to somewhat represent the community.

According to Mexico’s 2020 Census of Population and Housing, 19.4% of the total population self-identifies as Indigenous and 2% as Afro-descendent.

“For this first project, we thought that she had to be pretty,” says Aguilar, who has received criticism for this decision. He responds that NAT is the result of an experiment and that the experiment has been successful. “We don’t have the intention to hurt anyone's feelings,” he says. “NAT has very dark skin like most Mexicans. She’s not white. But at the end of the day, it wasn’t that we wanted to follow some kind of model. It was more of a happy accident that she looks like this.”

 

 

 

 

Sieh dir diesen Beitrag auf Instagram an

 

 

 

 

 

 

 

 

 

 

 

Ein Beitrag geteilt von Nat (@nat_ialife)

Pulse9 intentionally pursued an ‘ideal’, almost unreachable standard of beauty while designing Zaein, Kozibaeva says: “One of our core priorities was to craft a K-pop idol that reflects an idealised standard of beauty that many fans aspire to. Zaein, alongside other members of IITERNITI, features characteristics that can be described as incorporating almost unrealistically perfect aesthetics. This design decision aligns with the well-known trends in K-pop, where idols often display stunning, almost dream-like appearances that captivate and inspire their fanbases.”

 

 

 

 

Sieh dir diesen Beitrag auf Instagram an

 

 

 

 

 

 

 

 

 

 

 

Ein Beitrag geteilt von 제인 Zae-In (@zaein.iiterniti)

The problem with pretty AI avatars

Professor Newton thinks these decisions are deeply problematic. “This attitude always reinforces expectations,” she says. “It's always going both ways. They're simultaneously choosing somebody who is socially acceptable, someone who is the standard of beauty for their community, but then also setting the standard of beauty for the community.” 

This echoes criticisms directed towards generative AI more generally: because it ‘creates’ its output from an existing data set, it will tend to replicate biases and human prejudice. For (real) women on screen, including in news presenting, their appearance has been and continues to be a factor in how they experience their job, Newton has found in her research, with women more likely to experience sexual harassment, including via social media. 

How can companies offset this? “I would certainly advise them to not let AI choose what a typical attractive person is, I would advise them to look around their space and at the audiences that they serve and think really carefully about who the totality of those people are. They should think about what's appealing in real human beings to that audience and find a way to make their avatar fit that,” Newton says.

This trend of AI avatars resembling models may change in the future. Grupo Fórmula has set up a new company to develop AI avatars for other organisations. The initiative, Next AI Solutions, has already created two avatars for external customers: América, a female digital influencer for the Mexican football team Club América, and Alex, a male avatar to provide financial tips and updates related to the economy on behalf of financial services company Finamex. 

 

 

 

 

Sieh dir diesen Beitrag auf Instagram an

 

 

 

 

 

 

 

 

 

 

 

Ein Beitrag geteilt von América (@soyamericaia)

The development process to determine what the avatars look and sound like for these projects is based on both market research and the requirements of the client, Aguilar explained. “It depends on the characteristics of the project. We recommend different kinds of avatars, female, male and non-binary, but now it depends on what kind of product we are designing or what kind of audience the customer is trying to reach,” he says. 

Aguilar believes this technology will soon be more present in the media. As it becomes more common, he thinks that diversity in appearance will increase as the avatars will be created to serve more audience segments. For example, the team is currently working on an avatar to join popular host Maxine Woodside on her show Todo para la Mujer. Because the show’s audience is mainly composed of older women, the avatar will resemble a woman in her 50s. 

Is this a fad or a lasting shift? 

Aguilar believes AI avatars are here to stay. Grupo Fórmula is betting on an enduring appeal with the establishment of Next AI Solutions. “We believe it will be a very good source of income for the company,” he says. Grupo Fórmula is also using its characters, as well as some of its popular human presenters, to play into the growing appeal of news influencers.

When it comes to news, the Digital News Report 2024 found that people pay more attention to online influencers and personalities than to journalists and news media accounts on audiovisual platforms like YouTube, Instagram and TikTok.

NAT has both ‘professional’ social media accounts and a ‘personal’ Instagram profile for images of herself in the style of a glamorous influencer as do Zaein and her IITERNITI colleagues. “We are trying to make her an influencer, we hope one day she will have a lot of followers and a lot of views,” Aguilar says. The goal is to eventually secure brand collaborations like real-life influencers do. “We think it will be a good business for the company,” Aguilar argues.

Aguilar also thinks that at some point these avatars will play a more active role in presenting shows, alongside human hosts: “They will soon join real people in these shows. This is something we are working on.”

Many audience members are uncomfortable with Aguilar's work. Most of the negative feedback he receives is on Facebook, he says, while on LinkedIn most responses are positive and on Instagram there’s a mix. He attributes this to the demographic differences between platforms, with Facebook being more popular among older people.

Recent research by the Reuters Institute shows discomfort among audiences with the use of AI to create synthetic media, particularly artificial presenters and authors.

Aguilar believes this will change as people become more used to AI avatars. One of the main concerns people put to him is that humans will lose their jobs. He doesn’t believe this will be the case. While the NAT project started with two people, there is now a team of five behind the avatar, 25 across all AI avatar projects, and 43 working on AI technologies in the company more broadly, he says. 

“This creates new job opportunities for people who know how to use artificial intelligence technologies,” he says. He compared the fear of AI to the fear of steam engines in the 19th century: “At the end of the day, the technology proved to be good and people accepted and learned to use it.”

Kozibaeva agrees: “We are only scratching the surface of what is possible. Virtual humans like Zaein represent a fascinating glimpse into the future. As we make strides in refining these creations, we anticipate they will become an integral part of our tech-driven world.”

Today a human actor has to play Zaein’s body. But this is something Pulse9 is looking to phase out as their technological abilities improve. This would allow Zaein to become “a model that can produce steady, reliable content and maintain a continuous presence, attributes that are invaluable in today’s fast-paced media landscape,” Kozibaeva says.

Pulse9 says it “firmly believes that the human element remains irreplaceable.” However, the company describes Zaein as ‘the perfect employee’ for her “unparalleled longevity and consistency.” 

For Newton, there are three possible ways for this trend to evolve: it could simply prove to be too complicated, not providing the necessary results and dying out; it could lead to newsrooms leaning into AI to summarise the news at the expense of in-depth reporting; or it could live on as a gimmick. 

“You could have real newsreaders who are still doing the bulk of the work, but you've also got an AI person reading box scores from sports around the world, and that might be something that frees up a human from having to sit and read and could be something cheeky and fun for audiences,” she says. Even in this final scenario, she wonders if AI avatars wouldn’t become too distracting.

“I wonder whether people will be more interested in how good the avatar is than in what it says. They may be sitting there and thinking, ‘Do I really believe that the lips are moving?’ or ‘Wait, it's not blinking!’ and not even listening to the news.”

To find out more:

• Listen to our AI and the Future of Audio podcast episode about NAT, and read the accompanying Q&A
• Explore our recent survey report on public attitudes towards AI in news
• Read the Digital News Report 2024 chapter on AI in journalism