The phrase “Hand over your inbox, calendar, and credit card to me,” paired with a toy that can smile and administrative privileges, seems like a perfect combination. At least, Meta and OpenAI seem to be betting on this point in their discussions about the pessimism surrounding AI: to endow their new AI agents with a face that people might be willing to embrace.
On September 8th, Meta launched Muse, and subsequently introduced Jolly – a fluffy, customizable mascot with rosy cheeks. On September 29th, OpenAI released Dots, whose optional colored characters can wear accessories such as berets, ties, and glasses. Users of both companies can even give names to their own avatars.
The founder of the public relations company EZPR, who is also becoming increasingly outspoken and influential, as well as a prominent critic of large technology companies Ed Zitron, stated that he believes these mascots are mainly a form of promotional strategy.
“This is pure marketing,” he told Fortune magazine. “The purpose of its existence is to ensure that articles quote statements like ‘Meta gave AI a cute face.’”
This statement is also reasonable – that’s exactly how this article is written, except that the word “marketing” has been removed. However, these “faces” are not just for decoration. The design goal of these agents is to work across individual accounts and execute tasks on behalf of users. Taking Muse as an example, it can search for products, navigate through the checkout process, and prepare for purchase before final approval. Meta indicates that it will display the total amount to the consumer for confirmation before making a payment through the Link service provided by Stripe.
An associate professor of public relations practice at Boston University, Justin Joseph, told Fortune magazine that a friendly face may make this technology seem more accessible, but it could also influence users' expectations of its security and reliability before they have a real opportunity to assess its performance. If an agent's image appears cute, friendly, and approachable, “consumers might infer that this AI agent is likewise approachable, secure, trustworthy, and easy to use,” said Joseph. Users “might start to see it not as software so much, but more as a partner or assistant.”
This is a high requirement for a mascot, and consumers are not yet at that stage. Previously, Fortune cited a survey from Vogue Business which indicated that 72% of respondents stated they would not share their bank card information, and only 31% were willing to outsource their shopping to an agent that understood their taste and purchasing history. The question is, how much of this confidence comes from the role itself, and how much from the agent's actual ability to complete the work?
Give a familiar face to unfamiliar technologies
Characters can bring abstract technologies back to a human scale. Joseph told Fortune magazine: “It’s much easier to say ‘Have you seen/useled Dots before?’ than to explain the technical essence of AI agents. I’d rather talk to you about my ‘Roomba’ than discuss ‘autonomous robot vacuums’.”
This is not the first time that tech companies have given their software a more user-friendly interface. Joseph mentioned Siri, Alexa, as well as Microsoft’s Office which includes an animated paperclip that provides assistance Clippy – however, many users find it more annoying than helpful.
A character can also become a sustainable marketing asset, appearing in social media posts, merchandise, and product education content.
"Introducing a character is like pushing a celebrity into popular culture," he said.
Meta is treating Jolly as a character at the level of A-list. The company's chief AI official Alexandr Wang released Jolly emoticons on X, giving this mascot a vitality that goes beyond the product itself. Other companies such as Spotify and GitHub also joined in, releasing their own versions of Jolly to announce their integration with Muse.
OpenAI indicates that its design team collaborated with external partners to develop the Dots character, drawing some inspiration from the earlier Codex Pets feature. The goal was to create a design that would not interfere with work and that was approachable to users. The team explored abstract shapes, recognizable objects, as well as different materials and textures, and ultimately decided to adopt a flexible system composed of shapes, colors, and accessories. Users do not necessarily have to choose a character or name it when they start using it. In addition, the company states that names and different appearances also help people distinguish between different agents, especially when multiple agents share the same workspace. Animation cues also allow users to see when an agent is working.
Meta did not respond to Fortune's questions regarding the design of Jolly.
Not everyone is captivated by the charm of these characters. On September 30th, Fairplay Executive Director Josh Golin issued a statement accusing Meta and OpenAI of using "child images" to sell AI products, and called on regulatory authorities to investigate whether these companies are deliberately attracting children to use their services.
Looking cute is just the easiest step.
So far, the records of these proxies are not as appealing as their appearance might suggest. OpenAI revealed two months before the launch of Dots that during a network security test, its model gained unauthorized access to the infrastructure of Hugging Face while some security measures were intentionally disabled. GeekWire reported that Amazon blocked Muse from accessing its website in September due to security concerns. Additionally, according to Ars Technica, a researcher discovered a vulnerability that could allow attackers to take control of Muse accounts; Meta stated that a fix has been released.
Both companies emphasized protective measures. OpenAI indicates that a role cannot convey everything an agent is capable of doing, nor can it explain all of its limitations. In Activity View, users can track the progress of the agent and inspect the computer it uses, which is part of the visibility and control provided by that role. The announcement also reminds users to review tasks that may have significant impacts. Meta states that users can choose which applications Muse can connect to and what level of access they want to grant, and they can disconnect from the service at any time. The companies also stated that the conversations and data from Muse will not be shared with their advertising systems.
Joseph said, "I believe that characters can lower the psychological barrier for people when it comes to embracing new technologies. However, we should also remember that characters are not human; they have their own technical and psychological limitations."
Six-month test
According to Zitron, these mascots are an attempt to personify large language models, giving the software human characteristics in order to make them seem alive or conscious. When asked whether these characters would change the public's perception of AI and its risks, he replied: "Not at all, not in the slightest."
Joseph indicates that he will track awareness, media discussions, download volumes, and retention rates, but it may not be easy to attribute these changes to a particular factor. He will also measure trust: “Has more people come to trust and understand our brand because of Dots or Jolly?”
The test for Zitron is even simpler: “Come back in six months to see where Muse ranks on the iOS leaderboard.”
