AI chatbots are taking on recognizable personalities. Meta has introduced celebrity-inspired bots in its messaging apps, joining a wider group of products that invite people to talk with AI characters. The approach could make conversations feel more engaging, but it is still an open question whether that interest will last.
Personality becomes part of the product
At its annual Connect conference, Meta announced AI-powered chatbots for WhatsApp, Messenger and Instagram DMs. Available to select users in the U.S., they are designed to channel particular personalities and mimic celebrities including Kendall Jenner, Dwyane Wade, MrBeast, Paris Hilton, Charli D’Amelio and Snoop Dogg.
The launch is a bid to encourage engagement across Meta’s platforms, especially with younger users. The article points to a 2022 Pew Research Center survey: about 32% of internet users aged 13 to 17 said they ever used Facebook, an over-50% decline from the year prior. Character-based bots reflect a broader interest in AI that offers a persona, rather than only an answer to a task.
Character.AI, for example, lets users create AI companions with distinct personalities. Its examples include Charli D’Amelio as a dance enthusiast and Chris Paul as a pro golfer. This summer, its mobile app gained over 1.7 million new installs in less than a week, while its web app was receiving more than 200 million visits per month.
Character.AI said that, as of May, users spent an average of 29 minutes per visit. The company said that figure eclipsed ChatGPT by 300% as ChatGPT usage declined. The company’s growth also drew major backing: Andreessen Horowitz invested well over $100 million, and Character.AI was last valued at $1 billion.
Different characters, different uses
Other products illustrate how varied the category is. Replika, described as a controversial AI chatbot platform, had around 2 million users in March, including 250,000 paying subscribers.
Inworld is building a platform for creating more dynamic non-player characters in video games and other interactive experiences. The company has not shared much about usage, but its promise of expressive, organic characters has attracted investments from Disney and grants from Fortnite and Unreal Engine developer Epic Games.
These examples point to several possible draws: a character can give a conversation a recognizable identity, and a companion may be used differently from a tool built for a professional task. That distinction is central to the appeal. General-purpose chatbots such as ChatGPT and Claude can be useful, but they were designed to complete specific tasks, not necessarily to sustain an enlivening conversation.
The open question is whether interest will last
Meta’s launch signals that it sees value in personality-led bots. But early attention does not establish that people will keep returning. The novelty could wear off, as it can with other technology, leaving companies to find out whether the experience offers reasons to come back over time.
There is also a difference between engagement and usefulness. Visit counts and time spent show that people are trying these services, but they do not settle whether AI characters become a lasting part of messaging, gaming or everyday life. The products may develop in different directions, and their staying power remains uncertain.
Other research explores what machine learning can do
Separate projects show machine learning being applied to tasks beyond conversation. Researchers at the University of Edinburgh built a digital network based on observed insect neural networks. It navigated a small robot visually with limited resources, suggesting a possible approach for systems constrained by power and size.
Sony introduced a skin-color metric intended to describe color more comprehensively using a color scale and perceived lightness or darkness. The work found that bias in existing systems affects skin hue as well as lightness. Google’s RealFill, meanwhile, can fill missing parts of a photo using other images from the same scene. Its additions remain generated guesses, even when informed by those images.
Another research direction concerns earthquake aftershocks. The models discussed in the article characterize aftershocks after an earthquake; they do not predict earthquakes themselves. They are still only decent under specific circumstances, but may help seismologists process large amounts of data. The researchers emphasize that preparation matters, because knowing an earthquake is coming does not stop it.
Together, these developments show the breadth of current AI work, from more personable chatbots to image tools and scientific analysis. For personality-driven AI, the central test is simpler: whether a compelling character remains compelling after the first conversation.