Meta Sets up Chatbots With Personas in an Effort to Keep Users

 In an effort to increase engagement on its social media platforms, Facebook owner Meta is getting ready to roll out a variety of AI-powered chatbots with distinct personalities as early as next month.

According to three people with knowledge of the plans, the digital giant headed by chief executive Mark Zuckerberg has been developing prototypes for chatbots that can interact with its roughly 4 billion users in a human-like manner.

These individuals claimed that some of the chatbots, which employees have termed "personas," assume the appearance of various personalities. According to a person with knowledge of the plans, the corporation has considered launching one that takes after Abraham Lincoln and another that offers travel advice in the manner of a surfer.

According to the source, the chatbots might be live as soon as September. They will serve as a fun product for users to play with while also offering a new search function and recommendations.

The $800 billion firm is making the move as it fights off competition from social media upstarts like TikTok and tries to capitalise on the growing enthusiasm around artificial intelligence (AI) in Silicon Valley following the introduction of Microsoft-backed OpenAI's ChatGPT in November.

Chatbots might increase engagement while also gathering a tonne of fresh information on the interests of users, according to experts. This would enable Meta to target consumers more effectively with relevant content and advertisements. The majority of Meta's $117 billion in annual sales come from advertising. 

"Once users interact with a chatbot, it really exposes much more of their data to the company, so the company can do anything they want with that data," said Ravit Dotan, a researcher and AI ethics expert. 

She continued that the new advances pose questions about privacy as well as potential "manipulation and nudging". 

Meta chose not to respond.

Competing IT companies have already released chatbots with personalities. Large language models are used by, an Andreessen Horowitz-backed startup with a $1 billion valuation, to create discourse in the manner of people like Tesla CEO Elon Musk and Mario from the Mario franchise.

150 million of Snap's users have engaged with the "experimental, friendly chatbot" that makes up its "My AI" feature, which was introduced in February. The "early testing" of sponsored links within the functionality just recently started. 

On a Wednesday earnings call, Zuckerberg informed analysts that the business would provide more information on its product roadmap for AI in September at its Connect developer event.

According to Zuckerberg, AI "agents that act as assistants, coaches, or that can help you interact with businesses and creators" are what he envisions, and he added: "We don't think that there's going to be one single AI that people interact with." 

Additionally, he stated that the business was developing internal AI-powered productivity assistants for staff members as well as AI agents that may assist enterprises with customer support.

According to a source with knowledge of the situation, constructing an avatar chatbot in the metaverse would be investigated in the long run. This person said, "Zuckerberg is expending all of his effort and time on ideating about this. 

In generative AI, which can produce text, graphics, and code, Meta has been making investments. This month, it made available for purchase Llama 2, a substantial language model that may power its chatbots. 

According to two people with knowledge of the situation, Meta has been attempting to acquire tens of thousands of GPUs as part of setting up the infrastructure to serve the AI products. These chips are essential for powering substantial language models. 

Experts monitoring the chatbots for bias or the possibility that they can disseminate harmful information or "hallucinations" will likely pay close attention to Meta. 

The business has already undertaken small-scale, transient ventures into chatbots that have shown similar concerns. Researchers discovered that the 2021-released BlenderBot 2 Meta AI model immediately began disseminating false information. Although people continued to report that it produced inaccurate information and hate speech, Meta claimed that it made the BlenderBot 3—released in 2022—more resistant to this content. 

An insider at Meta claims that the business will likely incorporate technology that will screen user questions to make sure they are proper. In order to verify that what is said is true and avoids, for example, hate speech or speech that violates the law, the company may also automate checks on the output from its chatbots, the source continued. 


Previous Post Next Post