Meta Platforms started rolling out its built-in artificial intelligence (AI) assistant for WhatsApp in South Africa last week, exposing many more people to text-to-image generation and AI-powered online search features.
However, given its parent company’s notorious reputation for collecting vast amounts of personal data from users and, in some instances, failing to protect it adequately, some might be concerned about the implications this tool will have on their private conversations or in-app searches.
The Meta AI tool, powered by the company’s LLaMA large language model, first became available to select users in September 2023.
It has gradually been expanded to more countries over the past few months, with South Africa and India gaining the feature from last week.
Meta said the tool aimed to help its many users access Generative AI to enable them to be more creative, productive, and entertained — simply by sending a message.
“We want people interested in trying out this experimental technology to know some basics and what to expect,” the company said.
The core capability of the new tool is AI chats, which will respond to users’ questions about a wide range of topics, similar to a search engine.
This includes getting the perspective from dozens of “characters” Meta has created that respond in compelling ways.
To access this feature, users must tap the button with the blue, purple, and turquoise circle.
On Android, this appears above the “new chat” button in the bottom-right section of the home page, while iOS users will find it in the top toolbar to the left of the camera button.
Once you’ve agreed to Meta’s terms and conditions for using the tool, you will find it is also capable of generating images of ideas, places, or people from text prompts.
To use this feature, users can type in the prompt “/imagine” followed by an explanation of what they want the image to present.
Below are some of the images we generated using this feature. The phrases used for each image are in the caption below.
These features are only the initial AI-powered capabilities that Meta plans to integrate into WhatsApp.
It also plans to add support for more languages in the future and is adapting the WhatsApp API so that developers and businesses can use it to create AI-powered chat experiences for their customers.
Meta Platforms has previously been criticised for mishandling users’ personal data to improve its products and increase revenue.
Its previous behaviour regarding a WhatsApp privacy policy change for chats with businesses might have many concerned about the intentions of its AI tools.
Meta Platforms has provided details on how the Meta AI tool protects user privacy in the FAQ section of the WhatsApp website.
It explains the “most important thing” is that users’ personal messages with friends and family are off-limits to the AI.
“No matter how realistic these conversations become, talking to AI has different rules and expectations than personal conversations with friends and family,” the company said.
“AIs can read what is sent to them, but your personal messages remain end-to-end encrypted, so no one else, including Meta, can see them.”
“We’ll make chats with AI visually distinct so it’s clear these are different from personal messages and conversations.”
In addition, using the Meta AI Chat will not link your WhatsApp account information on Facebook, Instagram, or any other apps provided by Meta.
AI Chat promotion muddies the waters
Some users might be concerned that searching for a sensitive personal keyword or media file in their own chats might share this information with the Meta AI, as Meta is promoting the AI-powered tool in the regular search function of the app.
However, it is only when a user selects one of the suggested prompts under the “Ask Meta AI” section that they actually begin engaging with the AI tool.
“You have to take action to start the conversation by opening a chat or sending a message to the AI,” Meta said. “Only you or a group participant can initiate this, not Meta or WhatsApp.”
Meta has cautioned that users should not send messages with information they do not want someone else to know in AI Chats.
This is because this information is used to provide users with accurate responses or to improve the model.
“Even still, the AI is trained to limit the possibility of information about people, like names for example, from showing up in other chats,” Meta said.
It also warned that Generative AI was a new technology, and it was important to remember that responses might not be accurate or appropriate.
Loading ...
Join the conversation Autoload comments
Comments section policy: MyBroadband has a new article comments policy which aims to encourage constructive discussions. To get your comments published, make sure it is civil and adds value to the discussion.