Artificial intelligence at MetaNew AI functions in Instagram, Facebook & WhatsApp

The tech company Meta will now also integrate its AI assistant Meta Llama 3 into WhatsApp, Instagram and Facebook under the name Meta AI. The new functions are not yet available in Germany. However, it is only a matter of time before WhatsApp and co. will also be equipped with artificial intelligence here. klicksafe summarizes how Meta AI will change the user experience and what users should pay closer attention to.

Meta has presented the next generation of its text AI Llama 3. The Large Language Model (LLM) is said to be more powerful than most current models. It is currentlyonly available in some English-speaking countries. Meta has not provided any specific information on when users in the European Union will be able to access it. However, it is working on the market launch. A similar situation occurred recently with the launch of Threads. This service was also only available in the EU several months after its launch in the USA.

What can the new Meta AI chatbot do?

The integration of the AI bot is quite similar for Instagram, WhatsApp and Facebook Messenger. Users can ask Meta AI questions in the search slot. It then answers the questionsin real time. If the artificial intelligence is unable to answer a question, search queries are submitted to Google or Bing, for example. The results from these search queries are then summarized by the chatbot in a message. This means that users no longer have to leave the apps if they have a question that they previously asked via external search engines.

Users can also have Facebook and WhatsApp generate images in real time. These are created at high speed as they are typed. These images can then be sent or posted.

Basic information on use

The use of Meta AI is permitted from the age of 13. This age limit is based on the minimum age for all Meta platforms, which is 13. Meta obtains the age information mainly from the information provided by the user when creating the account. Anyone who has entered an incorrect date of birth can therefore use Meta AI even if they are under 13.

The integration of chatbots in other applications (e.g. My AI on Snapchat) has led to concerns that children in particular are not always aware that they are communicating with a language model and not a human being. According to Meta, the aim is to prevent such confusion. It should therefore be clear to users at all times that they are not interacting with a human, but with an artificial intelligence. To achieve this, WhatsApp chats with the AI, for example, are graphicallydifferent from those with human contacts. The AI should also never be able to contact users independently, but only respond when a request is made to it.

The General Terms and Conditions stipulate that Meta AI may not be used to create prohibited content. This includes, for example, content that describes the sexual abuse of children and content that deceives, endangers or infringes the intellectual property of others.

What data does Meta AI store?

Meta states that all conversations and requests can be used to further train the language model. This includes, for example, personal characteristics (e.g. personal preferences) that have been disclosed to the chatbot. This should allow the AI's responses to be tailored to the individual user. However, Meta recommends that you do not disclose any sensitive personal data when talking to the chatbot.

According to Meta, there is no need to worry about private conversations on the Meta platforms being used to train the language model. According to the company, messages in WhatsApp, for example, are protected with end-to-end encryption to prevent humans or machines from being able to read them.

Meta also states that chat content from Meta AI is shared with other companies. As things stand, these are Finnhub, Google, Microsoft and Wolfram. This should only involve chat content and not personal information. Unless this personal information was disclosed by the users themselves in the chat with Meta AI.

How to stay in control of your data

  • You can use the following command to see what personal information the AI has saved: /saved-details
  • If you want to download your saved data, enter the following command in the WhatsApp chat: /download-all-ai-info
  • On Facebook and Instagram, you can download your download your information here.
  • If you want the AI to forget all previous data from individual chats, you can use the following command in the chat in question: /reset-ai
  • If you want the AI to forget all previously collected information across all meta platforms, use this command: /reset-all-ais

How reliable is Meta AI?

As with all language models, the answers to Meta AI should also be viewed with great caution. Meta itself is of this opinion and states in its General Terms and Conditions:

"AIs and the Outputs are provided "as is" and"as available".Furthermore, we do not warrant that they will be safe, secure or error-free [...]. We are not responsible for the Outputs. You are responsible for checking the outputs for accuracy and suitability."
(emphasis added by klicksafe)

Even a brief test by klicksafe with the English version of Meta AI resulted in incorrect information. When asked whether there is a minimum age for using Meta AI, the language model incorrectly replied that the minimum age was 18. When it was pointed out that there was nothing about this in the General Terms and Conditions, the language model admitted that it had made a mistake and that the original information was incorrect. Subsequently, Meta AI was still unable to provide the correct information and then had to correct itself again when it was pointed out that the General Terms and Conditions stipulate a minimum age of 13 years.

This shows that the information provided by these language models is unreliable and must always be cross-checked. Our teaching material "How reliable is ChatGPT?" and our flyer for young people "Stay safe! 5 tips for AI in everyday school life".

The data used to train the language model may be incorrect or incomplete. It is also possible that the training data contains personal information. If you would like to know whether your personal information is included in the training data or would like to take action against it, please refer to the tips in the following box.

Personal information in the training data

If you discover that Meta AI knows personal data about you, you can take action against this. According to Meta, personal data can accidentally end up in Meta AI's training data. For example, if this information is publicly available on the Internet. In this case, you have the right to object to its use. You can also have data corrected if it contains incorrect or outdated information about you.

Further information on the topic of AI from klicksafe