Meta AI. What should you know about Facebook and Instagram’s AI-supported features? | AI in business #46

Meta AI. The artificial intelligence of Facebook and Instagram

AI ethics is about how AI-based technologies should be designed, developed, used and managed responsibly and following human values.

Facebook and Instagram are leveraging the capabilities of artificial intelligence in many fields. One of the applications available to English-speaking users is Meta AI, an assistant that can be spoken to 1-on-1 or in group chats using WhatsApp, Messenger, Instagram, and Facebook Stories. It can make recommendations, tell jokes, and answer questions.

The service is built on the Llama 2 model, which Meta released for commercial use in July 2022. Llama 2 is a large language model (LLM), trained on texts totaling hundreds of billions of words taken from the Internet. As a result, it can, among other things, translate between more than 200 languages or generate texts based on short clues. It can be used by companies to create chatbots, customer service systems, or text analysis.

In addition, Meta AI has the new Emu (Expressive Media Universe) model, which generates images based on text prompts. It makes it possible to create visualizations of concepts that cannot be captured with a camera. Just by typing in, for example, “Plush bear doing AI research,” the AI can generate fancy graphics based on that.

Source: Meta

With its models, Meta is introducing new characters for chatbots to converse with, as well as generative tools for Facebook and Instagram to create custom stickers or change the style of photos with text commands.

AI development in the Met ecosystem. What does this mean for companies?

However, the Met’s artificial intelligence is not just about improving Facebook’s functionality or making Instagram more enjoyable.

New Meta AI models

For companies, the development of artificial intelligence in the Meta ecosystem primarily means access to open models on which they can build their solutions. Indeed, Meta’s research labs are not only working on Meta AI, Llama 2 or Emu. Meta is making more and more AI models available under an open license. For example, SeamlessM4T for translation. This is an advanced language model for translating speech and text between multiple languages. It achieves better translation quality than previous Meta systems. It can process up to 128 languages and is available for free download.

Another example is Habitat, a physical simulator for training intelligent virtual agents. These agents learn to walk, manipulate objects and interact with the world, which could have applications in robotics or the metaverse.

AI in Meta Ads

Another area of Meta AI of particular importance to companies involves incorporating artificial intelligence in creating and publishing Meta Ads. Meta has announced that it is introducing new AI-based tools for all advertisers to create images and text faster. The capabilities include:

• Generating background images for products,
• Creating different versions of advertising text, or
• Automatically resizing ads and zoom in to fit Instagram or Facebook.

Source: Meta

The new features will be available to advertisers in Meta’s ad manager, and the release of the new features to all companies will be completed next year.

Overall, Meta seeks to advance AI through research and application to products. It also makes solutions available to the public, such as:

• PyTorch – a Python-based machine learning framework popular among engineers and scientists,
• Glow – a framework designed for rapid machine learning based on C++,
• ONNX – a machine learning model exchange format that allows models to be shared between different frameworks.

For companies, this means access to powerful AI tools that they can leverage in:

• ustomer service – for example, creating chatbots,
• marketing – including for ad targeting, or
• analyzing customer sentiment on social media – to analyze the emotional attitude of observers to new products introduced.

Artificial intelligence moderates Meta content

However, in Meta AI is primarily used for content ranking and moderation. On Facebook’s help center page, one can learn that in Meta AI is the first tool to prevent the publication of harmful content:

“Artificial intelligence technology is a key component of our content verification process. Artificial intelligence can detect and remove content that does not comply with our Community Standards before anyone reports it.”

For example, in the first half of 2022, Facebook removed more than 5 million pieces of content infringing copyrights, trademarks, etc. reported by users. Meta is also developing new methods for detecting such violations.

Security and privacy on Meta AI platforms

The question on many people’s minds about Meta AI’s great results is: how did Meta get the data to train its models? The answer can be found in Meta’s Privacy Center. Meta AI’s knowledge comes from:

• from public posts from Facebook and Instagram, excluding private content to respect users’ data,
• from licensed sources obtained from other suppliers, and
• From information publicly available on the Internet.

Meta says it trained AI responsibly, with safeguards like a ban on generating fake images of public figures. It also reveals details on security signals that help reduce the spread of harmful content. This is intended to increase transparency and user trust. At the same time, only 18% of users in the U.S. believe that Facebook protects their privacy and data. This is the lowest of all social media platforms and down from 30% in 2020.

To shore up its reputation and credibility, and to stop being associated with fake news, Meta has decided to prepare for the 2024 US elections. It has introduced regulations that require ads tying themselves to political campaigns to disclose the use of artificial intelligence in ads. Under the new policy, Meta AI will be visible through labels that appear on screens when they click on an ad.

Meta AI – summary

Meta has recently been striving to be transparent about its AI in digital interaction and marketing. At the same time, it is developing advanced AI models and Meta AI generative solutions. This is changing the way companies and users use social media. The company is making more and more AI tools available for business use, from chatbots to language models or simulators. By using these capabilities responsibly, businesses can streamline many processes.

Nonetheless, growing concerns about the privacy and ethics of AI point to the need to balance technological innovation with social responsibility. As Meta continues to develop its Meta AI, it will be crucial to monitor how these developments will affect the future of digital interactions and balance technological advances with user privacy.

If you like our content, join our busy bees community on Facebook, Twitter, LinkedIn, Instagram, YouTube, Pinterest, TikTok.