Billions of people use Facebook and Instagram every day to share the ups and downs of their lives, connect with people with similar interests, and find content they enjoy. To make every user’s experience on our apps unique and personalized, we use AI systems to determine what content is displayed based on user selections .
i wrote before About Your Relationship with Algorithms Meta is used to shape what we see on Facebook and Instagram and challenge the myth that algorithms are powerless over the content people watch. In that article, I wrote that I needed to be more outspoken about how this relationship worked, and that I needed to be more in control of what I saw.
We are now stepping up that effort by being more transparent about some of the AI systems that incorporate user feedback to rank content on Facebook and Instagram. These systems make the posts you see more likely to be relevant and interesting to you. We’re also testing new controls and making other controls more accessible, as well as making it clearer how you can better control what you see in your app. We also provide more detailed information to experts so that they can better understand and analyze our system.
It is part of a broader ethos of openness, transparency and accountability. With powerful technologies like generative AI advancing so quickly, it’s no surprise that people are both excited about the possibilities and concerned about the risks. We believe that the best way to address these concerns is to be open. In general, as these technologies are developed, companies are becoming more open about how their systems work, and are openly collaborating across industry, government, and civil society to develop them responsibly. We believe that it is necessary to ensure that It starts with giving you greater insight and control over the content you see.
Impact of AI predictions on recommendations
Our AI system predicts how valuable your content is to you, so you can view content faster. For example, sharing a post often indicates that you found the post interesting, so the prediction of sharing a post is one factor the system takes into account. As you can imagine, there is no single prediction that can perfectly determine if a post is worth it to you. Therefore, we combine various predictions, including predictions based on behavior and predictions based on user feedback received through surveys, to get as close to the right content as possible.
We would like to be more open about how this works. A model of transparency that Meta has been developing and advocating for some time is the disclosure of system cards. This provides insight into how the system works in a way that is accessible to anyone without deep technical knowledge.today we release 22 system cards for Facebook and Instagram. They provide information about how AI systems rank content, some of the predictions each system makes to determine which content is most relevant to users, and controls you can use to customize the experience. increase. These cover feeds, stories, reels, and other surfaces you visit to find the accounts you follow and your content. System cards also cover AI systems that recommend “disconnected” content from people, groups, and accounts you don’t follow. You can find a more in-depth explanation of the AI behind content recommendations. here.
To further explain what is exposed in the system card, the types of inputs called signals and the predictive models derived from these signals that help determine which content is most relevant from the network share. On facebook. The categories of signals we release represent the majority of the signals currently used in ranking this content’s Facebook feed. These signals and predictions are transparent centeralong with the frequency used in the overall ranking process.
We also use signals to identify and remove harmful content when we become aware of it. It also helps us reduce the delivery of other types of problematic or low quality content in line with our content delivery guidelines. Here are some examples of signals used to do this. However, there are limits to what can be safely disclosed. We want to be transparent about how we keep bad content out of people’s feeds, but we also need to be careful about exposing signals that people could easily bypass our defenses. I have.
Of course, just because you publish information on a website doesn’t mean that everyone will find it. That’s why we provide details directly in-app about why our systems predicted content to be relevant to you, and what kinds of activities or inputs may have led to that prediction. We are expanding “”.Why am I seeing this?The feature will be added to Instagram’s Reels tab and Explore, as well as Facebook Reels, in the coming weeks, after it was previously released for some and all feed content. advertisement Both on Facebook and Instagram. By clicking on individual reels, you can see detailed information about how previous activities informed the machine learning model that formed and delivered the displayed reel.
Expanded tools to personalize your experience
The tools available allow you to shape your experience on the app so you see more of the content you want to see and less of the content you don’t. To make this easier, we’ve created centralized locations on Facebook and Instagram where you can customize controls that affect the content you see in each app.Three-point menu of related posts and[設定]where you can access your Facebook feed settings and Instagram’s Featured Content control center.
Instagram is testing a new feature that allows you to indicate that you’re “interested” in recommended reels in the Reels tab. This will allow you to view more of your favorites. “Am not interested” feature is available from 2021. Learn more about influencing what you see across Instagram here.
There’s also a “show more, show less” feature so that you can customize the experience and the content you see. feature On Facebook, it’s available on all feed, video, and reel posts from the three-dot menu. We are looking into ways to make this feature more prominent. If you don’t want an algorithmically ranked feed, or just want to see what your feed would look like without it, “Feed” tab Facebook or choose Continue To switch to a chronological feed on Instagram, you can also add people to both favorites lists. Facebook and Instagram You can always see the content of your favorite accounts.
Providing better tools for researchers
We also believe that when it comes to research and innovation, especially transformative AI technologies, an open approach is better than leaving the know-how in the hands of a few large tech companies. That is why over the past decade, he has released over 1,000 AI models, libraries and datasets to researchers, allowing them to benefit from our computing power and advance their research openly and securely. I’ve been to Our goal is to continue to be transparent in order to make more AI models openly available in the future.
In the coming weeks, we will begin rolling out content libraries and APIs for Meta, our new suite of tools for researchers. Libraries contain data from public posts, pages, groups, and events on Facebook. For Instagram, this includes public posts and data from creators and business accounts. Data from the library can be searched, explored, and filtered through a graphical user interface or programmatic API. Researchers at qualified academic and research institutions pursuing research themes of scientific or public interest should seek safe and secure places for research, including the Interuniversity Consortium for Political and Social Research at the University of Michigan. You will be able to apply for access to these tools through partners who have deep expertise in sharing data. research. These tools provide the most comprehensive access to content published on Facebook and Instagram of any investigative tool we’ve built to help meet emerging data sharing and transparency compliance mandates. is also helpful.
By introducing these products to researchers early in the development process, we want to receive constructive feedback and ensure that we build the best tools to meet their needs. .
