Facebook has taken the wraps off a project called Learning from Videos. It uses artificial intelligence to understand and learn audio, textual, and visual representations in public user videos on the social network.
Learning from Videos has a number of aims, such as improving Facebook AI systems related to content recommendations and policy enforcement. The project is in its early stages, but it's already bearing fruit. Facebook says it has already harnessed the tech to enhance Instagram Reels recommendations, such as surfacing videos of people doing the same dance to the same music. The system is showing improved results in speech recognition errors as well, which could bolster auto-captioning features and make it easier to detect hate speech in videos.
Facebook says the project will help AI researchers avoid having to rely on labeled data and it's part of efforts to build systems that learn in a similar way to humans. As such, Learning from Videos will "enable entirely new experiences." The company didn't go into much detail about those except for a possible feature that would allow AI to find digital memories, including ones captured by augmented reality glasses. You could, for instance, ask such a system to show you "every time we sang to grandma," and it could surface those clips. Facebook, notably, is .
The company says the project is looking at videos in hundreds of languages and from almost every country. This aspect of the project will make AI systems more accurate and allow them to "adapt to our fast moving world and recognize the nuances and visual cues across different cultures and regions."
Facebook says that it's keeping privacy in mind when it comes to Learning from Videos. "We’re building and maintaining a strong privacy foundation that uses automated solutions to enforce privacy at scale," it wrote in a blog post. "By embedding this work at the infrastructure level, we can consistently apply privacy requirements across our systems and support efforts like AI. This includes implementing technical safeguards throughout the data lifecycle."
Understanding what's happening in videos can be an immensely difficult task for AI systems. They can include hurdles like background noise that makes it difficult to understand speech and language switching. Yet less than a year after starting the Learning from Videos project, Facebook is taking what the system has learned and putting it to practical use in other areas.