Facebook AI Research, the Menlo Park, CA-based company’s AI research arm made two important updates last week. First off, it open sourced Droidlet to enable the development of AI robots. The social media company is also using AI to make teenagers safer on Facebook as well as Instagram.
Facebook Open Sources Droidlet
Last week, Facebook open sourced a platform that enables a machine to learn the elements in the environment that they are in. Called DroidletOpens a new window , the platform enables developers to build ‘embodied agents’ using robotics that leverage computer vision, natural language processing (NLP) that are in a way conscious of their surroundings.
“There is much more work to do — both in AI and in hardware engineering — before we will have robots that are even close to what we imagine in books, movies, and TV shows,†wrote Mary WilliamsonOpens a new window , research engineering manager, and Arthur SzlamOpens a new window , research scientist at Facebook. “But with droidlet, robotics researchers can now take advantage of the significant recent progress across the field of AI and build machines that can effectively respond to complex spoken commands like ‘pick up the blue tube next to the fuzzy chair that Bob is sitting in’.â€
The idea is to teach machines the fundamentals of information processing, and thereby comprehension of human commands. Not only can the platform help in understanding real-world scenarios, such as the consequences of knocking over a soda bottle, it also can be used to teach machines to perform activities in simulated environments such as Minecraft gameplay or Facebook Habitat.
Facebook terms the platform as a platform for hobbyists as well as researchers to “build more intelligent real-world robots.†Facebook’s Droidlet goes beyond the different, specialized tools such as PyRobotOpens a new window (robotic), AllenNLPOpens a new window (natural language processing), DetectronOpens a new window (computer vision), etc., and consolidates different iterations of these technologies onto one single platform.
What is Facebook’s Droidlet Made Up Of?
Facebook describes Droidlet as a modular, heterogeneous embodied agent architecture made up of several different components or modules. Some of these modules are heuristic while others are learned through continuous improvements.
Droidlet’s high-level interface consists of the following:
- Memory system: Embodied agents, machines, or robots access stored data in the memory system located at the interconnection of different modules
- Perceptual modules: Tasked with processing the information received from the outside world such as an object, a pose, or stance in case of a living thing, etc. It also stores this information on the memory system.
- Lower-level tasks: These tasks basically bring about small changes in the agent’s environment. Commands are generally menial, such as, “move three feet forward†and “place item in hand at given coordinates.â€
- Controller: Prioritizes task execution depending on the state of the memory system
With Droidlet, Facebook intends to build towards the development of more capable high-level robotic agents through modularity. If one contributor is able to improve a piece of the system, it should improve all the robotic agents using that piece of the system.
Opens a new window
Source: Facebook
The interesting thing about these individual components is that not only can all of them be further trained for implementation within Droidlet, but also outside of it.
Facebook VP & chief AI scientist Yann LeCunOpens a new window told France24Opens a new window a couple of weeks ago, “In the near future, or further, depending on the progress we make, we could have intelligent personal assistants that understand nearly everything we say and are not frustrating to interact with. We could possibly also develop robots that are somewhat intelligent and autonomous: machines that we give a task to but don’t explain how to do it.â€
See Also: Facebook Blender 2.0: The Newest Chatbot That Recalls and Surfs the Web
Facebook Becomes a Little More Safer for Teenagers
Facebook and Instagram both, actually.
The world’s largest social network is bringing in big guns to curb the threat to teenager privacy amid some criticism the company faced. So Facebook is now using artificial intelligence to detect and thwart any attempts at infringing privacy of the platform’s young users.
Last week, Facebook announced that it will use AI to detect and stop any attempts at bypassing the platform’s policies of a minimum age requirement of 13. “We’re developing AI to find and remove underaged accounts, and new solutions to verify people’s ages,†said Pavni DiwanjiOpens a new window , VP of Youth Products at Facebook. “We’re also building new experiences designed specifically for those under 13.â€
Facebook’s new undertaking is applicable for its subsidiary Instagram, the fourth largest social network as well.
Age Detection Is Harder Than It Sounds
The company acknowledges the difficulties in detecting user age considering a lot of unqualified users, i.e., those below the age of 13 lie while registering on Facebook. The company explored collecting some sort of government-issued identification but hit a roadblock and listed more than a few “significant limitations†to this approach.
These include not having an ID, is not equitable, isn’t foolproof, accessibility to IDs is dependent on extraneous factors, while some can’t afford it or are uncomfortable sharing it. This is why Facebook is starting to look at subtle hints dropped by users while using Facebook.
“We look at things like people wishing you a happy birthday and the age written in those messages, for example, ‘Happy 21st Bday!’ or ‘Happy Quinceañera.’ We also look at the age you shared with us on Facebook and apply it to our other apps where you have linked your accounts and vice versa — so if you share your birthday with us on Facebook, we’ll use the same for your linked account on Instagram,†the company said.
Facebook’s attempts at controlling age misrepresentation also include collaboration with operating system (OS) providers, internet browsers and other providers for information sharing.
Besides interacting regularly with a group of online safety, child development, children’s/youth media, privacy, youth development, psychology, and parenting experts, the company said it is building new experiences for teenage users. Facebook also said it is collaborating with lawmakers and elected officials although none in particular were named.
New Facebook and Instagram Experiences for Teenagers
Based on the AI-driven technology, Facebook’s plans for age-appropriate experiences include:
- Instagram accounts categorized as those of users who are under 18 won’t receive messages from adults that they do not follow.
- Additionally, posts from these accounts won’t be shown to adult-owned accounts that have exhibited suspicious behavior in the past.
Closing Thoughts
Facebook AI Research or FAIR is making good use of the technology to enable others as well as itself to make sure products are a little bit better. This doesn’t mean it is infallible and completely effective. This is why the European Union sought to introduce AI regulations a few months ago.
Facebook’s LeCun believes this is a step in the right direction provided the application, and not the technology is regulated. So without giving a free hand to the use of AI for upholding teenage privacy, the research that goes into making this technology shouldn’t be constrained.
A more appropriate example would be to allow the research in making AI robots, but controlling how the tech is implemented. Fans of the three-decade-old Terminator, a wildly popular movie, would be quick to agree with EU’s new regulations to avoid the fate of AI robots taking over the world, which weirdly enough still remains a hot take on AI-related discussion.
Let us know if you enjoyed reading this news on LinkedInOpens a new window , TwitterOpens a new window , or FacebookOpens a new window . We would love to hear from you!