• Sunday, May 19, 2019

    What exactly is Facebook doing with AI?

    Comment

    If having high-end, self-learning computers and they are monitoring all your activities on the site makes you wonder a bit, you may not want to get too deep into specialized labs. Where is Facebook's AI study? Photo tagging, friend suggestions, fake news filters, timeline arrangements and many other Facebook features depend on some versions of AI. It means that no human can handle up to 2.19 billion monthly users. The scale and rate that Facebook is building AI for its products is well worth exploring.

    Image recognition

    Image recognition

    Face recognition and auto-tagging are just a small part of Facebook's machine learning ability. Facebook is actually using a 3.5-billion dataset with Instagram tags to help Facebook's software determine the content in the photo, whether it's beach (#beachlife!) Or a cat (#lolcats).

    Not only can Facebook tag and categorize your photos, it can also provide keywords to help describe images for the visually impaired and check for inappropriate content. Facebook even built a tool to help find the posture of a human being, which can become an effective method of judging the mood and behavior of users. This may sound a bit scary, but we are getting used to it.

    Proposals and classification

    Facebook proposes new friends to users, sure, but the proposals don't stop there. It also recommends timeline posts, news, events, groups, pages, products and more. Most of the content you see on your site is because the machine learning algorithm has decided that you want to see those content and prioritize them for you.

    Moderate content

    Moderate content

    Although these systems still have a lot of work to do, recent events have spurred Facebook to further promote the development of powerful content filtering systems that can identify fake news and hate. speech (speech is negative). Facebook will track links or documents that may spread false information and delete that information. Clearly, these algorithms have succeeded in finding and deleting more than 99% of terrorist propaganda content.

    Language

    Language

    Among other applications, Facebook is actually using AI to determine when someone posted suicidal thoughts on Facebook, contacting their friends and those who responded first, when needed. According to Facebook's report, this has begun to work, and it shows how powerful AI is, in an environment where it has access to human psychological data.

    Game is a great way to test AI. Drop them into an artificial situation, see how they work compared to other computers or people and see how they can really learn. Facebook has ELF OpenGo, similar to Google's Alpha Go Zero, as well as an extended ELF project, providing the foundation for AI game research. Facebook even developed an AI platform to help conduct research on AI playing StarCraft.

    Gaming

    Gaming

    Facebook's main AI site does not show you a series of flashy marketing materials about the future. However, there are a lot of pretty serious things happening, as you can deduce from the number of projects and groups. Facebook has developed tools such as PyTorch and (with Microsoft) ONNX, which are open source contributions to general AI research. Facebook is also involved in most other major AI-related companies in Partnership on AI, with the goal of using AI to benefit society and develop it responsibly.

    Research and development

    Research

    Your Facebook experience has definitely improved by AI, and it is likely that it will improve other parts of your life, as many of Facebook's research acts as open source and can be used. used by other researchers and developers. But Facebook doesn't tend to push this process too fast, because AI can go wrong.

    So what will Facebook do with all these powers?

    If you feel like you're in a science fiction movie, where robots check behavior, monitor your psychology and censor your interaction, you're not wrong. Targeted ads used predictions about you to sell you everything, but what if Facebook started using it to learn how to control your mood before displaying ads? Perhaps it would be a series of articles, color arrangements or subtle actions that stimulate hunger just before proposing pizza delivery? Or wait until you sympathize to promote a charity job? It may be a real and urgent concern than you think.

    See more:

    So will Facebook become Skynet?

    Đã có một công ty tên là Skynet, vì vậy Facebook sẽ phải mua lại chúng trước. Rồi sau đó có lẽ Facebook sẽ tiếp quản nhân loại, nhưng không phải với những con robot giết người, mà chỉ với những cú hích nhẹ nhàng. Có nhiều khả năng, chúng ta sẽ đón nhận những điều đáng kinh ngạc từ AI của Facebook, cũng như một số điều khiến chúng ta thậm chí phát điên hơn cả Cambridge Analytica. Ngay cả AI cũng không thể dự đoán được tương lai (hay ít nhất là chưa), vì vậy chúng ta chỉ có thể chờ đợi mà thôi.

    Xem thêm:

    No comments:

    Post a Comment