Slashdot Mirror

← Back to Stories (view on slashdot.org)

Facebook's AI Unlocks the Ability To Search Photos By What's in Them (techcrunch.com)

Posted by msmash on from the sounds-creepy dept.

An anonymous reader shares a TechCrunch report: Initially used to improve the experience for visually impaired members of the Facebook community, the company's Lumos computer vision platform is now powering image content search for all users. This means you can now search for images on Facebook with keywords that describe the contents of a photo, rather than being limited by tags and captions. To accomplish the task, Facebook trained an ever-fashionable deep neural network on tens of millions of photos. Facebook's fortunate in this respect because its platform is already host to billions of captioned images. The model essentially matches search descriptors to features pulled from photos with some degree of probability.

62 comments

  1. Still Not AI by Luthair · · Score: 2, Insightful

    Sorry, but this is still not AI.

    1. Re:Still Not AI by rainmouse · · Score: 1

      That really depends on how you define AI. If you want to define AI as purely something that can hold a realistic conversation with you then no, it's not. However if you agree that a machine that demonstrating the ability to learn from examples then I'm sure many would argue this is a form of AI, but that's why why careful people describe it as machine learning instead. Just like intelligent search was often called AI in the past by the press.

    2. Re:Still Not AI by Maritz · · Score: 1

      It's not AI until it's Skynet? Is it a true scotsman, maybe?

      --
      I do not want your cheap brainburning drugs. They are useless for work. And I am a working man today.
    3. Re:Still Not AI by AmiMoJo · · Score: 4, Funny

      Probably a good thing it's not too intelligent. Can you imagine if google was a person, the kind of sanity destroying crap it would have to deal with? Facebook's poor AI will be bombarded by dubious requests.

      --
      const int one = 65536; (Silvermoon, Texture.cs)
      SJW, n: "Someone I don't like, and by the way I'm a fuckwit" - AC
    4. Re:Still Not AI by Luthair · · Score: 2

      Just because a small number of PR people in silicon valley are misusing a term doesn't mean the rest of the world has to accept it.

    5. Re:Still Not AI by lucaiaco · · Score: 0, Troll

      Have you ever taken AI class in college? What do you think they teach you, how to develop conscious robots?

    6. Re:Still Not AI by lucaiaco · · Score: 0

      The scenario described in Her is also particularly convincing http://www.imdb.com/title/tt17... (the plot may sound stupid, but the movie is actually nice)

    7. Re:Still Not AI by spire3661 · · Score: 2

      Appeal to Authority fallacy. Colleges are not the final word on AI. College taught me a bunch of shit that was completely useless and behind the times too.

      --
      Good-bye
    8. Re:Still Not AI by 0100010001010011 · · Score: 0

      It's not worth arguing with them. 10 or 20 years ago being able to classify an image would have been AI to them.

      They'll be moving the goal posts long after NOT AN AI driving cars are talking them to their doctors appointments where a NOT AN AI does diagnostics on them.

      Their brains will be in a vat living on mechanically and they'll argue that it's not *true* AI.

    9. Re:Still Not AI by sexconker · · Score: 2

      Have you ever taken AI class in college? What do you think they teach you, how to develop conscious robots?

      I have. Machine Learning and AI were separate courses. In the Machine Learning course you get a lot of practical shit and some theory.

      In the AI course you get a lot of theory and some practical shit. Terms like "neural net" and "deep learning" are bandied about and there's talk about current research working to simulate a brain to some degree or use organic/fuzzy computing of some sort to mimic the brain's functions. On the practical side, it's all just the same "machine learning" shit from the previous course, but with clever feedback and bootstrapping mechanisms.

      I'm not knocking either course (they were pretty good and the same professor ran both courses). But the state of AI is no where near creating something that could pass as intelligence, even if some pieces can surpass human ability in certain tasks or mimic humans decently in a chat program after data mining billions of lines of conversation.

    10. Re:Still Not AI by Tablizer · · Score: 3, Insightful

      There is NO formal agreed-upon definition of AI and experts, philosophers, and pundits argue forever and ever over it. Givvittup.

      --
      Table-ized A.I.
    11. Re:Still Not AI by Anonymous Coward · · Score: 0

      Lick my balls, kid. I have been in CS for nearly four decades. Twenty years ago, it would have been a REALLY cool trick to do this sort of thing on a small scale on supercomputers. And we still would never have called it AI. I think that you should call us astroturfers now for good measure.

    12. Re:Still Not AI by Luthair · · Score: 1

      I did actually.

    13. Re:Still Not AI by Luthair · · Score: 1

      Until the computer makes a decision outside of its programming its just an algorithm, no matter how fancy.

    14. Re:Still Not AI by religionofpeas · · Score: 2

      An important part of intelligence is the ability to recognize patterns. That's what the computers are doing.

    15. Re: Still Not AI by Anonymous Coward · · Score: 1

      30% of the human brain is dedicated to the human visual system. Each eyeball has 100 million rods and cones and does basic data compression using contrast measurement of each point and its surroundings (red-hreen, blue-yellow, black-white). Add some more neurons to detect lines and points in varying directions and you have enough information to do shape recognition, motion tracking, depth perception, automatic face, text, font, object and location recognition. Get a computer to do all those and you have biometric face recognition, automatic machine translation and robot path planning and obstacle avoidance.

    16. Re:Still Not AI by Anonymous Coward · · Score: 0

      Your refutation of "AI includes Weak AI in common usage" was essentially "Only tech PR calls it that." So luaiaco says colleges call it that too. And you reply that colleges don't get to say what AI is.

      Well in addition to the companies who are actually developing this stuff, and the universities who are actually researching it, who do you think gets to make that call? Also it's in common usage in popular culture with this meaning and has been for decades. How much evidence will it take to convince you?

      The answer is that they're not wrong for using AI to describe Weak AI. You're wrong because you say Strong AI is the only AI.

    17. Re:Still Not AI by religionofpeas · · Score: 1

      Until a human makes a decision outside of its electro-chemical reactions it's just an algorithm too.

    18. Re: Still Not AI by hackwrench · · Score: 1

      Of course I have yet to see a rigorous definition of what intelligence actually is. Examples of it lead me to believe that one could argue that anything that calculates is intelligent.

    19. Re:Still Not AI by e3m4n · · Score: 3, Insightful

      Probably a good thing it's not too intelligent. Can you imagine if google was a person, the kind of sanity destroying crap it would have to deal with? Facebook's poor AI will be bombarded by dubious requests.

      My opinion of the average Facebook user is about as high as my opinion of people on AOL back in the 90s. If it were a full fledged AI there is no doubt that the assumption it drew from its average sampling of humans would definitely make it go SkyNet on all of us.

    20. Re:Still Not AI by ThePyro · · Score: 1

      Can you imagine if google was a person, the kind of sanity destroying crap it would have to deal with?

      Yes, and it's pretty hilarious.

      If Google was a guy

    21. Re:Still Not AI by Anonymous Coward · · Score: 0

      Until a human makes a decision outside of its electro-chemical reactions it's just an algorithm too.

      Wrong. Just wrong [or not even wrong].

    22. Re:Still Not AI by Anonymous Coward · · Score: 0

      There is NO formal agreed-upon definition of AI and experts, philosophers, and pundits argue forever and ever over it. Givvittup.

      Regardless, one thing is for sure: AI won't come from computer sciences. It will most probably be born in a physics lab. Expect the unexpected.

    23. Re:Still Not AI by Anonymous Coward · · Score: 0

      thus spoke the alt human

    24. Re:Still Not AI by zlives · · Score: 1

      there is technical AI and there is marketing AI
      technical AI cannot keep up with the marketing AI.

    25. Re:Still Not AI by Anonymous Coward · · Score: 0

      The concept of AI isn't something that changes over time. It has always meant sentience.

  2. 1K by Maritz · · Score: 1

    Simply type your 1000 words and Bob's your mother's brother.

    --
    I do not want your cheap brainburning drugs. They are useless for work. And I am a working man today.
  3. It would be even easier if ... by BarbaraHudson · · Score: 1

    If would be even easier if Facebook didn't strip off any embedded info you add to images (such as your name) so that they can sell them to others. Time to watermark the sh*t out of everything.

    --
    "Transparent" is a shit show that trades on every stereotype going. A man in drag is NOT a transsexual.
    1. Re:It would be even easier if ... by Anonymous Coward · · Score: 0

      So sorry you're having trouble making a living selling copies of porn. Have you tried performing live?

    2. Re:It would be even easier if ... by BarbaraHudson · · Score: 1

      If everyone watermarked their pictures, it would make it a bit harder for scammers to just grab anyone's pic and claim it's theirs. There should be an automatic watermark plug-in so you can mark all photos before uploading.

      --
      "Transparent" is a shit show that trades on every stereotype going. A man in drag is NOT a transsexual.
    3. Re:It would be even easier if ... by SuricouRaven · · Score: 1

      It's actually a perfectly reasonable action - bizarely, for the sake of privacy. The embedded information in images can include when and where it was taken, which is information that could be used against a person.

      Facebook want to spy on you. But it is not in their interests to aid other organisations or individuals in spying on you. Not without paying Facebook for the information, anyway.

    4. Re: It would be even easier if ... by Anonymous Coward · · Score: 0

      Once scammers knew how a watermark was applied they know how to remove it. The only way to watermark something is to have a unique method for each user that is unobtrusive and relevant to each scene. Eg graffiti on a street scene.
      Objects can be airbrushed out. Low order bits can be filtered out. Perhaps a checksum, upload time and filename could be registered and logged.

    5. Re:It would be even easier if ... by Anonymous Coward · · Score: 0

      If everyone watermarked their pictures, it would make it a bit harder for scammers to just grab anyone's pic and claim it's theirs. There should be an automatic watermark plug-in so you can mark all photos before uploading.

      Why don't you build one?

  4. two girls and a cup? by Anonymous Coward · · Score: 1

    gets a hit?

    mmmm, brazile nuts and chocalate.

  5. Facebook catching up to old Google features by Anonymous Coward · · Score: 1

    Google Photos has quietly done this for years. It's documented on their help pages:

      https://support.google.com/photos/answer/6128838?co=GENIE.Platform%3DAndroid&hl=en

    The example search is for "dogs," but it can find many more things. It's distractingly documented next to a face/person-tagging feature, but it really does do Zork-like search queries for recognized objects, or relationships among objects, within your photos.

    It's too bad Google demonstrated they couldn't be trusted to run a social network without pushing things on people. I feel Facebook is almost as bad, but since they are less powerful, so far, people probably made the right decision not to move to G+. But it's too bad. Google has a lot of unused capability, and major discoverability problems. This has always been true.

    1. Re:Facebook catching up to old Google features by arth1 · · Score: 1

      Google Photos has quietly done this for years.

      Not the same thing. Google Photos only lets you search your own images for machine guessed labels, not the images of everyone.

    2. Re:Facebook catching up to old Google features by Anonymous Coward · · Score: 0

      Not the same thing. Google Photos only lets you search your own images for machine guessed labels, not the images of everyone.

      Their other product, "Google Search", lets you search the images of everyone. HTH. HAND.

      It seems the only thing Facebook has contributed to this problem is granting access to their walled garden full of images. Am I wrong? Is there a minor contribution over what Google has already:

        done,
        and open sourced (the model, untrained): https://github.com/tensorflow/models/tree/master/im2txt
        and sold (trained model) on cloud: https://cloud.google.com/vision/

      but a minor contribution which has nothing to do with AI? or is it really no contribution at all, just pure necktie damage?

      Either way, I'm just saying, blowhard press releases should cost you credibility, not gain it. At least with that much I expect we can agree.

    3. Re:Facebook catching up to old Google features by arth1 · · Score: 1

      Their other product, "Google Search", lets you search the images of everyone.

      Yes, but those do not, as far as I know, have machine generated tags based on image analysis, only tags based on the filename, alt text and other text in pages that link to it. I.e. it won't find a dog picture if it's labeled "cat", and nothing in the text on any pages references it uses the word "dog" or synonyms.

    4. Re:Facebook catching up to old Google features by religionofpeas · · Score: 1

      Google Photos can sort my photos by category. There's Sky, Lakes, Cars, Screenshots, Flowers, Bridges, Forests, Beaches, Hiking, Birds, Towers, Fog, Trains, Cliffs, Sunsets, Boats, Selfies... And I didn't name or sort any of them myself.

    5. Re:Facebook catching up to old Google features by religionofpeas · · Score: 1

      Never mind... read too quickly.

  6. Hot chicks by Anonymous Coward · · Score: 0

    Hot chicks.

    1. Re:Hot chicks by FFOMelchior · · Score: 1

      https://goo.gl/images/MafWo1

    2. Re:Hot chicks by Tablizer · · Score: 1

      http://www.thepoultrysite.com/...

      --
      Table-ized A.I.
  7. Scotsmen are... intelligent. :) by fyngyrz · · Score: 1

    It's not AI unless it's (A)rtificial and (I)ntelligent. You'll know when it's intelligent. Because it'll exhibit actual characteristics of intelligence. Not the ability to filter images based on content.

    But go ahead, keep calling your toaster and your thermostat intelligent. It's very amusing. Despite the fact that it's not very... intelligent.

    --
    I've fallen off your lawn, and I can't get up.
    1. Re:Scotsmen are... intelligent. :) by religionofpeas · · Score: 1

      The other day, my toaster called me intelligent.

    2. Re:Scotsmen are... intelligent. :) by fyngyrz · · Score: 1

      True story:

      A few years back, sometime overnight, a new cat I had rescued peed in my toaster.

      In the morning, dull witted and bleary eyed, I threw a couple slices of rye bread in there, and pushed down the lever. The heating elements immediately vaporized the cat urine, while altering it chemically into something that smelled so bad I could not possibly adequately describe it.

      As I reeled out of the room coughing and gagging, I actually thought I heard that toaster laughing at me.

      I think that damed burst of vapor was actually hallucinogenic.

      I kept the cat anyway. The toaster, however, I disposed of. With prejudice.

      --
      I've fallen off your lawn, and I can't get up.
    3. Re:Scotsmen are... intelligent. :) by Tablizer · · Score: 1

      The other day, my toaster called me intelligent.

      I installed Kissup 2.0 on it. It's lying to you. Nor does it make very good toast.

      --
      Table-ized A.I.
  8. Yeah, but the data it's given to work with by hackwrench · · Score: 1

    I'm already seeing people use the tag people in the photo tool to tag people who the picture merey hint at a trait in. If a picture reads "Earth girls are pretty" somebody will tag it with the profiles of some girls they think are pretty.

  9. Re: Appeal to authority by hackwrench · · Score: 1

    Luthor made a mix of an appeal to authority and no true Scotsman by identifying a group that he determined were not an authority and since they weren't an authority their use of the term was misuse.

  10. Re: agreed upon formal definitions by hackwrench · · Score: 2

    Nor for that matter is there one for intelligence.

  11. like... Google Photos? by ooloorie · · Score: 1

    Seriously, this is news? Google and others have had this for years.

  12. new improved twist-top spinach can by epine · · Score: 1

    Intelligence, by any definition, is the ability to spot the word "unlock" in the story summary while ignoring the word "AI".

    Because we don't have "unlocks" every day, but we do have this tiresome "what the fuck is AI, anyway?" shit show more often than Popeye eats E. coli tainted spinach.

    New improved can, same old dubious irrigation method.

  13. Does it stand a chance against the Ostrichinator? by ffkom · · Score: 1

    After all, deep-learning NNs have been demonstrated to be fooled very easily by this brilliant little tool.

  14. It's bitztream by Anonymous Coward · · Score: 0

    The autism-hating, custom EpiPen-hating, Musk-hating Slashdot troll!

  15. Re:Biased reporting by Applehu+Akbar · · Score: 1

    This site is so pro-Trump it isn't even funny. You need some real editors that are impartial and don't have an agenda to push.

    Gee, that must by why Slashdot has this one editor who relentlessly promotes coal. Whenever a story mentions coal in any way, he posts it.

  16. So, by BitztreamNotARealNam · · Score: 1

    How's life in the hypocrite lane?

  17. CognitionX by Matthew3015 · · Score: 1

    Thought you might want to check out the daily/weekly newsletter from CognitionX with over 4K subscribers. We covered this story and many more. We provide you with the latest and greatest news related to AI. Subscribe to stay up-to-date: http://cognitionx.com/news-bri...