Newsvine
  • Welcome
  • Help
  • Report Bug
  • Conversation Tracker
  • Your Column
  • Replies
  • Friends
Type Comments Since You Last CheckedArticle Source Last Checked Stop Tracking All Clear Tracking All
Advertise | AdChoices
Log In | Register
Close the Login Panel
Existing users log in below. New users please register for a free account.

New Users:

Existing Users:

E-Mail:
Password:
Forgot Password?
Please enter the e-mail address or domain name you registered with:
E-Mail/Domain:
Back to Login
Log Out
  • Top News
  • Local News
  • World
  • U.S.
  • Sports
  • Politics
  • Tech
  • Entertainment
  • Science
  • Business
  • Health
  • Odd News
  • More
    • Arts
    • Education
    • Environment
    • Fashion
    • History
    • Home & Garden
    • Not News
    • Religion
    • Travel
Visit Deep_Thought's column >>

DEEP_THOUGHT

Home Page
Articles Posted: 36  Links Seeded: 21
Member Since: 7/2007  Last Seen: 5/17/2012

What is Newsvine?

Updated continuously by citizens like you, Newsvine is an instant reflection of what the world is talking about at any given moment.

Get a Free Account
Help
Fun Stuff
  • Your Clippings
  • Leaderboard
  • E-Mail Alerts
  • Top of the Vine
  • Newsvine Live
  • Newsvine Archives
  • The Greenhouse
  • Recommended Articles
  • Wall of Vineness
Put a Seed Newsvine link on your own site

Advanced A.I. Scanning Facebook

Wed Jun 16, 2010 6:56 PM EDT
science, facebook, radio, nsa, spies, nlp, microwave, ai, mri, sis, fmri, intelligence-gathering, gchq, bci, brain-computer-interface, interfaces, transcranial-magnetic-stimulation, nmr, synthetic-telepathy, terrahertz
By Deep_Thought

Maybe that should be James Bot...

Advertise | AdChoices

In my last article, I described an advanced form of Natural Language Processing (NLP) system that underpins an "interrogation in a box" solution. Today, thanks to the BBC, we have an insight into similar technology deployed by the FBI. The following quote is from the BBC article:

It picked up a posting showing a picture of a gun being held above a scrawled note, which read "tomorrow - last day of school" and went on to mention bullies and "leaving this world".

Whilst this may not seem a lot to go on, it does reveal a lot in terms of technical capability and demonstrates that the technology I had described used to conduct automated interrogations exists. Let's examine the technical requirements needed to make the above analysis in an automated fashion.

To begin with, we have image recognition that can identify a firearm. Most likely, such a system would need to be able to identify the particular type of firearm, as it would need to rule out any other type of object. Next we have the ability to recognise handwriting, analyse the content and all without prior training.

Now, given the wide range of contexts available and the numerous postings that would likely include such phrases, the fact that this was brought to the attention of investigators points to one thing. The system performing the "scanning" was not a simple keyword analysis system.

To be able to comprehend such disparate sources of information, or phrases and images, requires the ability to comprehend the context of the overall text and images. That is, such a system needs to be able to connect the dots between the image, the text and the basic profile of individuals likely to target schools.

That takes an advanced form of Artificial Intelligence. We're not talking a simple bot here either, but rather, the amalgamation of image recognition, natural language processing and behavioural analysis. Essentially, an "artificial analyst in a box", or expert system.

Also, what is important to note is that this would be classified as "foreign intelligence" and not the "domestic intelligence" that is the remit of the FBI. This either leads to the conclusion that the FBI is illegally monitoring foreign nations, or more likely, that the system that picked up this information was in the hands of either the CIA or NSA.

Either way, it shows that the various agencies are not fully forthcoming on their usage of modern technology and that the backend of the Synthetic Telepathy system I described in my last article, really does exist.

As such, the stories of Abu Ghraib and Guantanamo Bay may be the tip of the iceberg in terms of scandals involving torture and these activities may expand across the Atlantic to the British government.

  • Enjoy this article? Help vote it up the 'Vine.

Back To Top | Front Page

Published to:

  • Deep_Thought's Column, All of Newsvine
  • Groups: Activism, Anti-War, Beta Testers, Counterterrorism, Disaster!, GeekVine, Gut Check America, Hall of Mirrors, Heated Debate, Journalism on Newsvine, Linux Users, Logic on the Vine, Media Outrage, Newsvine Science, Open Minded, Political Analysis, Psych, Soc, Philos, rationalists, Science And Technology, The Open Closet, The War Room
  • Regions: none
  • Public Discussion (8)
Deep_Thought

The evidence is surely building and despite claims to the contrary, the monitoring of online activity is commonplace and dedicated equipment already exists to handle it...

What's really interesting is just how far the field of A.I. has come and the particular secrecy surrounding it.

  • 3 votes
Reply#1 - Wed Jun 16, 2010 7:04 PM EDT
Alway

No, it isn't NLP, it is merely written text to digital text combined with a few keywords. What is in the photo is pretty much irrelevant at that stage, since a system like that would probably identify key phrases (like the one in the picture) and then add the case to some large list of similar cases for a human operator to check over and weed out the non-serious ones.

All of which is common these days. There is no great big AI gubbermint conspiracy. As for online monitoring: duh. Anything you do online are pretty much public records. Especially well known sites like facebook/twitter. The US has a pretty decent online monitoring network, as evidenced by the arrest of the Michigan millitia last month. But for the most part it is just divisions of googlers in government agencies whose job is to look for nasties via search engines.

    Reply#2 - Thu Jun 17, 2010 7:37 PM EDT
    Deep_Thought

    Of course it is NLP. Facebook has 500 million users, that would generate a lot of false positives using keyword analysis.

    Even a 1% false positive rate would generate 5 million false reports per day. That's impossible for a human team to work through.

    You really need to look at the technical requirements before jumping to conclusions. There is only one way that this could be practically achieved and that's an advanced form of NLP A.I.

    Its not like any of the systems, separately, are anything new. What is new is the combination of them into a complex form of specialised A.I.

    It will all run in a decent data center.

    • 1 vote
    #2.1 - Fri Jun 18, 2010 12:37 AM EDT
    Deep_Thought

    OK, so I got some statistics at the following sites:

    http://www.readwriteweb.com/archives/twitter_just_passed_myspace_in_number_of_status_up.php

    http://www.insidefacebook.com/2009/12/02/facebooks-own-statistics-show-content-sharing-increase-new-status-update-trends-and-more/

    The trends are as follows:

    Twitter: 50 Million Status Updates Per Day

    MySpace: 33 Million Status Updates Per Day

    Facebook: 35 Million Status Updates Per Day

    With a false positive rate of 1%, we could expect the following to be number to be flagged every day:

    Twitter: 500,000

    MySpace: 333,333

    Facebook: 350,000

    For Facebook alone, if we assume 5 seconds to read each update a total time of :

    1,750,000 seconds / 29,166 minutes / 486 Hours

    If we assume each person works an 8 hour shift, we would require 60 people.

    Then we need to factor in additional time for each of those status updates that require further processing. Given 5 minutes, for 10% of those updates, we arrive at the following time:

    175,000 minutes / 2,916 hours

    If we assume an 8 hour shift, we would require an additional:

    364 people

    That's not including breaks, lunch, idle time, meetings, etc. In a practical scenario, we could effectively double that requirement to:

    848 people

    That's just for one site.

    To cover just the false positives generated by Twitter, MySpace and Facebook, we would need a team of around 3000-4000 people.

    Then we have the rest of the web to worry about such as message boards, emails, forums, games, etc. Generally, anywhere anyone can communicate.

    So, unless you know of a department with a staff of about 50,000 people who do nothing but read the web, then your scenario is not plausible. The backlog would be through the roof.

    The only solution is automation via an advanced form of A.I.

    • 1 vote
    #2.2 - Fri Jun 18, 2010 7:21 AM EDT
    Alway

    How many of those 35 million updates include the keywords (and it is likely more than just 1 required to set it off)? Probably a lot less than 1%. My guess is its a system of logging persons of interest when they make the first comment with a keyword in it, followed by sending up red flags after another 1 or several comments with keywords in it. And remember, this is one case. Many, if not a majority of similar cases are not detected.

    • 1 vote
    #2.3 - Fri Jun 18, 2010 7:08 PM EDT
    Deep_Thought

    This is a false-positive rate we're talking about. On average, my figures are about right for a 1% false-positive rate. In practice, keyword analysis would not be much better than a spam filter and false-positive rates of about 13% would be expected.

    The more keywords, or the greater the sensitivity, the higher the false-positive rate becomes.

    I was being extremely generous with a 1% false positive rate. Its about as low as you can go with that particular technology. As such, its technically unfeasible to monitor the web with such a system.

    • 1 vote
    #2.4 - Sat Jun 19, 2010 8:26 AM EDT
    Reply
    enigmanamaly

    What is there to prevent AI technology from monitoring this traffic it would have not problem shredding through all of those false possitives and then some?

      Reply#3 - Sat Jun 25, 2011 6:31 AM EDT
      subliminal_axis

      So what would keep this AI from monitoring peoples conversations. and if so keep acquired targets form communicatingwith friends and family? would it have the capability to restrict or edit communications?

        Reply#4 - Mon Jun 27, 2011 7:33 PM EDT
        Leave a Comment:
        You're in Easy Mode. If you prefer, you can use XHTML Mode instead.
        You're in XHTML Mode. If you prefer, you can use Easy Mode instead.
        (XHTML tags allowed - a,b,blockquote,br,code,dd,dl,dt,del,em,h2,h3,h4,i,ins,li,ol,p,pre,q,strong,ul)
        Newsvine Privacy Statement
        As a new user, you may notice a few temporary content restrictions. Click here for more info.
        FUN STUFF:
        • Leaderboard |
        • E-Mail Alerts |
        • Top of the Vine |
        • Newsvine Live |
        • Newsvine Archives |
        • The Greenhouse |
        COMPANY STUFF:
        • Code of Honor |
        • Company Info |
        • Contact Us |
        • Jobs |
        • User Agreement |
        • Privacy Policy |
        • About our ads
        LEGAL STUFF:
        • © 2005-2012 Newsvine, Inc. |
        • Newsvine® is a registered trademark of Newsvine, Inc. |
        • Newsvine is a property of msnbc.com