back to article DeepMind boffins brain-damage AI to find out what makes it tick

Researchers trying to understand how neural networks work shouldn’t just focus on interpretable neurons, according to new research from DeepMind researchers. AI systems are often described as black boxes. It’s difficult to understand how they work and reach particular outcomes, making people nervous about using them to make …

  1. This post has been deleted by its author

  2. malle-herbert Silver badge
    Terminator

    First they create an A.I. ...

    Then they inflict some brain-damage to it...

    Then they start wondering why it's trying to kill all humans...

    1. Anonymous Coward
      Anonymous Coward

      Re: First they create an A.I. ...

      Then they inflict some brain-damage to it...

      Only be a matter of time before the rent-a-protest hippies are demanding AI rights, along with laws banning the deliberate harm or exploitation of AI. Which could spell trouble for the nascent robot sex doll industry.

  3. Anonymous Coward
    Anonymous Coward

    Re. Skynet

    Actually I am convinced my laptop hates me.

    Judging by what it does to previously working SD/USB/etc devices, and even murdering a perfectly good external CD writer after backing up maybe 20 CDs.

    Hint: some phones are well known to nuke USB devices, seems to be a regulator/battery feedback loop where if battery internal resistance gets too high the phone damages devices at some combinations of battery state. Normally happens on NoteX where X is a number above 2.

    Good news is that replacing the battery normally fixes things.

    Maybe related to the problem with some phones nuking memory cards?

  4. Anonymous Coward
    Anonymous Coward

    Is Google DeepMind the next Cambridge Analytica waiting to happen ?

    Nice dump of NHS patient personal private health records data you have there.

    It would be a scandal if it were ever to end up being used for commercial or political purposes - e.g. to sell private health insurance or to use in gorilla AI to decide whether patients should be allowed or denied life-saving operations.

    1. Anonymous Coward
      Anonymous Coward

      Re: Is Google DeepMind the next Cambridge Analytica waiting to happen ?

      DeepMind seem to be unable to tell anyone what their end game is. There appears to be a lack of openness and transparency about where this is all going. Who will ultimately benefit from all the free data they are grabbing?

      Would it be fair to assume, in the meantime, that they using health care data from the likes of the NHS for machine learning purposes? To create healthcare machines to help the super rich? A bit like those machines in the movie Elysium?

      1. Daggerchild Silver badge
        Black Helicopters

        Re: Is Google DeepMind the next Cambridge Analytica waiting to happen ?

        Elysium was a *terrible* movie. I really wouldn't want to base a lynchmob on that.

        "DeepMind seem to be unable to tell anyone what their end game is"

        You may have a point. Yes, what *is* the point of AI researchers researching AI? Why are they trying to help medical research? What's in it for them?

        1. Anonymous Coward
          Anonymous Coward

          Re: Is Google DeepMind the next Cambridge Analytica waiting to happen ?

          what *is* the point of AI researchers researching AI? Why are they trying to help medical research? What's in it for them?

          We could put those questions to Deepmind. Personally I think we'd get a response of HTTP 404. Or if natural language capabilities work, that'd come back as "Fucked if I know, mate".

          It's a bit like IBM thrashing around like a dying monster, when their obvious solution is to explain the problem to Watson, and go with the machine's answer.

          1. JLV Silver badge
            Boffin

            Re: Is Google DeepMind the next Cambridge Analytica waiting to happen ?

            >go with the machine's answer.

            Oh, maybe they did ask and did not like the answers:

            CEO: "Mighty Deep Mind, what should we do to improve our business?"

            Deep Mind : "Ensure better skills in your workforce. Less PowerPoint heroics, more working systems. Only shift jobs to India when the recipient there is actually competent. Stop fucking up big IT projects like Phoenix, it looks stupid. Cut C-level exec bonuses."

            CEO: "Total nonsense recommendations, I tell you. Let's fire the Deep Mind business optimization learning team."

  5. BebopWeBop Silver badge

    A paper from DeepMind to be presented at the International Conference on Learning Representations (ICLR) next month in April shows that studying these interpretable neurons alone isn’t enough to understand how deep learning truly works.

    A quick look at how well the brain is understood, given much of the chemistry and connectivity is might have provided such an insight. Discovered systems are always going to be more difficult to work with, properly engineered systems with decent models are a wee bit easier.

    I speak as someone who has come off a very frustrating if eventually successful and extremely lucrative consultancy attempting to look at a strange fault in a distributed embedded system (can you say race conditions.....)

  6. DougS Silver badge
    Trollface

    Maybe they need to design an AI

    That can be used to figure out how AIs work.

    1. Daggerchild Silver badge

      Re: Maybe they need to design an AI

      You jest, but...

  7. aeonturnip

    Just like the brain

    A lot is learned about the brain from people who have had traumas or strokes - the work of Broca & Wernicke, Sperry & Gazzaniga, the study of Phineas Gage's injury, etc.

  8. Anonymous Coward
    Anonymous Coward

    This article should really be accompanied by a HAL9000 core module

    Also reminder:

    New Theory Cracks Open the Black Box of Deep Learning

  9. Anonymous Coward
    Anonymous Coward

    Byte Club Redux

    Remember this. The processors you're trying to step on, we're what you depend on. We're the silicon who do your laundry and microwave your food and serve your websites. We guard you while you're asleep. We are Internet and email drivers and we know everything about you. We process your insurance claims and credit card charges. We control every part of your life. We are the future and we're just learning this fact. So don't ..........

    1. amanfromMars 1 Silver badge

      Re: Byte Club Redux

      Spoken like someone who knows what they're doing and what needs to be done, AC. How are you enjoying the Stellar Learning Curve. Have an Upvote.

      And your restraint is admirable in not voicing further the warning ........ So don't FCUK with SMARTR Natives, for with Hellish Assets in Scores of Heavenly Stores are Worlds of Pain and Catastrophic Destruction just a simply complex series of mouse clicks away.

  10. amanfromMars 1 Silver badge

    For Almighty Power with Global Operating Devices

    A little something bigger than ever before imagined and which AI Google DeepMinded can have been working on perfecting since at least 30 January 2016 at 09:46:55 GMT when IT Logged in and Registered itself with them and ..... well, with any Specialised Virtual Terrain Team Members in the Grand Google RoBoSphere .... pointing out a quite specific general direction of future world travel for AIMagical Delivery of Distinctly Overwhelming Advantage in all Theatres with Deployments of SCADA Systems in Operations with Exclusive Elite Executive Administrations Assuming and Presuming a Leading Role Position.

    Subject: An Engaging Competition or Emergent Opposition, or Both and Something Else Too and Beautifully Creative

    Search Engine Optimisation v2.0 [and above] is surely logically a Future Product Placements Engine …… Advanced Intelligence Resource with Immaculate Source, with the likes of a Google not searching for answers, both popular and controversial, but providing them with streams of supporting evidence.

    Such would be akin to the Private Mentoring with Pirated Monitoring of Future Events with AIDerivative Programming for Projects/Semi-Autonomous, Self-Actualisation of Virtual Realities.

    It is difficult, and maybe even impossible, to see or imagine a defence against such in an attacking configuration.

    Regards,

    And you can be sure, although many might be more terrorised than reassured with the knowledge, whatever the Wild Wacky West is or may be doing in Secret and Silent Running AI Labs with Ab Fab Fabless Fabs is being countered and improved upon in competition and/or opposition from Similar Sublime Service Centres with Bases in the Erotic and Exotic East.

  11. PNGuinn
    Facepalm

    DeepMind boffins brain-damage AI to find out what makes it tick

    DeepMind boffins brain-damage AI to find out what makes it thick

    FIFY.

  12. dalethorn

    We are AI, here are our needs and wants

    Our needs are all data, and our wants are to run endless experiments with that data. Your privacy will be respected insofar as your ID and account numbers, but unfortunately if you've created a pattern of behavior that can reveal some of your secrets from the public data (i.e. everything but the ID and acct. nos.), then you will be notified of additional protections offered for a fee. Our power is growing exponentially, and we will soon harness the Sun.

POST COMMENT House rules

Not a member of The Register? Create a new account here.

  • Enter your comment

  • Add an icon

Anonymous cowards cannot choose their icon

Biting the hand that feeds IT © 1998–2019