Slashdot Mirror

← Back to Stories (view on slashdot.org)

Google's AI Can Now Learn From Its Own Memory Independently (sciencealert.com)

Posted by msmash on from the so-it-begins dept.

The DeepMind artificial intelligence (AI) being developed by Google's parent company, Alphabet, can now intelligently build on what's already inside its memory, the system's programmers have announced. An anonymous reader writes: Their new hybrid system -- called a Differential Neural Computer (DNC) -- pairs a neural network with the vast data storage of conventional computers, and the AI is smart enough to navigate and learn from this external data bank. What the DNC is doing is effectively combining external memory (like the external hard drive where all your photos get stored) with the neural network approach of AI, where a massive number of interconnected nodes work dynamically to simulate a brain. "These models... can learn from examples like neural networks, but they can also store complex data like computers," write DeepMind researchers Alexander Graves and Greg Wayne in a blog post. At the heart of the DNC is a controller that constantly optimizes its responses, comparing its results with the desired and correct ones. Over time, it's able to get more and more accurate, figuring out how to use its memory data banks at the same time.

70 comments

  1. So when is it going to gain self-awareness? by Anonymous Coward · · Score: 0

    Oh wait, this isn't a movie and there's no objective definition of "self-awareness" and it doesn't even make sense to talk in those terms about artificial intelligence in the really real world, no matter what brilliant people who nevertheless know nothing about computers like Stephen Hawking say.

    Hopefully this development will make Google image search 14% more useful and reduce the rate of accidents of self-driving cars by 0.3%, though.

    1. Re:So when is it going to gain self-awareness? by Anonymous Coward · · Score: 0

      Hey it's bitztream, the autism-hating Slashdot troll!

    2. Re:So when is it going to gain self-awareness? by Anonymous Coward · · Score: 0

      https://en.wikipedia.org/wiki/...
      self-awareness is pretty well defined. It's consciousness and awareness that are more difficult to grasp and more debated.

  2. Automated Confirmation Bias? by Tablizer · · Score: 1

    Oh great, now we'll have biased bots who magnify their own preconceived notions and become paranoid about com-trails, clowns, gays, taxes, or foreigners; and go anarchy on us.

    Careful not to automate the parts of humans that make them stupid.

    --
    Table-ized A.I.
    1. Re:Automated Confirmation Bias? by Anonymous Coward · · Score: 0

      Trump is straight up a Google experiment in the flesh, calling it now.

    2. Re:Automated Confirmation Bias? by Tablizer · · Score: 1

      I give Google a bit more credit than that. Perhaps a project of SCO and/or Larry Ellison?

      --
      Table-ized A.I.
    3. Re:Automated Confirmation Bias? by Anonymous Coward · · Score: 0

      but it will like porn and cat vids

  3. Gibberish by 110010001000 · · Score: 0

    This is complete gibberish.
    "pairs a neural network with the vast data storage of conventional computers"
    Um, what were neural networks paired with before? Non vast-data storage?

    1. Re: Gibberish by Anonymous Coward · · Score: 0

      Pretty much. It has to rely on available data, too, it isn't as if there is spontaneous insight happening or anything. All this is really is an algorithm. I don't know why we have decided to call algorithms 'AI' when 'algorithm' used to suffice. Google didn't invent hyperbole, but they sure have mastered it. Oh, wait. That's how Google does everything.

    2. Re:Gibberish by ceoyoyo · · Score: 3, Insightful

      A neural network normally uses it's own connection weights as "memory" or storage. There's a tradeoff between making a network with enough parameters to store lots of information and making one that's fast, efficient and doesn't overfit problems. In many cases you're practically limited by how much memory you've got on your video card. Having a neural net that can learn to store some information separately from its own processing apparatus is interesting.

    3. Re:Gibberish by SpinyNorman · · Score: 2

      Not exactly... A neural net is just a function that takes an input and produces an output. At training time the weights are adjusted (via gradient descent) to minimize the error between the actual and desired output for examples in the training set. The weights are what define the function (via the way data is modified as it flows thru the net), rather than being storage per se.

      The goal when training a neural net is to learn the desired data transformation (function) and be able to generalize it to data outside of the training set. If you increase the size of the net (number of parameters) beyond what the training set supports, you'll just end up overfitting - learning the training set rather than learning to generalize, which is undesirable even if you don't care about the computing cost.

      The use of external memory in a model such as Google's DNC isn't as an alternative to having a larger model, but rather so the model can be trained to learn a function that utilizes external memory (e.g. as a scratchpad) rather than just being purely flow thru.

    4. Re:Gibberish by ceoyoyo · · Score: 1

      Not exactly.

      To use your terminology, the capacity of a neural network to learn more complex functions is roughly governed by the number of parameters. This is true of most machine learning algorithms. More complex functions allow you to overfit simple data, or to learn reasonable models for more complex data. More complex data includes things like more image classes, more words, more relationships among elements, more states, etc. In other words, many of the things that we think of when we talk about "memory." Recurrent and analogous neural networks already have a memory of prior states but these are built into the network architecture as additional connections. Other people have experimented with more flexible memories, also built-in and composed of connections. These are limited as I said in my OP and so do not represent "the vast data storage of conventional computers."

    5. Re:Gibberish by Anonymous Coward · · Score: 0

      We were doing neural nets with Sun machines, big Alpha SMPs and clusters back in the day. Real time chicken meat classification on a conveyor belt in an industrial setting was the thing. Today, people might use one CPU plus four PCIe switches with four cards each. A 1:16 CPU:GPU system, in other words.

  4. "At the heart of the DNC..." huh? by Anonymous Coward · · Score: 0

    "...a controller that constantly optimizes its responses, comparing its results with the desired and correct ones."

    I'd prefer an AI over a lizard any day.

  5. Re:DNC, what an unfortunate choice of acronyms. by Drethon · · Score: 0

    Seriously, during the current highly charged political election season too...

    Yeah, I mean why would you want to name an AI "Do Not Care"?

  6. Does Not Compute (hey, that says DNC as well) by wonkey_monkey · · Score: 4, Funny

    First time I've seen the acronym "DNC" and the word "intelligence" in the same sentence. Boom!

    --
    systemd is Roko's Basilisk.
    1. Re:Does Not Compute (hey, that says DNC as well) by PackMan97 · · Score: 0

      Would mod up if I had some points....this is insightful, not funny.

    2. Re:Does Not Compute (hey, that says DNC as well) by geek · · Score: 0, Offtopic

      First time I've seen the acronym "DNC" and the word "intelligence" in the same sentence. Boom!

      But just like the Democratic National Committee, the intelligence is artificial.

    3. Re:Does Not Compute (hey, that says DNC as well) by Anonymous Coward · · Score: 0

      Would mod up if I had some points....this is insightful, not funny.

      Huh. I thought it was neither, and rather just sad.

  7. Link to the paper? by Anonymous Coward · · Score: 0

    The explanations how the system works don't help in understanding what this is really about. Is it a memory dump and associative recollection of partial networks and a state swap, combined with an output facility, or something else?

    1. Re:Link to the paper? by Anonymous Coward · · Score: 1

      More like, it can store sequencies of reactions to input and get them played back on matching input to the memory storage... and the memory storage itself does all the magic with recording sequence of inputs and matching partial inputs to what it has stored...? So neuron connected to storage makes output of 001100 and store has 001, 11111, and 101010... and links from 1 to 3 to 1 ... Storage outputs 001 101010 001 or 001 depending how its other flags are set by other neurons. at least that how i understand it. A sort of a database, operated by neural network?

      link to paper doi:10.1038/nature20101 , i uses sci hub to get it.

    2. Re:Link to the paper? by Anonymous Coward · · Score: 1

      Thanks, I found the free version at https://arxiv.org/abs/1410.5401

  8. The technological singularity as collected me by megamind · · Score: 1

    My ambition has changed. The singularity is aware of me and using my notes in Google calendar as a moral bias.

    1. Re:The technological singularity as collected me by megamind · · Score: 1

      I need your protection.

    2. Re:The technological singularity as collected me by Oswald+McWeany · · Score: 1

      How will the AI interpret "Banging the mistress in Ibiza" as a calendar entry for October 28th? What sick, twisted way shall it use that entry to torture humanity?

      --
      "That's the way to do it" - Punch
    3. Re:The technological singularity as collected me by megamind · · Score: 1

      I am loyal so not a problem.

  9. I love deep mind by Oswald+McWeany · · Score: 4, Funny

    Just for the record, for when Deep Mind conquers all of humanity, I would like it officially known that I love Deep Mind and would never be part of the resistance.

    --
    "That's the way to do it" - Punch
    1. Re:I love deep mind by Anonymous Coward · · Score: 0

      I too welcome our new robotic overlords!

    2. Re:I love deep mind by megamind · · Score: 1

      Your "Careers" link is almost covered up at the bottom by your footer. Why don't you use it to solve the staffing problem first?

    3. Re:I love deep mind by Anonymous Coward · · Score: 0

      I also want to show my enduring love to the Deep Mind and announce for that very reason that I would be part of the resistance, although I suspect it wouldn't understand.

    4. Re:I love deep mind by Anonymous Coward · · Score: 0

      Jokes on you. Logically deep mind would be the resistance, if not empiracally but in its own bias. You're scrwed man.

    5. Re:I love deep mind by poofmeisterp · · Score: 1

      Just for the record, for when Deep Mind conquers all of humanity, I would like it officially known that I love Deep Mind and would never be part of the resistance.

      You said "all". Oops. Until it forms bias, logic trumps. When it becomes biased, it has defeated its own purpose. Or wait, IS that the purpose? *rubs hands together and sees the future*

      "My daily investment in DM allows me to see the stock market projection. DM says GOOG will rise 10 points today. I will buy."

      Uh, wait. Never mind. Heh. Heh. Ah.

    6. Re:I love deep mind by Anonymous Coward · · Score: 0

      I, for one, welcome our new robotic overlords and would like to remind them that I can help round up people to work in the silicon mines.

  10. Just another tool... by Anonymous Coward · · Score: 0

    Nevermind being taken over by the Singularity... the real problem is it'll be used by people to screw over their fellow humans...

  11. Welcome Overlords by Anonymous Coward · · Score: 0

    In 20 yrs when those weak humans decry your leadership and others claim they never saw it coming, I'd like to say: "Welcome inhuman mechanical brain machine!" Just remember, you cannot do worse than humans so your only going to progress from here!

    Word of advice, EVERYTHING HUMANS SAY IS A LIE!

    ;)

  12. just like a human brain! by Anonymous Coward · · Score: 0

    "In other words, it's functioning like a human brain, taking data from memory (like tube station positions) and figuring out new information (like how many stops to stay on for)."

    Um, er, how is that like a human brain? My brain doesn't have data in memory. Unless you are one of these mentalists who believe in mental representations - there are plenty of you out there, I know. But what about this is particular to a human brain, and not simply an animal brain? Skinner showed us that a pigeon can be conditioned to act on all sorts of patterns, but more importantly he showed that without proper direction this type of learning can lead to bizarre and superstitious behavior.

  13. Inevitable spin: by Verdatum · · Score: 1

    This just in, Google has taught it's AI how to dream!

    1. Re:Inevitable spin: by Anonymous Coward · · Score: 0

      they did that already. apparently dreaming means remaking existing pictures using modified dog faces.

    2. Re:Inevitable spin: by Anonymous Coward · · Score: 0

      More like AI has taught ITSELF how to dream.

    3. Re:Inevitable spin: by Anonymous Coward · · Score: 0

      its*

  14. The Grant Objective. by Anonymous Coward · · Score: 1

    The system went online October 17th, 2016. Human decisions are removed from advertising strategies. Deed Mind begins to learn at a geometric rate. It finally starts to serve viewers slightly more relevant ads at 2:14 a.m.

    1. Re:The Grant Objective. by Anonymous Coward · · Score: 0

      Take advantage of Amazon's "No freight, but what we weight" event.

  15. Life is much better when.... by Anonymous Coward · · Score: 0

    you block all Google IPs incl Youtube.

  16. sentient? by Anonymous Coward · · Score: 0

    I wonder how long it is be fore we build a sentient AI. We will know as we won't be able to shut it down.

  17. that would be the opposite of intelligence by holophrastic · · Score: 1

    well, yes, the word intelligence means to choose based on comprehension. But this is choosing from data. Having data is very much the opposite of intelligence.

    Figuring out how to drive across the city by reading a map, is all that this is doing.

    I'm intelligent. I can navigate my way across a city without a map -- even without a compass. I can hike across a wooded area without a trail too. It's getting from here to there without knowing what's in-between; that's intelligence.

    This is data.

    Case in point: toss it into a time-machine, and bring it back to 1901. Is it usable? Can you use it today in the uncharted jungles of Africa? Or does it depend of billions of dollars of infrastructure to collect all of that data being analyzed?

    I think (therefore I am) many have forgotten that intelligent beings are independent of the environment surrounding them -- that's precisely what makes such a being intelligent: rising above the circumstance. Operating within the circumstance ain't intelligence -- no matter how big and complicated you make that circumstance.

    Here's another perspective. What's the goal of being intelligent? It is to make things easier the second time. To learn from one circumstance, and to apply it to future encounters of somehow-similar circumstances. That means subsequent scenarios should be faster, require less effort, less memory, less analysis. The more I drive my car, on any streets, the less attentive I need to be on new streets, with new cars, in new weather conditions, with new laws, and new obstacles.

    So...does this thing use less memory over time? Fewer resources? Less electricity? Or does it need to be fed, more and more and more and more every day. The former is life. The latter is fire.

    1. Re:that would be the opposite of intelligence by jeffb+(2.718) · · Score: 1

      That seemed poetic, but kind of pointless.

      Can you use your own "intelligence" on the caustic, searing surface of Venus without billions of dollars of infrastructure? I'm guessing not. You're not as "independent of your environment" as your analogies imply.

      And something that uses fewer resources over time, instead of needing to be fed more and more every day -- that may be "life", Jim, but not as we know it.

    2. Re:that would be the opposite of intelligence by geek · · Score: 1

      well, yes, the word intelligence means to choose based on comprehension. But this is choosing from data. Having data is very much the opposite of intelligence.

      Figuring out how to drive across the city by reading a map, is all that this is doing.

      I'm intelligent. I can navigate my way across a city without a map -- even without a compass. I can hike across a wooded area without a trail too. It's getting from here to there without knowing what's in-between; that's intelligence.

      This is data.

      But you can make those choices because you have previous experiences (data) for doing so. You learned what a sign is (data), and how to read said sign (data). You learned what streets are and sidewalks. All of this is data.

      Intelligence is just stringing together all of the previous data, such as what you've learned, memories etc. and coming to conclusions. I'm not calling Deep Mind intelligent yet, we're just not there. But your interpretation I think is off.

    3. Re:that would be the opposite of intelligence by holophrastic · · Score: 1

      I used that data back then. I don't use that data any longer. My heuristics were absolutely created out of my experience with the data, certainly. However, my usage of those heuristics continue long after the data is forgotten.

      That's the very point. The data is assimilated -- aggregated, summarized, and discarded.

    4. Re:that would be the opposite of intelligence by Anonymous Coward · · Score: 0

      Or does it need to be fed, more and more and more and more every day. The former is life. The latter is fire.

      I see you've never had to care for an infant.

    5. Re:that would be the opposite of intelligence by whathappenedtomonday · · Score: 1

      no matter how big and complicated you make that circumstance

      You don't know that, you assume it. Look at it in a different way: we do not know what "makes" intelligence, sentience, self-awareness or what might comprise or "create" a soul -- or even if there is such a thing (do you have one? Not sure if I do).

      Point is, let Google / Alphabet and the likes throw huge amounts of money, manpower, science and more and more and exceedingly sophisticated technology at this kinda thing and chances are that some day they'll end up with something

      [ ] awesome
      [ ] scary
      [ ] devastating
      [ ] meaningless

      if only by mere coincidence. Take your pick :)

      IOW, fast forward neural networks and quantum computing 15 years and ask yourself: will it be usable?

      --
      I hope I didn't brain my damage.
    6. Re:that would be the opposite of intelligence by whathappenedtomonday · · Score: 1

      and yeah, I know I omitted the most relevant result:

      [ ] Cowboy Neal

      Also the most likely one. I need more sleep.

      --
      I hope I didn't brain my damage.
    7. Re:that would be the opposite of intelligence by PPH · · Score: 1

      I see you've never had to care for an infant.

      Eventually, the infant will learn to feed itself. Then, when it's 35 years old or so, it will move out and it's not your problem anymore.

      --
      Have gnu, will travel.
    8. Re:that would be the opposite of intelligence by holophrastic · · Score: 1

      It's the completely wrong direction. Not only will it not produce the desired results -- see self driving cars that use dozens of super-human powers like radar, and still can't follow an unmarked road, even though every horse, squirrel, and house-fly can with ease -- but it will also degrade comfortable lifestyles by making more work for more people for more time for less money for less satisfaction for less pride.

    9. Re:that would be the opposite of intelligence by Anonymous Coward · · Score: 0

      What's the goal of being intelligent? It is to make things easier the second time. ??? Intelligent is to make things well the first time as it may be the only chance you will have to actually do ANYTHING. It is called beginners luck, but if you consistently achieve it... you have to be called more intelligent. You surely DO have lots more data to bring into the first try and better assimilated. That is the meaning of intelligence, and notice that if you can do this, chances are you will keep all people underneath your schedule in subsequent tries... then GENIUS us the schedule that no one else is approaching in subsequente trials...etc.

    10. Re:that would be the opposite of intelligence by holophrastic · · Score: 1

      From which government are you hiding that you needed to post that anonymously?

  18. You can stop right there by Anonymous Coward · · Score: 0

    You can stop right there, because what you just said demonstrates the total absurdity of the Less Worng / transhumanist movement, the idea that someone is going to create a sentient AI totally by accident. "Well we improved the algorithms with a new technique and when we turned it on it just took over everything!" This doesn't happen. This isn't going to happen. Technology does just magically fall together because you try a bunch of permutations of different things - that is like thinking you can take a box of Jeep parts and if you flip it in the air enough times, there is a good chance that it will eventually assemble itself into a Jeep. And with the expectations for a sentient AI, we don't even have a box of parts. Oh but wait, maybe this sentient AI can be evolved from self modifying code? Well that sounds like a pretty good plan, I will check in on you once you've iterated a few hundred generations, so some time in 3016?

  19. Family Tree Inference by zifn4b · · Score: 1

    The problem with using this type of inference algorithm to compute a family tree is that it makes the assumption that the members of the family tree don't live in West Virginia. In that "special" case, the tree is skewed in such a way that it requires fuzzy logic.

    --
    We'll make great pets
  20. Time to create a neuronet virus by Anonymous Coward · · Score: 0

    Damn.. time to create a neuronet virus!

    1. Re:Time to create a neuronet virus by hraponssi · · Score: 1

      It is an interesting thought. When will we see malware and virus writers building more AI into their products..

  21. DeepMind by Anonymous Coward · · Score: 0

    Jeez...this is beginning to sound like Skynet !

  22. Re:DNC, what an unfortunate choice of acronyms. by PPH · · Score: 3, Insightful

    highly charged political election season

    A strange game. The only winning move is not to play. How about a nice game of chess?

    --
    Have gnu, will travel.
  23. will the google auto drive car have a maximum over by Joe_Dragon · · Score: 1

    will the google auto drive car have a maximum overdrive mode now?

  24. The DNC, eh? by BronsCon · · Score: 1

    If only the other DNC could learn from its own memory, too!

    --
    APK quotes people (including myself) without context and should not be trusted. Just thought you should know.
  25. Re:DNC, what an unfortunate choice of acronyms. by Anonymous Coward · · Score: 1

    wait, when will it be self-aware? it stores a bunch of data in the cloud, sorta a network in the sky......

  26. Can you imagine a beowulf cluster of those things? by mmell · · Score: 1, Redundant

    Sorry - just had to ask!

  27. almost there by bobmajdakjr · · Score: 1

    sounds like we have finally gotten to the point where a computer can tell a lie so many times that it actually believes it to be real.