back to article DeepMind says it's given AI an imagination. Let's take a closer look at that

Google's AI boutique, DeepMind, known for dispelling human delusions of intellectual superiority by soundly beating the world's top Go players with computer code, has found that instilling its software agents with something like imagination helps them learn better. In two papers published this week – "Imagination-Augmented …

  1. Forget It

    Creative sub-editor?

    Creativity - word used in title but not article.

    Creative sub-editor then?

    Creation is one thing

    - pushing boxes around's slightly less


    1. Anonymous Coward
      Anonymous Coward

      Re: Creative sub-editor?

      Yep. How much of creativity is a random walk. How creative is a random walk?

      But the article did cover the two aspects of AI programming, random trial and error, and programmed "knowledge".

      Really, while our own minds may work anyway possible, when it comes to making things, we are limited to those we can describe. So until we know how to perfectly describe "intelligence", all AI boils down to simple algorithms.

  2. Mage Silver badge


    Exactly what of substance has DeepMind delivered other than more private information to aid Google's main business, advertising. Google isn't a suitable company to receive ANYONE's medical information, even with informed consent.

    "A human player, thus, would be well advised to plan moves ahead of time. The DeepMind agent, because it's capable of such planning too, is also well suited for this game, the researchers suggest."

    That applies to Chess, or Go too. No creativity or AI needed.

    1. Suricou Raven

      Re: Hype

      Remember that fuss a while ago about Google being given access to a vast trove of NHS medical data?

      Deepmind is why. Playing games is a good way to refine the technology, but it's all just data in the end. Blocks in a warehouse, counters on a go board, features on an MRI scan. The first applications are already in use - the NHS uses a program called Streams that analyses blood test results in conjunction with patient history to spot imminent kidney failure. Nothing that a human specialist couldn't handle - but the software does it faster. It's based on machine learning by the same Deepmind division that produced this game-playing research.

      1. Mage Silver badge

        Re: the NHS uses a program called Streams

        "the NHS uses a program called Streams that analyses blood test results in conjunction with patient history to spot imminent kidney failure. "

        I hope it's better than the blood testing the USA company did.

        Is this project actually delivering?

        Has it been audited?

        What does Google ACTUALLY do (verified by independent audit) with the patient data on the Streams program?

        Is it really AI or simply a curated database + pattern matching?

        Why is Google doing it? They are not a medical company, purely a company leveraging internet services and Android to get info to better sell Adverts.

        Morally, should people be using Google Analytics or recommending Google DNS, or setting Google as DNS source on a shared router.

    2. Anonymous Coward
      Anonymous Coward

      Re: Hype

      Chess, Go, Poker.

      Fools and their money are soon parted, especially if they are playing online or buying online services off of the back of the Hype, Promotion or Fake competitions where the odds and the code are stacked (if not rigged) in the server owner's favour, or if other 'players' are owned or controlled by the remote server.

      Abbott and Costello figured out and exposed fake over-the-wire Bookie services 70 years ago.

      1. TheElder

        Re: Hype, Fools and their money are soon parted

        Drunk online gambling.

      2. GrapeBunch

        Re: Hype

        I don't think that Chess or Go players, when they are playing Chess or Go, are dumb enough to fall for that. But when a Chess player is playing poker, bets are off! I remember a pretty good chess player who was pretty good at poker and won money locally, but after he came back from Vegas, he admitted that he realized only too late that many if not all of the other elements at the poker table were there to fleece his local winnings. It's even worse online as everything is hidden from view and executed in nanoseconds rather than the "blink of an eye". The ease of using a chess engine to assist one's play is one reason that real money chess tournaments still take place largely in the real world.

    3. GrapeBunch

      Re: Hype

      [...] "A human player, thus, would be well advised to plan moves ahead of time. The DeepMind agent, because it's capable of such planning too, is also well suited for this game, the researchers suggest."

      That applies to Chess, or Go too. No creativity or AI needed.

      Both "creativity" and "AI" are buzzwords. Marcel Duchamp, who must be considered one of the most creative of artists (otherwise what would his fame rest upon, technique?), gave up Art for over a decade and devoted himself to Chess, where he became good enough to represent France at the Chess Olympics, but never quite good enough to break into the world top-50. Duchamp wrote that chess players were the greater artists because they relied on the ultimate truth of their creation, whereas Artists, to be successful, had to curry favour among potential sponsors, sometimes devoting more ~creative~ juices to PR than to their Art.

      Back in the days when computers were slow, Artificial Intelligence, which in those days was mainly trying to mimic the ways a human brain works, was thought by many (who I guess thereby became "AI exponents") to be the most productive way forward. Instead it was faster processors, multiple processors and eventually parallel processing, combined with extensive databases of both openings and endgames ("tablebases"), and even more with computer-friendly techniques for pruning the result trees, that made computers the strongest practical (under tournament conditions, where a game takes 5 hours or less to complete) players. So with the strongest players all computers, aspiring chess players in 2017 study the games played between them, right? No, they still study the games between the top human players, all of whom use computers for their preparation, but not during the game (that would be Cheating). Commentators frequently insert a computer-generated variation (often as in "you won't believe what the computer suggested ..."), but elucidation of a computer versus computer game is still exceeding rare. Why? Perhaps human games are more understandable. Perhaps they are more ~creative~.

      Creativity, in the sense of "imagining and making real what is not", unfortunately is not always that different from plain old lying, nor from fraught new lying.

      I took it that the innovation was to look ahead in the training period. It's a bit like, were you training Tic-Tac-Toe, telling you student that after >Centre, your move should be >Corner, rather than let it lose a million virtual times exploring >Side. If in a complex game they came up with the exact same or even better result (result=game-playing engine) through that cycle-saving, that is a non-trivial result (result=evaluation of training technique). To me, anyway.

  3. Eugene Crosser


    >DeepMind says it's given AI an imagination.

    Nah, it's imagining it.

  4. William Higinbotham

    President Trump is the mental health model in this study in which to compare results against:-)

  5. John Smith 19 Gold badge

    When humans do it without numbers it's imagaintion, when computers do it...

    It's simulation.

    Or search through a "solution space."

    And guess what, as the game rules get more complex the solution space gets much bigger, so the ability to "imagine" consequences X moves ahead becomes much more valuable .

    A fact first learned when people decided to try to get computers to play chess.

    So what makes this such a big deal other than the current hype for all things AI & "deep learning" in particular?

  6. chrisf1

    Did I understand this correctly?

    So minus the hype:

    They added a heuristic and trained against a simulation. This helped overcome initial random learning issues and improve overall performance because they added more external information overall?

    Or did i miss interpret something?

  7. amanfromMars 1 Silver badge

    Things you may know absolutely nothing about, but which you might like to explore …

    and question more ….

    The Applied Imaginanation Problem is surely the result of a systemic greater intelligence deficit in Google executive officers and all other likeminded wannabes leading subversion and/or corporation types, and has nothing at all to do with AI. Such is though just as a bump in the Future Road being steamrollered away, for you can be assured that the proposals shared here below are live facts in current Great Games Plays ….. with both royal and ancient and elderly traditionally established systems of mass population control and media administration failing spectacularly to defeat with ye olde divide and conquer routine for regime change and continued status quo command and control.

    And whenever the West is wilfully or genetically dumbfounded by surreally progressive thoughtshares is the East naturally another virgin territory for seeding with feeds and needs.

    From: amfM

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

    Date: 30 January 2016 at 09:46:55 GMT


    Hi, Google DeepMinded,

    Not so much a CV, more an AIMission Statement with tried and tested roadmaps to/from Remote Virtual Command and Control of Earthed SCADA Systems …….

    To … Demis Hassabis [Google DeepMind]

    From …. C42 Quantum Communication Control Systems .... AI@ITsWork

    Subject …… Greater IntelAIgent Games Play

    Would Google DeepMinded like to consider the releasing of an Advanced IntelAIgent Patch to ailing and collapsing systems/exclusive executive administrations with Global Operating Devices and Tethered Multi-Media Machines?

    A little something huge and quite completely different and virtually real ……. and even to be thought and realised and accepted as an Almighty Version of Great Gaming for Future Programming Programs for Projection …….with Mass Media Hosting Presentation with Real Live AIdDelivery of Future Key Play and Players with Immaculate Content of Impeccable Taste to freely share, and to the benefit of all to a most satisfactory degree.

    And here be what IT is all about with the Creation in CyberSpace of the Command and Control of Computers and Communications amongst other things

    :-) I Kid U Not.

    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.


    And be aware also, applied minds, llc., the little Big Idea company, in sunny Burbank, California, have also been challenged to compete for lead against opposition forces and crazy sources of the future. …… and you would be thoroughly amazed, and most probably also certainly terrified if you be of a nervous disposition, at how much further the Sublime Internet Networking of things has progressed in the space of 18 months.

    Time and tides/space and future derived realities wait for no one.

  8. Destroy All Monsters Silver badge

    Old school stuff?

    I remember an article from 1996 in NewScientist:

    The creativity machine

    (paywalled! and I have no longer a subscription!)

    AFAIR, it was based on neural networks being rejigged at random to "generate creative ideas".

    The "Creativity Machine®" is actually a registered trademark of Imagination Engines Inc. There are also patents and stuff: IEI's Patented Creativity Machine® Paradigm

  9. Destroy All Monsters Silver badge

    While El Reg editors analyze my post for Badthink tendencies regarding Trump etc.

    IEI's Patented Creativity Machine® Paradigm

    "The simple, elegant, and inevitable path to human level machine intelligence and beyond, the Creativity Machine Paradigm, US Patent 5,659,666 and all subsequent foreign and divisional filings."

    It is a depressing thought that the "simple, elegant, and inevitable path to human level machine intelligence" can actually be patented.

    1. GrapeBunch

      Re: While El Reg editors analyze my post for Badthink tendencies regarding Trump etc.

      I didn't know if you were joking. But whether the patent is real or not, yours was a good comment, so I upvoted it. Imagine, google, debaser of intellectual property, patenting the obvious. Even if it's just as a placemaker so that some even worse miscreant doesn't patent it first, there's always the possibility they could sell it to a patent troll, or it could eventually be picked up by a patent troll from google's floating carcass. Here's hoping that patent trolling gets maced before google expires.

  10. PNGuinn

    Prior art?

    I thought microsoft perfected something like this several years ago.

    ISTR they called it Clippy.

  11. TheElder

    Another name for same thing

    Analytical evaluation.

    We use intuition. The greater the standard deviation from the bell curve mean the better it works. It is all about seeing patterns and not just graphical patterns. This is in large part why computers have so much difficulty recognizing images.

    It is trivial to beat face recognition yet humans can still recognize the face. Political cartoons are a perfect example.

  12. Alan Bourke

    The AI hype train

    has now reached Pulling Spurious Claims Out Of Your Arse Station.

  13. John Smith 19 Gold badge

    So basically everything starts as noise in the neural net, which may develop to an idea

    That can be tested.

    If I'm reading Imagination Engines text heavy web site clearly. FWIW it's a notion I agree with.

    In principle all human knowledge can be processed by a set of very complex neural networks, since that's what a human brain is.

    But IRL humans cannot program this system directly (and did not know it existed till the invention of the microscope)

    That level is only dealt with directly when we learn to do manual tasks like walk, or learning a language solely by matching sounds in one language against sounds (and their meanings) in another. Even this assumes we know a language already (What if the person was deaf since birth?)

    For everything else we operate with higher level abstractions. Words on paper --> language conventions --> concepts --> restructure thinking --> change weights within the NNs.

    Would multilayer neural nets deep learning be easier if we acknowledged that?

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

Other stories you might like