back to article DeepMind boffins are trying to help robots escape The Matrix and learn for themselves in the real world

Google DeepMind is trying to teach machines human-level motor control using progressive neural networks – so that the robots can learn new skills on-the-go in the real world. The idea is to build droids that can constantly learn and improve themselves, all by themselves, from their own surroundings rather than rely on lab- …

  1. Anonymous Coward
    Anonymous Coward

    Where are these projects that scare the Zuck out of Bill Gates & Elon Musk???

    .................Meanwhile in another AI article this week:

    "Soon computers will be routinely inventing, and it may only be a matter of time until computers are responsible for most innovation," Abbott said. "To optimize innovation – and the positive impact this will have on our economies – it is critical that we extend the laws around inventorship to include computers."

    .................Make me a robot that can manage to 'take the stairs', any stairs, all by itself, dependably without falling over.....

    1. Dave 126

      Re: Where are these projects that scare the Zuck out of Bill Gates & Elon Musk???

      >Make me a robot that can manage to 'take the stairs', any stairs, all by itself, dependably without falling over.....

      That's feasible, though its more of a mechanical engineering problem. Unless you're talking about a bipedal robot, in which case yeah, the devil is in the control system details.

    2. Alister

      Re: Where are these projects that scare the Zuck out of Bill Gates & Elon Musk???

      "Soon computers will be routinely inventing, and it may only be a matter of time until computers are responsible for most innovation," Abbott said. "To optimize innovation – and the positive impact this will have on our economies – it is critical that we extend the laws around inventorship to include computers."

      Dear God, why would you possibly think that removing responsibility for innovation from humans is a positive thing? Or that innovation is somehow something that needs replacing with a machine analogue?

      Tell you what, lets replace all the intellectual professions with computers, and all the professions that require physical labour with machinery, and watch the human race stagnate and die.

  2. allthecoolshortnamesweretaken
  3. tiggity Silver badge

    Tailored motor skills

    I for one welcome our individualized preferences motor task machine learning sex doll overlords

  4. jpjeral

    IHMO, what is missing is a progressive refining of the simulated world to make it a better match with reality - that is, after a reality incursion, make this refinement as new things have been learned about reality, which show that the simulation is not exact, spend time retraining in this new simulation, and back to reality with updated skills.

    Cheers, jpjeral.

  5. Anonymous Coward
    Anonymous Coward

    as someone who has a cold currently

    Alice in wonderland and matrix aside long live the red pill in the picture (Dayquil) and fsck the blue one (Nyquil regular ie the give you a hang over pill)

  6. joed

    so much for perfection

    robots will be no better than humans after all

  7. jpjeral

    Comment on "...learning in the real world" (re-posted and continued)

    Another choice would be to add, in addition to the preset goal, a new goal to attain while in the real world: "getting as much information as possible about the real world". As this is a very open-ended goal, care not to "steal" all resources from the original goal. (Inspired from the short story "Réflexe spontané" a.k.a. "Le robot déchainé", by the brothers Arkady and Boris Strougatski).

  8. harmjschoonhoven

    There was once a German professor

    who had trained his dog to recognize slides with geometric figures. He toured from town to town to show the clever dog. Until one day the lamp of the projector failed and the dog still made the right barks; he was trained on the sounds the slides made when they were put in the projector. Something simular happened when neural networks were first used to train driverless cars. The side of the road was recognized not by the curb, but by green grass and the car refused to cross a bridge.

  9. SeanC4S

    There was another paper recently that showed evolution could be just as effective as back propagation. I've been doing a little more work on scale free optimization: http://www.freebasic.net/forum/viewtopic.php?f=5&t=25133

    While it is not possible to fill a human up with 5 liters of vegetable and have them run for a week, it seems like you will be able to fill up a humanoid robot with 5 to 10 liters of methanol and have it run for a week or 2:

    http://www.freebasic.net/forum/viewtopic.php?f=17&t=25086

    You would get around 20 Mega Joules per liter. I presume the conversion efficiency into mechanical work would be reasonable. Kinda bad news in terms of job and our physical security.

  10. jpjeral

    Comment on "...learning in the real world" (continued)

    Shame on you, stupid users! You will read a comment only if it contains the word clinton, trump, scandal, scoundrel, arkassian and everything in that vein. F**k you, degenerates!

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