back to article With Alphabet's legendary commitment to products, we can't wait to see what its robotics biz Intrinsic achieves

Alphabet today launched its latest tech startup, Intrinsic, which aims to build commercial software that will power industrial robots. Intrinsic will focus on developing software control tools for industrial robots used in manufacturing, we're told. Its pitch is that the days of humans having to manually program and adjust a …

  1. jonathan keith Silver badge
    Terminator

    ROTM

    Combined with Google's ongoing AI training program, this is going to end well...

    1. Dinanziame Silver badge

      Re: ROTM

      At least they don't own anymore the terrifying dancing terminators from Boston Dynamics

  2. Pascal Monett Silver badge

    "the AI can work out the best way to achieve its goal"

    No. The Statistical Analysis Machine is going to use and break a million industrial robots in order to find the best way to do one job without breaking either the machine or the piece it is supposed to work on.

    IBM finally made a computer that could soundly beat a master chess player - but only because said computer played 60 million games against itself before being confronted to a human, which doesn't have a chance in hell of playing more than a few thousand in his lifetime.

    Industrial robots are built for one task. A riveting robot is not going to paint, a painting robot is not going to cut metal.

    Intrinsic might be an interesting experiment in applied computing science, but I doubt companies will want their expensive equipment "finding out" how to do the one thing they were bought for.

    1. Korev Silver badge
      Terminator

      Re: "the AI can work out the best way to achieve its goal"

      Intrinsic might be an interesting experiment in applied computing science, but I doubt companies will want their expensive equipment "finding out" how to do the one thing they were bought for.

      I guess the vendor of the robots might be more interested though

    2. cyberdemon Silver badge
      Terminator

      Re: "The Statistical Analysis Machine is going to use and break a million industrial robots"

      I'd guess that most of the learning is done in simulation - only once the 'statistical analysis machine' has analysed enough data from the simulation to be able to control a simulated robot, will it ever be allowed near real hardware.

      For a good demo of how this works: https://www.youtube.com/watch?v=aTDkYFZFWug

      Google have realised that all they need to make terrifying robots is LOTS OF COMPUTE, LOTS OF DATA, and they have both of those.

    3. Irongut

      Re: "the AI can work out the best way to achieve its goal"

      I imagine its going to be particularly fun working alongside welding robots that no one knows what their programme is or why they are doing what they're doing. Or any of the other very dangerous jobs we give to production line robots.

      1. parperback parper

        Re: "the AI can work out the best way to achieve its goal"

        looks like the AI has determined that if it flails about at random intervals with the welder going full whack it keeps the meatsacks away.

        No operator has made any settings changes for a while, so it seems to be a very stable and hence optimal behaviour.

  3. Boris the Cockroach Silver badge

    AI to program

    industrial robots?... is it April the 1st again? this covid pandemic has really screwed up my sense of time..

    We dont want the robots to program themselves or analyse the task or do anything other than what we tell them to. because they are bloody expensive when they break.

    our way is that repeat work is saved as programs, the machine tool is loaded with a part number, the robot is loaded with a program matching that part number, and away we go making bits (I've simplified the process a lot here )

    Any new work is designed up and tested as best we can off the machine, followed by teaching the robot howto hold and move the part.. everything done nice and slow so we can see the problems as they occur so where is the AI going to fit in?

    20 years learning machining and programming? or is it just going to find 'best fit' from a huge database of knowledge without the specific knowledge that always catches us engineering types out

    Or put it a shorter way "would you fly on an aircraft thats had its flight control software written by an AI?"

    1. Chris G Silver badge

      Re: AI to program

      From the sound of their blarb they are a little short on hard manufacturing experience.

      Maybe they could start with a pick and place pill packer before bringing an entire factory to a halt.

    2. John Brown (no body) Silver badge

      Re: AI to program

      "our way is that repeat work is saved as programs, the machine tool is loaded with a part number, the robot is loaded with a program matching that part number, and away we go making bits (I've simplified the process a lot here )"

      They are trying to dumb down the programming of the robots and off load the "learning" to cheap AI so employers can use cheap labour instead of very expensive engineers and programmers. When it goes wrong, there will be no one left who knows *why* it went wrong.

      1. Boris the Cockroach Silver badge

        Re: AI to program

        Quote

        "so employers can use cheap labour instead of very expensive engineers and programmers. When it goes wrong, there will be no one left who knows *why* it went wrong."

        I heard on the grapevine there was a factory up north that went down that very route.

        Very limited range of stuff being made, so they hired a bunch of engineering whizzes and programmers to build the production cell (I think 6 £750 000 machining centers plus robots and pallet loaders to link them)

        Having commissioned the cell all the expensive workers got the P45/Pink slip and they hired a bunch of min wage people to follow the instructions left to them.......

        Went swimmingly well for 2 weeks or so.... then an automatic tool measurer reported a defective tool, said min wager did not replace it correctly and 1 machining center took a £100 000 crash as a result.

        One weeks down time while it was brought back on line........ then it happened again with another weeks down time..... could have hired a well paid engineer/machinist for 5 yrs with the money they lost on that cell ....

        Not that we cared......it was a rival company after all

  4. JacobZ
    Facepalm

    Apocalypse

    Google will kill this program in about three years.

    Because the one thing the robot apocalypse really needs is orphaned robots.

    1. Flocke Kroes Silver badge

      Re: Apocalypse

      At least some unemployed elephants have a home to go back to. I can just imagine these robots desperately begging for change outside railway stations with no passengers because everyone is working from home.

  5. xyz123

    Google AI - now we have a way to cancel products before they've even been thought of, let alone reached the market!

  6. Anonymous Coward
    Anonymous Coward

    With Alphabet's legendary commitment to products?

    I'm a user, I'm not a product, so I'm just Alphabet's lunch.

  7. Steve Davies 3 Silver badge
    Big Brother

    How long will this last?

    Given Google's world renowned habit of announcing something that was going to change the world and killing it off only a few short years later.

    Will this just turn out to be another flash in the pan?

  8. Anonymous Coward
    Anonymous Coward

    There should be a lot of room for innovation in the "last cm/inch" of movement and optimization of path movement, combined with tendon like (active) springiness and damping in joints. "Solving" that sort of control problem could well be more efficiently done using available "learning" methods than trying to solve analytically.

    Another way to look at it is simply solving complex non-linear regression problems using a lot of data. In that way it's no stranger than calculating weather a week in advance using supercomputers and simulation.

    That's a lot different from an autonomous driving system trying to calculate whether or not another vehicles lane changing activity is "bluff", and whether it should charge on through anyway.

    One thing I noticed in autonomous Telsa driving videos was that the turn control is rather jerky. The video author himself, recording his own driving, even commented on that phenomena. It seems to me getting rid of that jerkiness would be an excellent application of AI as a non-linear regression problem solver.

    1. John Brown (no body) Silver badge

      "One thing I noticed in autonomous Telsa driving videos was that the turn control is rather jerky. The video author himself, recording his own driving, even commented on that phenomena. It seems to me getting rid of that jerkiness would be an excellent application of AI as a non-linear regression problem solver."

      Maybe it's an artefact of "digital" steering and not using small enough steps.

  9. elsergiovolador Silver badge

    Money

    How small business can compete with something like this?

    Government all can think of is slapping more tax on the workers (check the latest proposal to increase the NI). This example clearly shows that taxing corporations through workers does not work.

    Workers cannot save meaningful amount of money to start their own business or focus on their own R&D and they are essentially forced to build someone else's dream.

    If only those corporations paid taxes like everyone else.

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 © 1998–2021