back to article IDC shock prediction: Someone might build Skynet in next few years

Those crazy tech shamans at IDC have been sniffing the data centre cooling system exhausts again, this time breathlessly informing us that in three years "30 per cent of commercial service robotic applications" will take the form of "robot as a service". We have no idea either. IDC's "Top 10 Predictions for Worldwide Robotics …

  1. Steve Button Silver badge

    House cleaner?

    Even if a fully automated robot cost £100,000 which could do the daily chores (washing, cleaning, vacuum) it would still make sense to get one of these assuming you could share it with one or two of your neighbours. Very rough calculation. Assuming the thing would last 5 years. Or if you could just pay a couple of hundred every month, and get "robot as a service" ... or what in meatbags would be called "a cleaner"

    But a cleaner, which you don't need to vet and would not go through your personal stuff and steal things. (assuming it's secure).

    Just a thought. Someone just needs to invent the thing.

    1. Wyrdness

      Re: House cleaner?

      I'm working on a robot cleaner that *does* go through your stuff and steal things. But only very occasionally, so that you don't suspect it.

  2. TeleC

    Someone?

    "No one User wrote me! I'm worth millions of their man-years!"

  3. Anonymous Coward
    Anonymous Coward

    Your choices are:

    A) hire a regular, immigrant cleaning person for some $/month

    B) subscribe to RaaS or some other share/rent-a-bot service for many $$$/month

    3) leave the fucking mess where it lies and let the cats or ants take care of it

    F) set fire to your flat and get a new one with new clean junk in it!

    Hmmm... 3, please.

  4. SeanC4S

    This paper suggests that highly dimensional neural nets do not (effectively) contain local minimum traps:

    https://arxiv.org/pdf/1406.2572.pdf

    One implication is that extremely naive training algorithms, simpler than even those tried in the early 1980's are likely to be effective with current deep neural nets.

    That is, simplified simulated annealing with no need of occasional energy increasing steps, or equivalently random hill climbing with a cooling rate is sufficient onto the day for online or batch training of neural nets.

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