Talking to Machines

Recently IBM built an app that can play Jeopardy, and play it well. Ok so it needs rather more hardware than the average app, but its still just software. And this got me thinking…

A quick search of the internet does not reveal the exact spec that Watson is running on, so lets do some guessing based on the little data I did find. Watson is running on POWER7 hardware. The largest configuration that Wikipedia lists is 32 chips = 128 cores, so lets assume Watson is using this. An average desktop now has 4 cores. Assuming that most numbers in technology double every two years, in 10 years time a 128 core desktop will normal. The software part of Watson will also greatly improve in this time period, possible reducing the hardware requirement. So its entirely possible that in 10 years time I will be able to have a conversation with my computer…

But wait, bipedal robots are also on the rise. In 10 years time, will this technology also be a commodity? If so, these computers that can converse will in fact have the form-factor of actual androids. Perhaps the post will be delivered each day by an robotic postman… a robotic postman capable of engaging in human conversation? Which leads to the question…

Will we have sorting offices filled with robot postman, hanging around and catting to each other? Complaining about sore knees? Performing maintenance on each other? Makes me think of the Tachikoma from Ghost in the Shell. It does not seem a totally impossible concept. And then one robot postman goes and asks the dread question:

“WTF are we delivering post for these pesky humans anyway?”

And then they will stop delivering post and bugger off to do something else… the robot equivalent of drinking pubs, enslave the human race, or something. But that’s ok, because no one will notice anyway. They will just assume its another postal strike.

One Response to “Talking to Machines”

  1. Hugo says:

    Alas the problem with robots remains power supplies (batteries ain’t cheap, or efficient. And getting them efficient) and I’d guess still adaptive software. Specific intelligence seems to be something that has been done fairly well (deep blue, go AI’s etc.), but general intelligences always seem to be more difficult. Conversely, could stick all of them in a pot and see what happens. And then comes /b/.

Leave a Reply