The Core

Why We Are Here => Traffic => Topic started by: DogBoy on December 31, 2016, 03:51:19 PM

Title: Still thinking Google's AI is harmless?
Post by: DogBoy on December 31, 2016, 03:51:19 PM
So there I was... innocently dragging a partial image of lithium boat battery (with a file name of battery-trash.png) into the search box, hoping Google Image Search might identify the manufacturer.... but instead of finding that, it came up with this... "Best guess for this image: battle of el alamein" (SEE ATTACHED)

Me: "Hey Google, Can you help me find a battery?"
Google: "Unfortunately, we seem to be all out of batteries, but we have some death and destruction, you might find of interest..."

...and yes, I was logged out; that's default search.
Title: Re: Still thinking Google's AI is harmless?
Post by: DogBoy on December 31, 2016, 04:05:35 PM
I think it's possible they are combining the term 'battery', with 'battle', as a misspelling, and seeing the "+" as a flag.  I see the same thumbnail repeated at the bottom of that SERP, that led me to the page ATTACHED... + is the only thing that jumps out at me.
Title: Re: Still thinking Google's AI is harmless?
Post by: DogBoy on December 31, 2016, 04:19:17 PM
yup, its seeing it as a flag...

When I took a snapshot of the top half of the battery, and named it the exact same name, it came up with 'hardware'.... with was more in-line what I was thinking:)

So be forewarned, that thing is quick to go to war:)
Title: Re: Still thinking Google's AI is harmless?
Post by: ergophobe on January 01, 2017, 12:39:54 AM
The US nuclear arsenal is impervious to interference from the Google AI... or at least the Minuteman III is, since the launch sequences run on IBM Series/1 from 1976 and require 8.5" floppy disks or pressurized phone lines to exchange data

Of course, they are IBM Series/1 from 1976 and did at one point detect 2200 Soviet ICBMs inbound and Brezinski woke up Carter to tell him had something like six minutes to decided whether or not to initiate launch... and then a minute later NORAD called back and said "Ha ha! Sorry guys, little mistake. It's just a bug."
Title: Re: Still thinking Google's AI is harmless?
Post by: buckworks on January 01, 2017, 01:52:11 AM
FWIW, "battery" is very much a military term in some contexts.
Title: Re: Still thinking Google's AI is harmless?
Post by: ergophobe on January 02, 2017, 04:05:55 AM
True Buckworks! It's still an odd result because if it were interpreting "boat battery" as a military thing, I would expect to see the big artillery from a battleship. What it's showing instead are things that are visually similar (squarish, tan thing with + signs on them, also know as Iron Crosses). It seems to think the square box with the + sign is a German tank with an iron cross.

It could be seeing it as mechanized artillery and be thinking "artillery battery" but I think Dogboy is right - it's just not that good at figuring this out yet