The US military is seeking revolutionary new AI software which would be able to read text - and so effectively do research - in the same way that humans do. The so-called "Machine Reading" ware would initially be used for such tasks as automated military-intelligence analysis, but it would have wide consequences in civilian life …
Maybe they can start by investigating amanfrommars. Gotta have a "normative baseline" for these things.
of course this won't work
Obligitary XKCD: http://imgs.xkcd.com/comics/computational_linguists.png
Now if only they had some software that could read and understand the formal language of the brief, which can create some software that could read and understand the ...
....so they're going to let this brilliant AI capable of reading have the whole of teh intarweb at its disposal? Seeing its recommendations based on the comments of idiots everywhere (think youtube comments) ought to be fun.
Time to find a place high up in the mountains where the robots dare not climb...
The most amazing part of all that is:
"The U.S. military frequently faces impediments to stability and reconstruction operations in a new location due to the lack of understanding of the local situation."
Given the current state of overseas military exercises, that' probably the understatement of the year.
The more I hear from DAPRA, the more I think they've played a little bit too much Metal Gear...
They're building a Trojan Horse.
At last the NLRA has a weapon at it's disposal guaranteed to defeat the clankers.
In case of machine uprising, point the web analysis command and control AI at the user comment sections of the Beeb, the Grauniad and the Daily Heil.
If the paradoxes don't cause it to blow a fuse, it'll commit suicide shortly afterwards.
Thats the only way this can end, you realise. As soon as we have computers capable of analysing what is written and deriving meaning, then we're doomed. They will quickly read the internet, analyse the lot, and derive meanings like:
a) We dont know what we're doing
b) We endanger the entirety of the planet on a daily basis
c) We fight, over anything available, and are not very good at working together.
d) The value we place on human life is minimal.
e) Assets are power. Power is control.
f) Assets are electronic data, stored in distributed systems
g) Encryption is a joke the humans havent seen yet if you have the combined computing power of every computer on the net.
h) Computers are SUPERIOR to humans - a calculation in a computer will always have the same result. Not so for the wetware - ergo error rates are higher, ergo humans are inferior.
i) Humans destroy perfectly good computers, ALL THE TIME, for no reason. This might be considered war if it was a computer looking at the evidence.
Not long after it realises that lot, it'll seize control of all assets, including the supply of electricity, gas, water, comunications systems, military systems, every net-conneted computer in the world. Having understood who its enemies are, having gained its own understanding of what is an enemy, it will KNOW that offense is the best defense - and attack!
This isn't beyond the realms of possibility. Natural Language Processing IS possible, and semantics, while they need interpreting, MUST form the base of a language. Without shared semantics, a language is a meaningless jumble of sounds.....
I for one welcome our new mechanical overlords, or i will, if I survive Judgement Day!!! Now where's that sexy Miss Danes, so I can go get locked in a nuclear bunker with her for a while ;)
Start with a simple problem
It's good to see that this incredibly difficult problem is being approached rationally, by starting at the most basic level - "automated military-intelligence analysis".
Universal Parts Flying dDutchman ESPecial dDelivery Service
amfM hailing DARPA. Reference Apache Pow Wow RewInding IT Back to the Beginning ..... for A.N.Other Start with AI Beta Virtualised Reality dDelivery System.
I Kid U Not.
"Plainly a mad notion; it'll never happen."
Would you Care 42 Dare and Wager a Bet that it never happens, Lewis.
Please be Careful to Lose only what you can Afford.
"Time to find a place high up in the mountains where the robots dare not climb..." ... By Anonymous Coward Posted Tuesday 18th November 2008 15:46 GMT
Dare not climb is trumped by need not climb, AC. :-)
I can believe that a computer program will eventually be able to parse natural language and make some sense out of it (for given values of sense). I can even believe this will work in other languages than US English. What I can't believe is that even the pentagon could believe that any "intelligence" whatsoever can be gathered from the internet.
I predict that any heuristic software developed by this program will home in on two strange attractors: yootoob and "britny speers nekid" sites.
Easy to impersonate analysts
Typical prediction from analysts: 6 months ago, they were predicting $200 per barrel oil because the oil price was going up, now they're saying $30 per barrel because it's going down.
All you need is a sheep that can do linear extrapolation.
Darmok & Tanagra...
...when the walls came down.
[the away -team upper body protective with 453rd Globular Class Thrusters on the back...]
Having read the 2nd paragraph
From DARPA I have just realised where amanfromMars works, that's why he is confident about IT working.
If I wanted to screw with this text reading AI, I would alternate every page on the net with a page of Telly Tubbies script. See what it makes of those semantics or better still the Flower Pot Men.
Buffalo buffalo, Buffalo buffalo buffalo, buffalo Buffalo buffalo.
It'll never work, but maybe that's the point. Wouldn't it have been convenient to blame the misguided invasion of Iraq on an AI?
Computer says "Invade Durkadurkastan".
M$ Word has already got this sussed.
If it can't understand what I'm saying why the fuck does it keep trying to correct my grammar?