DARPA Builds Smarter Version of Microsoft's Clippy 140
holy_calamity writes "Microsoft's animated paperclip may be long dead, but a $150m DARPA project has resurrected the idea of a virtual assistant. AI researchers from more than 60 institutions worked on the project entitled CALO. CALO is designed to help ease the bureaucratic burden of the military. A consumer spinoff, Siri, is coming to the iPhone later this year. It responds to conversational voice commands to take over multi-step tasks like choosing and booking restaurants or cabs."
The problem of inductive bias in knowledge trees (Score:2, Insightful)
I have read the paper and am not sure if the researchers have solved the problem of inductive bias, which is the bane of "artificial intelligent" learning on this scale. Basically, suppose you teach monkeys Shakespeare using a tree system of rewards versus noxious odors. This is analogous to the binary decision map tree that the computer system uses. A human might adapt to Milton, or even Cervantes, but a "intelligent" monkey will just start screeching and throwing feces, i.e. Clippy's inane "advice."
But of course any monkey would be better than Donald Rumsfeld. So I guess we're safe.
Smart = Unpredictable (Score:5, Insightful)
Re:Smart = Unpredictable (Score:3, Insightful)
Smart interfaces are a bad idea, because you can never be sure how they will respond.
To a degree, they can only do what they have been programmed to do.
Dumb interfaces are predictable tools so they require less brain power to use than the two-way dialog of smart interfaces.
I wouldn't say dumb interfaces require less brain power at all. In fact they might require more because you might learn something doing it.
With dumb interfaces I can fire off a long string of commands without having to stop and think between each one.
Only because you already know those commands and have them memorized.
This improves productivity more than any supposedly intelligent interface will.
After the learning curve of entry, sure.
So has anyone asked the question... (Score:5, Insightful)
Why don't they just work on easing up the bureaucratic burden in the first place?
A: Likely because it's impossible. An aging and entrenched organization, with no incentive to compete, receives the same amount of tax payer money per year no matter what they do.
My friend works for a branch of the millitary as an accountant, and oh the stories. Just watch Office Space and multiply it by ten. It's comedy gold. I laugh and tell her to quit but she's addicted to the huge paycheck.
Re:A consumer version? (Score:3, Insightful)
Re:Is it time yet... (Score:5, Insightful)
I disagree. I think Clippy, as a personal assistant, was by definition a bad idea.
The computer is an extended tool. Which itself is an extension of your body.
So the idea is, to make it a powerful but fully transparent thing. Like a Mech suit with Matrix interface.
Our hands even have their own special "highway" path around the slow areas of the brain, because of our habit of extending our bodies trough them.
Which means that separate entities in that space give you essentially a split personality. Much like Dr. Strangelove's hand.
Additionally, you have to communicate with that entity in probably the most inefficient and senseless way possible: Text. Or even speech!
Even a keyboard and a mouse, primitive as they are, are still much closer to a brain-computer interface, as putting another layer of a chatting bottleneck below that and the program.
Re:Is it time yet... (Score:4, Insightful)
Similarly I feel AI researchers should focus more on human augmentation, and delay the "create a new entity" stuff.
There are lots of problems if you actually end up creating a new entity - ethical, social etc. It's like forcing ourselves to answer hard questions before we are ready.
Re:Is it time yet... (Score:3, Insightful)
There will be a time.. and a place....
you'd BEG for those directions!