And Jinxter's on its way...
Enjoy The Pawn and The Guild of Thieves in the meantime.
How do you train AI agents in language and understanding? Easy – drop them into a Zork-alike dungeon and let then find their own way out. While the rest of the Microsoft world squeaked with excitement about new hardware and free collaboration tools, the boffins at Redmond’s Montreal Research Lab have looked to history as way …
I used to play about with a text adventure construction kit (don't recall it's actual name) on my first PC - the only way beasties would kill you with was one 'huge teeth' description, which I kept forgetting, A spy epic I wrote for it was a little freaky with beautiful spy ladies coming at you with knives/guns but at the last minute, taking you out instead with a huge set of sharp teeth.
Any AI being forced to play that and against all odds, becoming self aware is going to need therapy.
"If it ever learns to quit vi then we know we're in trouble."
When you consider that the only thing I know about how to use vi is how to quit it, using "killall -TERM vi"* from another terminal, if AI learns the killall command, we are indeed in deep shit.
* "killall -KILL vi" if I'm particularly annoyed at the time. And yes, I am aware of ":q" I think it is, but killall works for most other things as well.
Text based, I get, i fondly remember (well, small flashes in between the long cold-sweat nightmares of fruitless and mostly unrewarded frustration) but turn-based always puts me in mind of the only slightly less frustrating Civilisation, Colonisation or UFO and their ilk from the early to mid-nineties.
If the AI were any good you could plug it into any text adventure (maybe not Corruption, perhaps something from the other end of the decade).
As it is, it seems the deck has been loaded because care has been taken to generate the descriptions and parser with known or easily learnable words.
... would be to set an AI the test of getting this thing up and running in the first place, like our plucky correspondent. i.e. In the face of incomplete and inaccurate instructions can an AI actually achieve an outcome it has been given in the form of a verbal or written instruction ?
The answer of course is "Don't be ridiculous."
With that in mind, I here-by declare the end of supposedly tech-literate journalists throwing the term "AI" around as if it is either meaningful or relevant to any current technological endeavour.
Biting the hand that feeds IT © 1998–2019