Slashdot Mirror
Digital 'Ghosts' To Guide Students On Campus
Hambone.dk writes "The students at Copenhagen's new IT University will soon be guided by invisible, but talkative digital agents, known as ghosts or Disembodied Location-specific Conversational Agents. The ghosts are to compete amongst themselves for privileges such as better vocabulary or the ability to clone themselves. Ignored ghosts can die out completely. This project is a lot more serious than it sounds at face value - several papers have been published already."
..helping steer blind/disabled students around campus?
Very impressive text to speech technology, but I didn't see much in the way of a demo on the site given.
The grass is only greener, if you don't take care of your own lawn.
I mean, it doesn't have an AI (Microsoft instead opting to go with their proprietary Artificial Stupidity) but doesn't it somewhat seem like a prototype for this idea. I mean, an artificial helper that guides you around software isn't too different from one that guides you around a physical location. But in all seriousness, instead of location specific ones, wouldn't you rather have a personal ghost? You decide it's appearance on your PDA/Wearable Computer/Whatever, you adjust its personality via programming or learning capabilities. You get to the campus, and it wirelessly logs onto a local server, gets a layout, and comparing your schedule develops a path to where you need to be, and on demand (or wim, if so programmed) gives you directions? Sure, location specific ones are a neat idea, but personal ones seem like they'd be alot more useful.
1. Ghosts are mostly invisible or only vaguely visually manifested 2. Ghosts are often bound to a specific location which often has a very special relation to the ghost 3. Ghost owe their twilight status to some unfinished business and they are therefore active and striving 4. Ghosts only appear when called upon or if they feel an urge to manifest themselves
These ghosts sound a lot like the microsoft word paperclip. Is that damn thing going to start talking now?
Heh. But seriously, they have a lot in common - listen to these samples of background ambience to accompany the voice. Here's a bunch of nerds that actually understands how ambient sound can change the atmosphere completely.
For instance, in System Shock - the first time you hear SHODAN speak, she's just giving you a standard greeting ("Welcome back to Citadel Station.."). But all the time there's this evil drone in the background that starts low but keeps intensifying. SHODAN makes no threats, nor states anything evil - but that ambience, the drone, tells you that something's definitely wrong, and you should be very affraid:)
This has got to be one of the coolest things I've seen on slashdot. Not only is text to speech interfaced with most of the services of a fully wifi college ( the elevators, printers, music are accesible by ghosts ), but they've given each ghost a unique personality and history to boost! almost makes me want to learn danish and transfer to Copenhagen!
Will they play nice, or do what ever it takes, to survive?
How do you code competitveness? (or spell it for that matter)Free Mac Mini Yeah, it's
They have a good web designer and have clearly purchased the top of the line speech synthesizer (which has recorded canned audio clips narrating a few snippets of text for them)... they claim "all the voices you encounter on this site are generated by computers." Congratulations. Kraftwerk has been doing this trick since the 80's. Musical stings to provide ambience for different "ghost activities..." Little PHPbb posts about each ghost's personality that sounds like something cut from Starship Titanic's promotional materials...
- processing) that's been going on in countless CS departments around the world for decades...
The papers on their site that I've skimmed were extremely "light." They were at least suggestive of interesting ideas (albeit ones that have a nothing to do with AI and everything to do with human-computer interaction... "ambience," new ideas for interfaces, which seem promising or at least interesting). Their "main paper" is a 404.
So they're not exactly leading with the great breakthrough that makes their ghosts possible. Can anyone more familiar with the project comment? It looks like a lot of fancy dressing on the same kind of waste-of-time vanilla AI project (yet-another-unambitious-stab-at-natural-language
What's the real meat of this project? Have they really accomplished anything of interest from an AI or user interface perspective? Or is their main accomplishment an unusually skillful PR coup for themselves?
Want to Know How to Cheat the GPL? Read On!
Sure, it's weird to say that, but it's technically true :)
My thought EXACTLY. There are so many young people out there who don't know what has been done before. I just came from their web site, and the demonstration "interaction" between Alice and the supposed user reminded me STRONGLY of using ELIZA years and years and years ago. They have spruced it up, sure, and added good text recognition and speech, but it still seems to be based fundamentally on an Infocom-like adventure game text recognition system underneath. The bit where, and I paraphrase, she says she "appreciates his polite style, but it is ambiguous: please use what, where, when" instantly clued me in to that. Hell, I remember writing an adventure game engine in BASIC back in the 80s that did a decent job of parsing human text and figuring out what was wanted. I would have hoped that in the intervening DECADES, natural language recognition would have progressed a hell of a lot more than what appears to be the case. I'm sure in fact that it has, actually. Maybe just elsewhere.
I was at an ACM presentation not long ago, and some PhD was talking about this cool project he had done. Likewise, it looked functionally VERY similar to how one interacted with those old text Infocom games. Lots of the young kids ooh'ed and ahh'ed over it. I mentioned this similarity afterwards, and he actually got a bit irritated. I suppose people are supposed to have forgotten about that, so it can be re-invented in the 21st century.
Larry
Recent museum guides like at the Dallas Noeller Sculpture Museum use mp3 players with RFID readers. The mp3 gives random access sound loops, so you aren't tied to a sequential audio tape. The RFID tags on art works give you the location index.
This reminds me of the "guide" that the little girl has in William Gibson's "Mona Lisa Overdrive" .. it was a character that accompanied her everywhere to help her find her way around London.. If they could package these ghosts in portable devices that people could carry it would be essentially the same thing.
Once again sci-fi predects the future....
I really find this attitude disappointing, especially since its here on Slashdot. Although everything you say is technically correct, I'm not sure I understand what is the point of dissing an experimental project -- because it is experimental.
This is not a commercial product. Clippy is. And here is the big difference.
An experimental project like this is all about moving to the next step. The step where it becomes a reality. If you're dissing this project now, because a CS faculty is conducting a research project (which CS Faculties are supposed to do), which is actually interesting, and has potentially really great possibilities, then I'd hate to see how you expect progress to be made in computers at all.
Oh maybe you're just upset that it's not a Linux project.
Either way, I think this is really cool, and I love the evolutionairy aspect of it.
This is one project, I'll be keeping track of.
Right on. As I just wrote in a previous reply, the underlying technology involved in the interaction appears to be a mix of ELIZA and an Infocom-like adventure game text recognition engine using keywords, like what we wrote as kids in BASIC on our home computers back in the 80s. All the glitz on top, and the tie-in to other systems, appear to be the only real meat of the system. That is, of course, interesting. But the technology is certainly nothing terribly impressive, unless they've hidden the good stuff.
Larry
Rearranging the Soup of the Day sign at diners never gets old. There was a big bookstore across the street from my high school. They only built the sign 10 or 12 feet off the ground. I used to stand on top of my car and change the sign all the time. Having 2000 of your peers greeted in the morning with a sign that reads "Eat More Cock" was great.
Hint for young pranksters: Steal a page from the movie Sneakers. Recreate the target sign using Scrabble tiles. Rearrange the letters in private. Standing in front of a sign trying to think of a clever phrase will get you busted.
-B
This bring up the idea of an audible body language for these ghosts. Effectively background music or noise but once you got used to it you could use it to determine the ghost's mood, intentions, playfulness, etc. Whether or not the ghost actually has any feelings is moot as long as the interaction with the ghost can replicate more real human interactions. Ghosts might then have a auditory presense when they're around so that we can know that it's there instead of being surprised with a disembodied voice. An interesting question is would the ghost be allowed to change it's own "body language" or is it a fixed entity like our own presence in a room. And if it can change it's "body language" like we can, is it programmed that way from the get go, or does it have to learn it's own habits and change them accordingly. This could be an effective way to teach a ghost how to lie. The real question in all of this I guess would be what does PMS sound like?
I was able to get to the main paper (at least, I think it was the right one), but the focus seems to be on UI and the disembodiment thereof, not on any of the actual trivialities of interpretation and response. Next to last page has a sentence on the challenges of giving verbal commands to such a system (10 words a minute--not really a conversation).
And to slip in "personalities" and the genetic algorithm business just muddies the waters.
There's absolutely nothing in here about AI. I do think the UI stuff--the locality of personalities is interesting, so that might be somewhat original...
the rest is just so much fluff.
--Have a good night's sleep. Don't forget to brush your tooth.
Some of the pages have quite impressive voice inflections.
Does anyone know what they might be using to generate the voices? I poked around for a while but never found anything on what actually creates them.
Of course, this is a university campus, so by definition the "badly trained ghost" will be the MOST popular as people will be interacting with it two and three times to enjoy the joke, and they'll tell their friends about it as well. Meanwhile the "history of the lawn mowing patterns used in The Quad" ghost will 'self-terminate' a lonely, brief existence with but a wimper.