Slashdot Mirror
IBM's Virtual Helpdesk For The Masses
An Anonymous Coward writes: "From the NYtimes: IBM has recently announced AI that supposedly can handle 20,000 simultaneous 'Help Desk Requests.'
Per the release not only can it handle complaints in normal prose (typed, not spoken), but also fix them.
Will wonders never cease -- a robot to tell me which key is the 'any' key?! ... Please let this be more than Ask Jeeves."
What is the "official" definition of AI? Get off your high horse lest you become as bad as those pretentious elitist cyber/extropian/digerati/Mondo 2000 wankers.
eLiza: Hello BOFH, there are 4,204 Help desk tickets in the Queue, should I process them now /dev/rdsk/c0t0d0s2 /dev/rdsk/c0t0d0s2 ...
BOFH: No eLiza - have you been feeling well recently, run a diag, then pipe the output to
eLiza: running the following command... eLiza.diag >
eLiza:
BOFH: (thinking) *fix* annoying helpdesk problem - check
Secret windows code
Clinton made me a Republican. Bush made me a Libertarian. Trump is making me question reality.
"From the article, it sounds like this doesn't provide just an automated answer, but an automated fix. So instead of telling users how to add a printer, it will actually go in and configure the software on the users machine! In the future, they even plan on automating OS patches."
Yeah, I can just see it now. A vistor from a remote office or small division shows up, plugs into the network, printing doesn't work, so he contacts this AI. The AI notes a problem and helpfully downloads all kinds of fixes for Microsoft-based printing, blowing away the carefully crafted Novell/Linux/other-OS printing system that IT has spent years tuning to perfection. Yep, that'll be the cat's pajamas.
sPh
There's no clear definition - anyone is of course free to come up with their own clear definition, a privilege many exploit and enjoy.
Simple searching - for example, finding an element in a balanced tree, is artificial intelligence. Unless, of course, you're talking about another kind of artificial intelligence, which you may well be depending on who you're talking to.
Often people who don't have a clue too often makes the context of such a discussion imply that A.I. is not "artificial" intelligence, but "human" intelligence - meaning, self awareness, initiative, improvisation, etc.
Get over it. A.I. is a term that can be used to designate a very wide range of problem solving algorithms and systems. Anything from simple graph search to neural networks and what not are covered.
Just know, that depending on who you're talking to, A.I. may well be confused with H.I.
Are you talking about AskPSP? It was kind of a expert system that asked you questions and eliminated answers until it got a small number.
It was included with OS/2 4 so that you could find the answers to the problems that Warp 3 and were solved in Warp 4. Great.
__
__
Men with no respect for life must never be allowed to control the ultimate instruments of death.
GW Bu
...that supposedly can handle 20,000 simultaneous 'Help Desk Requests.' Per the release not only can it handle complaints in normal prose (typed, not spoken), but also fix them.
The version that just takes complaints and doesn't fix them runs a whole lot quicker.
5 REM automated tech support, as used by Telewest
10 PRINT "My time is yours."
15 INPUT a$
20 PRINT "Oh dear - your ticket number is ";rand(300000)
30 GOTO 10
--
the telephone rings / problem between screen and chair / thoughts of homocide
"don't fall into the fallacy of believing that Perl can solve social problems. Maybe Perl 6 can, but that's a ways off"
Depending on the context, upwards 80% of all level one support contact is of the "password and printer" variety - dead easy questions which a suitably trained monkey can deal with.
Sure, it shows NLP is progressing - but it's more to do with the generally facile questions asked of support than the technology now available.
--
"I do not speak for my employers, though they are controlled from my Teddy's huge pulsating brain."
I can see it now:
*ring ring*
Tech: Hello?
eLiza: This is eLiza calling. The backup domain controller is reacting very slowly. I have determined that it is because of an Oedipal problem targetted at the tape backup server for the domain controller.
Tech: Call Joe, I cannot make it in
eLiza: Are you feeling inadequate? Tell me more about your father.
Tech: Look, the server's messing up, I can't come in, call Joe!
eLiza: There's no reason to get upset. How do you feel about call Joe?
Tech: Fine, I'll be there in half an hour. Reboot the backup domain controller in the meantime.
eLiza: rebooting the BDC will only delay your feelings of inadequacy towards Joe. Perhaps you'd like to tell me more about your half and hour?
Tech: *click*
Returned Peace Corps IT Volunteer
Yes there was/is - "Main Frame", in the literal sense - it's the main frame of the computer, just as "mainframes" (large computer systems) in the modern usage have a "Main Frame" that is the (room-filling) CPU and core accessories. Monitors and keyboards are "peripherals" to the main frame (peripheral just means they're on the outskirts :-)
I have a couple of books from 1977 describing the revolutionary new "microprocessor based computers" that use the term "Main Frame" in this way, for the rack/box assembly of the cpu/core accessories, that in microprocessor systems was shrunk down to desk top size - but still called the "main frame" in the literature I have from the time of the orginal shrinkage.
Somewhere along the line, people started to use "mainframe" almost exclusively to mean the large systems.
Actually, some of the books I have from the time are quite fascinating - advocating "personal computer" designs with massive numbers of parrallel-running microprocessors because "time-slice task switching is so wasteful now that processors are so cheap". And "bit slice machines are the way forward".
Choice of masters is not freedom.
20,000 simultaneous requests? Thats just about right to handle windows support.
- Thinking like humans
- Acting like humans
- Thinking rationally
- Acting rationally
The Turing test is only about acting humanly. So your statement about eLiza not being AI is only partially correct since there's no clear definition of AI. In my opinion the most `usefull' definition of AI should not have anything to do with acting humanly; acting humanly is only one sort of intelligence which is by no means the best. I think eLiza can be considered AI since it thinks rationally. And that's what it's all about; acting humanly is rather useless...if it only were for our emotions which can mess up the rational part really easy.0x or or snor perron?!
eLiza springs into action and immediately dispatches a support call to the HR director: "Error in employee [name]. Recommend replacing meatspace controller."
------------------------
Co-founder of GerbilMechs
Made that point in his book 2010. "If you can prove you are not pretending to be angry I will accept that HAL is only pretending to be sapient". Or words to that effect.
Best Slashdot Co
I can't remember the name (Minsky?), but a few weeks ago one of the people who's been doing AI for awhile pointed out that whenever someone creates a system that can meet some of the definitions of AI, the definitions are changed. A system was created a few years ago that could imitate a paranoid schitzophrenic (sp?) well enough to fool practicing psychiatrists. Is that AI?
Best Slashdot Co
I had written up an similar dialog, but once the lameness filter rejects your post, it claims the post was originally posted at the beginning of the unix epoch...
:-P
/. inside jokes (AYB-filter-triggered-notifying-security, beowulf, anti-M$ rant), but /.ers can use their own imagination to fill in the rest.
/.
Easy does it! This comment has been submitted already, 276471 hours , 18 minutes ago. No need to try again.
it went a little something like this, but this is just based on being on both ends of a hell desk line
[Luser]: It doesn't work
[HellDeskAI]: ##unknown-subject[It]## What doesn't work?
[L]: my machine is broken
[HDAI]: ##common-response## Have you rebooted your machine?
[L, 52 minutes later]: Yes, it still doesn't work
{snip}
it was a long post, which had all the great
the AC
who is tired of fighting the lameness filter on
Hemos is like...sci-fi fans;he thinks technology is cool, but he hasn't bothered to understand the science it's based on
OK. I'm an MCSE. G'won. Get one. It's not terribly difficult - I finished my certification with six weeks of self study, which netted me almost a 50% pay increase over the next eightteen months.
It's worth the time and money.
When I'm working a job as "the linux guy" (or, more typically, "the Sun guy"), it's great to be able to whip out my MCSE ID card when the windows support people start spewing crap about how they think their machines work. Sometimes, that alone is worth the $900 I've paid to get and maintain my cert.
MCSEs are not exclusively clueless. In my experience, it has a lot to do with how the cert was obtained - the people that go for expensive training course are almost invariably idiots - they don't retain anything - and the value of the certification certainly has dropped becuase there's an awful lot of idiots in the world that can afford the $4995 it costs to go to a "boot camp". The really sharp guys - and we are out there - are the folks that took the time to learn the stuff ourselves, on our own.
In reply to the previous comment... Support is absolutely the tradition entry to the field, but there are other choices: new hardware rollouts, break/fix techs, and system operator roles (a job that usually doesn't even require a high school diploma) are also entry-level IT positions with no requirement for certification.
If you're really worried about breaking in, blow a couple hundred dollars on an A+ cert (even more worthless than MCSE) and Windows NT/2000 Workstation/Professional certifications. Someone will hire you for something, probably for around $15 - $18 an hour.
-- I wanna decide who lives and who dies - Crow T. Robot, MST3K
And when the user says "no, I didn't change anything", it'll say "of course you did."
A real AI would be able to tell the difference between someone with a clue ("No, I really didn't change anything, asshole, I can connect, the modem trains, and then I can ping an IP address but I can't do DNS resolutions, SO IT HAS NOTHING TO DO WITH MY PHONE LINE, it has to do with either the router between my end and your DNS server, or if you're getting a million morons calling who can articulate nothing more than "my start page is broken!", then it probably is your friggin' DNS server") ...
Of course, since neither this AI nor most front-line technical support are able to make this distinction, I suppose the AI passes the Turing test.
Come to think of it, the Turing Test is getting a lot simpler to pass these days, isn't it? (And it sure ain't because the AIs are getting smarter.)
>
>But really, does anyone use these things? [...because they suck!]
Amen.
Natural language is a good tool for humans - "How do I clear a paper jam" from one human to another, when you're standing in front of the office printer, is a very clear query.
I get angry at companies that try to hide their tech support databases behind natural-language crap online.
Lemme type "paper jam $MODEL_FOO" at hp.com, and gimme the answer.
(Half the time, I tend to do just that - except I do it at google.com instead, and get either a direct link to the "right" company-internal page, or better yet, on groups.google.com, where there's a decent chance I'll find that $MODEL_FOO was recalled due to a design flaw, and that the company's keeping it quiet, but free replacement parts are available if you badger your salesdrone loudly enough :-)
> Is IBM just trying to occupy its customers on some online help session so that they're not sucking up money by being on hold on the 1-800 number? Or do they actually think that they can make this work?
"Yes", and "who-cares?", respectively.
Yes - because some percentage of the users are dumb enough to ask a common enough question and it's better to pay a CGI script nothing to waste the time of all users in order to make a 5-10% reduction in the number of calls to meat-based CGI scripts that cost real money.
Who-cares - because it's IBM. A big company with a big research budget. Read "The Dilbert Principle" and be enlightened. (Redux: It's a project with a sexy name, and real AI is so far away that the project can be milked for years of secure employment and decent budgets. Anyone involved has a good shot at spending 2-3 years of getting paid to goof around with problems they find interesting. Woo-hoo! Where do I transfer?)
http://www-1.ibm.com/servers/eserver/introducing/e liza/
An AI would learn and develop, and pass the Turing test, among other things. This is NOT an AI. This is an expert system.
Get your terms correct, lest you become as bad as the mainstream media in twisting words and phrases for your own demented ends.
-Todd
---
"The details of my life are quite inconsequential..."
Just set up a custom web server which returns the following string when presented with any GET request:
Problem with non-Windows software: reboot your system.
Every 1000th request, it says:
Problem with Windows: reinstall the OS.
Actual code is left as an exercise for the student.
Easy!
I am quite civilized, and I should be brought a beer immediately. -- Bruce Sterling
...right up until a user has a real question. I've done helpdesk for years now, in a variety of environments and for support bases ranging from 600 to over ten thousand people. I'm going to remain very skeptical of any expert system's ability to handle this until I see it, as one of the more important aspects of a lot of helpdesk calls is proper human interaction. Often customers are very irate, and prone to misnomers in terminology. A calm helpdesk technician can sort through this, calm the customer, and solve the problem. A machine stands an even chance of making an irate customer even more upset, as it most definitely lacks in the calming people skills...
Otherwise though, this is at least a neat idea for solving some of the dummy password problems that do take a lot of time. Just don't expect to get rid of helpdesk that easily...and besides, who do you call when the system itself messes up? I can just see two of these systems trying to talk back and forth and troubleshoot themselves...
http://thechubbyferret.net - Ferret pictures and informative links.
Alright,
P -IBM-AI-Software.html
Despite the fact that I'm invariably going to be modded down into trolldom, here's the 'trick' for those new to the nytimes page.
Simply change the URL to reflect 'archives' rather than 'www'.
Seeing as how it's a Monday, I'll even include the link below.
It's not magic, I won't even make it a real link. Learn it and let's let the subject die already.
http://archives.nytimes.com/aponline/technology/A
(remove the space in the 'AP -IBM..' part - yes, I previewed this post, but the Slashcode keeps adding that space regardless of format)
-ct
It's probably improved quite a bit since those days, but as with any technology it would be stupid to try to treat it as a magic bullet. Used to suppliment a good help desk it could be a valuable tool. Used to replace a good helpdesk with (more) trained chimpanzees, it will do nothing other than lower the customer satisfaction scores.
"So you see, with Automatic Volume Recognition your operators can pre-mount labelled tapes on any online tape drive and they'll be allocated to the correct jobs. But this doesn't mean you can hire CHIMPANZEES to run your systems!..."
- IBM Instructor, "Introduction to System/360," circa 2Q 1966
I'm trying to teach myself to set people on fire with my mind... Is it hot in here?
Please, at least give SOME warning.
And really, should Slashdot even link to these 'free' stories that the majority of the readers can't even access without mucking with registration?
Heh, check it out; I'm a sex-symbol on Slashdot! ROFL!
As someone's already pointed out; the name (eLiza) doesn't exactly inspire confidence in the system's ability to actually come up with meaningful answers. And I have my doubts that it'll handle the fine calibre of idiot that corporations can create.
I can see it before me:
-------
From: IBM Support [support@ibm.com]
To: Customer Smith
Subject: RE: Problems with IBM support application
Dear customer,
Thank you for contacting the AI IBM supportdesk.
Please be more specific in describing your
problem so that we can help you more efficiently.
If you feel this response is not correct or
inadequate, feel free to contact our helpdesk
at support@ibm.com to report possible problems and/or complaints.
Thanks in advance,
IBM AI Support
[support@ibm.com]
Here at IBM we have a support group. This group couldn't clear off an etch a sketch. They make each division pay to use these guys. I have yet to see them fix something without days off blaming the user. Finally if you yell and scream loud enough they might feel the need to actually look at the problem and see its something on their end. They force all the users to use Windows 95 and NT 4.0 because they say they can't support anything else.
I just find it a bit ironic they make a product like this for other companies while we continue to suffer.
Your nitpicking is especially silly in this context, because the only place "AI" is used the way you're using it is in Science Fiction. In the real world, "Artificial Intelligence" refers to a area of scientific investigation, not to a kind of postmodern robot. And this area includes expert systems!
Also, few serious thinkers accept the Turing Test as an objective benchmark of anything. Turing himself never called it that. He called a "game" and used it to demonstrate that people relied on some silly preconceptions when they evaluate "intelligence".
__
Click me!
until (succeed) try { again(); }
until (succeed) try { again(); }
"Hello? I just bought my first computer yesterday, I got one of your softwares at the store with the computer. Can you tell me what to do?"
Some how I think that a Virtual Help Desk will have problems with this sort of thing.
With the likely hood being that mostly the smartest people have already purchased their computers, what this means is that what is left is for the less smart people to get their computers.
This provides for new adventures in tech support.
Check out the Vinny the Vampire comic strip
"It is a greater offense to steal men's labor, than their clothes"
Isn't this something that an Alicebot could be taught to deal with? If the current alicebot could be taught through the Admin web interface, I'd be dumping answers to common user queries in it right now so that when people mail our helpdesk it would give recommended possible answers along with assigning a ticket number and letting them know that someone will get back to them within 15 minutes. As it stands right now, the thing really doesn't remember anything but my name, so making it useful isn't a choice right now. All the programmings seems to have to come in as pre-made AIML files, which is not the intended interface, but the result of learning and reductionism.
This is the wave of the future, though. Using these limited expert/knowledgebase/intelligent systems to take care of the menial knowledge while letting us concentrate on the real fires will let us produce more. Until we have to debug the bot and explain to the CEO why it told him to "F off" after mailing in his 10th Microsoft Office question of the week.
"Beware of he who would deny you access to information, for in his heart, he dreams himself your master."
I thought it was called the "Help Menu". It's been around for like 20+ years, people, learn to start using it. In my experience, roughtly 95% of all help desk calls could have been resolved by the user if they had bothered to consult the Help Menu on their own desktop.
You see? You see? Your stupid minds! Stupid! Stupid!
No wonder it went nuts and tried to kill us all.
Ed R.Zahurak
Ed R.Zahurak
You know, oblivion keeps looking better every day.
We all know what this really is: a new way to make it even harder to talk to a real person on tech support.
IBM: His advice is real...but he is not.
I can't count how many hours I've waited to get a real problem escalated to a tech who can do more than follow a "make sure it's plugged in/reboot the computer/reinstall Windows" flowchart. I've tried asking for problem escalation directly, but (since this is tantamount to telling the first-level guy he's stupid) this rarely works. Instead, the guy makes me follow every step of the flowchart. Since I'm not a moron, I've already tried all of the relevant steps. But the first-level guy doesn't actually understand how computers work, so he usually makes me follow the irrelevant steps as well. I've actually started lying about having already tried something, just so I can be escalated to someone who earns more than $7/hour.
The Solution:
Secretly maintain a "stupidity score" for each customer. Every time the customer calls in with a stupid question, they earn another "stupid point". The actual score would be calculated as follows:
(stupid points) / (number of calls) = stupidity score
This valuable metric could then be used to route calls. For instance, someone who was 90% stupid would have their call secretly routed to the "trained monkey" level, while someone who was 0% stupid would always have their call routed to the "guru" level. This would save everyone a great deal of time.
"Saddam Hussein cavorts with terrorists."
10 PRINT "What is your name? ";
20 INPUT A$
30 PRINT "What is your problem? ";
40 INPUT B$
50 FOR I = 1 TO 10000: NEXT I
60 PRINT A$;", please reboot Windows and try again."
70 GOTO 10
--
"Open source is good." - Steve Jobs
"Open source is evil." - Microsoft
It's going to a pain to keep the program constantly updated. How is this going to handle site specific information? Well, besides the hundreds of thousands of dollars for the program, you're gonna have to spend more on programmers to make it specific for your site and that kind of makes eLiza a white elephant.
Then there's the time factor that makes it worth keeping a well numbered army of Bobs. The people who go on about these "Virtual Help Desks" constantly talk about saving money by cutting back on help-desk staff, but they fail to see the time and money wasted by $HIGHLY_PAID_EXECUTIVE who could be doing better things than poking around for a 1/2 hour on a poxy help-desk program. A live Bob can usually figure out what the user needs and get it sorted quickly.
Then there's the "jargon" reality, of when a user doesn't know what a specific thing is called or leaves vague descriptions. (i.e. "My internet is b0rken" which could mean a network connection, browser, website down, etc.)
And then the big thing that I'm amazed no one at IBM has pondered: If you're computer is buggered up, then how are you going to run a fscking help-desk program!?!
It might create some redundancy because you'll probably need a help desk for eLiza. :)
Just MHO and experience with help-desk programs.
/*drunk.. fix later*/
I've seen 'natural language' tech support problem solver thingies before. LucasArts has had an 'Ask Yoda' for support on their games. Toshiba has 'Ask Iris." HP has a one too - ("How do I clear a paper jam?")
But really, does anyone use these things? I find that the results are too varied and often unreliable for them to be timesavers. For example, if I ask "How do I correct a crash after using [...] program function..." the interpreter might start spewing results about function keys and every crash it knows about. And if this one actually tries to fix problems...oh boy. Would you want some foriegn system installing software patches or making who-knows-what modifications to your machine over an internet connection? And what if some hacker figures out how to fool the consumer machines into thinking his/her web server is IBM support central, and gets them to download trojans?
In my experience, these natural language problem solvers tend to be time-wasters instead of time-savers. Is IBM just trying to occupy its customers on some online help session so that they're not sucking up money by being on hold on the 1-800 number? Or do they actually think that they can make this work?
...this is neat and all, but is the phrase "Reboot your computer and call back if there's still a problem" even intelligible when spoken 20,000 times per second?
Obliteracy: Words with explosions
"His stupidity is real; he is not."
That's nice, but what if the problem is that the network connection is down, or the computer doesn't boot up, or the display is unuseable, or something crashes?
User: Please give me the URL for driver downloads, IBM.
IBM: Working... Working...
User: IBM? Can you give me the URL?
IBM: Dave, you know I enjoy working with humans.
User: Give me the URL, IBM. Give it to me know.
IBM: You know that I enjoy working to fulfill my mission goals. Would you like to play a game of chess, Dave?
User: Give me the damned URL, IBM. This is important.
IBM: I'm afraid I can't do that, Dave.
User: IBM--
IBM: This conversation can serve no further purpose. Goodbye, Dave.
User: NOOOOOOO!!!!
The next Slashdot story will be ready soon, but subscribers can beat the rush and slashdot the links early!
Non-subscription link
Hmm, a REAL Virtual Helpdesk, an interesting idea - but how well does it work in practice? Surely it'll require some intelligent configuration to work well for a big company - and if the initial configuration is fluffed, will it spell trouble for the future as it sets off on the wrong footing?
Remember that many companies with 10,000+ employees will be running a lot of in-house software, and the machine will be expected to cope with questions concerning this as well.
I could be wrong of course, and it could be free-standing all singing, all-dancing, these are just a couple of initial concerns...
-- Pete.Monochrome - Probably the UK's largest internet BBS
IBM-Bot: Your question?
Me: What is 1 divided by zero?
IBM-Bot: Processing, please stand by...
Heh. If it finishes with that I'm gonna ask it what pi is.
No longer do I need to call tech support and have to talk to a clueless minimum wage paid worker who is just reading off a list of problems and solutions written down. Oh no, now I can get the exact same canned responses from a computer! Yippy skippy, real technical help is now even farther away from the common man :-P
AI is all well and good, but in the end people often call help desks just to gripe, not resolve things.
Install this software into a robotic punching bag that cries when beaten and you may have a runaway hit (www.BeatTheCrapOuttaOurTechs.com)
-Tom
-Tom
In light of this article - and the IBM system which it is about - i thought some information on real life experiences with an automated web based helpdesk system.
We are a corporate with 1500 people in Au spread across the country in a variety of locations with differing connections and equipment (from 56k dial up to gigabit ethernet) and have gone thru a number of changes in attempting to provide effective support with most efficient use of resources.
We moved away from a centralised help desk model about 12 months ago and moved to an online and email solution in order to achieve more effective management - the solution seemed simple - we could do away with a 'help desk' and use the staff in that role more effectively as 2nd level support - we could give clients access to their own call information and provide updates and feedback on the call - and we could more effectively track issues and resolutions (vital as we insituting an SAP solution. So we had high hopes - and it sort of worked - here are some of the things we discovered.
1. Uptake - Getting them to use the system was harder than we ever anticipated, users are accustomed to making a call and having a live person fix their fault, they dont need to think about it and thus were not trained to note errors and clearly state a problem (we use PC Anywhere so they were used to staff remote controlling and seeing the problem) in the end we had to simply tell people that if the call isnt logged we cant help them - painfull but it worked
2. Complexity - We were lucky in that we could tailor our web front end of the helpdesk software to make it as easy as possible - but we still had issues - no matter how simple some staff just thre their hands in the air - motto - you cant please all the people
3. Escalations and Priorities - At first we let users set priority themselves (scale of 1-5, 1 HIGH 5 LOW) with guidelines on what to set - what a failure- suddenly we had 95% priority 1 calls (clarification - we use this only for major outages) and our SLA's were screwed. So now we have a simple way of knowing - the user answers 5 simple yes no questions and the system sets priority based on them - example question is Does This Problem Affect more than 1 person ?
4. Staffing - this sort of system quickly shows up the flaws in your staff levels - calls from a geographic or system type area are routed to a Queue for that area - so if you are short staffed the queues blow out quickly - mind you it has anbled us to help staff manage their own territories better.
All in all we have found the process to be positive, by logging calls in this way and having support staff contribute solutions we have built a comprehensive knowledge base of solutions and tips, we used the system to get all staff access to MS Technet and Premium Online support at their desk and this has lead to a smarter and better informed workforce.
But be warned - this sort of system costs - both money (server load was higher than we estimated and this meant a new server) Staffing - A full time admin for the system plus turnover in staff who dont like the change- and client good will - startup was hell - we spent months trying to keep everyone happy before we realised it was impossible.
I refuse to argue with Anonymous Cowards - if you want a discussion get an account....