Profile

unixronin: Galen the technomage, from Babylon 5: Crusade (Default)
Unixronin

December 2012

S M T W T F S
      1
2345678
9101112131415
16171819202122
23242526272829
3031     

Most Popular Tags

Expand Cut Tags

No cut tags
Saturday, November 14th, 2009 11:51 pm

It's not so much the strip, as the footnote.  Computer voice recognition is so bad that whenever I find myself talking to a robotic voice-recognition menu system, I immediately just start repeating "Just give me a human" until it gives up and forwards me to a human.  If I'm calling one of several potential sources for something, and I hit a voice-recognition menu system, I'll hang up immediately and try the next supplier.

Companies and government agencies deploy computer voice-recognition menu systems because they're convinced it'll save them money and let them reduce the number of humans they're paying.  But at the current state of the art, what it mostly does is lose them customers who are sick and tired of dealing with artifical stupids, and frustrate and enrage the people who don't have a choice about whether to use the AS.

Tags:
Sunday, November 15th, 2009 07:53 am (UTC)
Actually, it does save organizations money. It doesn't recognize voice commands as well as a human can, but it recognizes most of them most of the time, which means an entire tier of support people can, uh, go look for more interesting jobs.

Sometimes that means the difference between providing a service and not providing it. The SF Bay area's 511 service uses voice recognition and can be maddeningly stupid sometimes, but most of the time it gives you up-to-date road information, for the routes you're interested in, for free. If it sometimes mistakes "Berkeley" for "Oakley," well, we're none of us perfect, and human operators can be annoying and stupid sometimes too, and if 511 had to employ humans, it wouldn't be there, not for free at any rate.

But that wasn't what made me want to post a comment. I wanted to mention that pattern recognition in both audio and video is one thing humans still do better than computers -- witness www.galaxyzoo.org/. 400 years of computer technology meets 65x10^6 years of primate wetware evolution -- and loses.

I find it comforting, myself. A cheap calculator can calculate a square root before I've even found a pencil; my Palm beats me at chess; but damn few computers can distinguish at a glance a spiral from an elliptical galaxy.
Sunday, November 15th, 2009 07:18 am (UTC)
A long time ago, I adopted my own defense against the artificial "stupids". As soon as I realize that a number I called is being answered artificially, I hit "0", and every time I detect another layer of artificial stupidity, I hit "0" again. This will nearly always lead to a real human being. My big gripe is the time wasted, because the only phone I have is a cell.
Sunday, November 15th, 2009 09:55 am (UTC)
I've found this website to be rather helpful in bypassing the AS: http://gethuman.com/
Sunday, November 15th, 2009 02:55 pm (UTC)
I'll second the vote for http://gethuman.com/ -- the simple EXISTENCE of such a website is an indictment of the VR AS mania.

I need to find an actress friend's blog entry on this. The ONLY way she has found to get the various VR AS systems to process her request was to use her "Illiterate, Drunken Hillbilly #2" voice. (The fact that she has more than one such voice in her professional arsenal terrifies me...) She found that the more distinctly she spoke, the less accurate the idiot machine proved to be -- "Prim, Proper English Professor" proved to be the least successful, with "BBC Standard Received" a close second.

Since she and I exchanged views on this, I've cultivated an "Uneducated Inner-City Man" voice for VR AS systems on the East Coast. My success rate in navigating VR AS systems has improved markedly.

While she and I are only a pair of data points, the implications are far from encouraging.
Sunday, November 15th, 2009 05:47 pm (UTC)
Mrak, Alaric is English...
Sunday, November 15th, 2009 06:35 pm (UTC)
I know. :) I was pointing out that the VR systems are skewed toward "uneducated American".
Sunday, November 15th, 2009 10:10 pm (UTC)
I was thinking more along the lines of you had ~3~ data points... ;-)
Sunday, November 15th, 2009 06:59 pm (UTC)
So the American Airlines frequent flyer reservations number asks you a few questions before sending you to a human -- domestic vs international and your frequent flyer number, so you get routed to the right department. It works pretty well.

But the frequent flyer number recognition system works GREAT. Specifically, it understands the phonetic alphabet, so you can tell it "two seven lima november six five three" and it gets it. This is about plus eleventy million (of course it's a much simpler problem than ad-hoc recognition).
Sunday, November 15th, 2009 11:17 pm (UTC)
Someone was thinking. :)
Sunday, November 15th, 2009 10:57 pm (UTC)
I've found that any time I get an automated system that expects me to respond in spoken voice, if they present a list of options ("To check your balance, say 'Balance'...to make a payment, say 'Payment'..."), I just hit the number that corresponds to the list position of the item I wanted.

I haven't come across an automated system where that hasn't worked.
Sunday, November 15th, 2009 11:20 pm (UTC)
Recognition is only part of the problem. After recognition comes the problem that it doesn't matter how well the system recognizes the three choices "apple", "orange", and "banana", if those are the ONLY three options, what you wanted was a bagel, and the system offers no way for you to talk to a human who can comprehend that you're trying to get help doing something that's nobody thought to put on the menu. (Hitting 0, the usual escape, did not work.)
Tuesday, November 17th, 2009 05:10 pm (UTC)
Nothing to do with voice recognition, but the real problem is that phone trees are stupid in the first place. Putting a voice recognition overlay on top of the underlying phone tree only makes it more stupid.

My company's CEO *hates* these things. His philosophy is that we should answer every single phone call with a human and that human can figure out where to send your call (and this human should be located in the US or UK for the most part, but that is a separate issue).

If you call in and all the Customer Service people are busy, it rolls over to a different group of people until it finds a human to pick up the phone. For instance, the sales line eventually rolls over to the cell phone of the VP of Sales. It is rare that a call ever makes it that far before someone answers it.

Having said all that, the single most annoying thing about having an automated system is how none of the information you tell it makes it to the eventual person you talk to. I can't count the number of times I've told a phone system an account number or something and then had to repeat it to the eventual agent. This is not rocket science and modern phone systems can handle this kind of thing, even across multiple locations. (my kinda crappy little phone system takes care of support/sales being located in Campbell, Ann Arbor, Atlanta, Fresno and the UK just fine).

Tuesday, November 17th, 2009 05:23 pm (UTC)
My company's CEO *hates* these things. His philosophy is that we should answer every single phone call with a human and that human can figure out where to send your call (and this human should be located in the US or UK for the most part, but that is a separate issue).
I'm 100% behind your CEO on that.
Tuesday, November 17th, 2009 05:33 pm (UTC)
Oh yeah...he's crazy, but it is very easy to get behind him on this issue.