I have been an opponent of the speech recognition movement for while now, albeit not a fervent one, but definitely not a fan. For me, it’s simply a matter of bad user experience. When I would encounter the technology, it rarely works according to my expectations and in fact, generally causes me frustration and embarrassment. First, I feel frustrated because it slows me down and impedes my forward progress. Whether it is cool, cutting edge or reminiscent of Star Trek, it doesn’t matter. I don’t like how the interaction seems to always turn into a self-service dead end, where I am yelling for an agent to get on the line to help. Also, because of reliability issues, I’m very tentative when using these systems. I anticipate that they won’t understand me, so I act slower than normal in order to ensure the system’s success, which is really pointless and against the goal of helping me get things done efficiently.
Secondly, I feel embarrassment because talking aloud to a computer system feels foolish and uncomfortable. I don’t want people around me to know who I’m calling and what I’m doing. It’s private. But even when I’m all alone, I still find it awkward to speak out loud to nobody. I don’t know what’s going to be asked of me next and this is unnerving as well. Yes, there have been advancements in speech systems so that they can be more conversational. However, what I have found, is that the more the system tries to be human, the bigger the failure. I wish the system wouldn’t try to be human at all because it so clearly isn’t one and can’t be one. When I interact with the system and convey my thoughts to it so that it understands me, the process insults my own human intelligence. Instead, make it plain and simple and unsophisticated. Despite the fact that I’m physically talking, it doesn’t feel like a conversation at all. And when I run into a dead end, there’s no Conversation Bridge to get me speaking with a real, live person who can truly help. For these reasons, the technology just isn’t useful part of an overall customer service conversation strategy.
Talked Off The (Speech) Ledge
When my VHT colleague and I were recently talking about his new iPhone 4S, I asked him if he liked Siri in sort of a smug and skeptical manner. He said he did and I said, “Why? It just takes you to a bunch of links on the Internet.” My friend’s response was rather interesting and probably not what Apple had in mind. He said, “For dead simple tasks, it works really well.” So I asked him for an example and he rattled off a few, such as making a phone call, setting a calendar appointment, playing a song and sending a text message. He proceeded to demonstrate and told me about how, admittedly, it takes a little practice. But that once you’ve figured out how to consistently repeat success, it’s quite convenient and handy, especially when driving. I found my friend’s demonstration quite impressive and I saw myself using these features one day. After a few more minutes of discussion I actually agreed with him that the simple stuff works well and could be useful.
So while I started out pretty negative about speech in general, my colleague helped turn me around…a little. He got me to the point where I’d agree that speech can be practical and useful…sometimes. Now, if I could just call all those companies where I know they have speech systems and practice with them for a while…then when I really need to get something done, I can call back and it’ll go quickly and smoothly. 😉
I share your frustration with most customer service uses of speech recognition. Most just replace a touch tone menu with a speech menu, not helping much if at all. This isn’t a fault of the speech technology, but of the application design. The first Macintosh was a failure; it took Jobs a second try to prove the effectiveness of the Graphical User Interface developed at at Xerox’s Palo Alto Research Center (PARC). There are a few excellent implementations of customer service that have customers telling agents, “Can I speak to your automated system?”
The key is what is behind the speech recognition. Speech recognition, in effect, produces text. It’s what you do with what was said that produces the efficiency. The key technology that Apple brings to Siri is Natural Language Processing (NLP); the speech recognition technology is actually licensed from Nuance Communications. The combination produces Speech Understanding, and it sounds as if your frustration is that the call center doesn’t understand what you want, so just sticks to a series of overly structured inquiries that are simple enough that they can be understood with no back-end technology. After all, the result of your inquiry is usually some data retrieved from or entered into some computer program, and a computer can do that much faster than an agent typing.
Speech recognition has gotten better over the years. Try some of the PC dictation software. It’s uncannily accurate, in part because it understands what words go together (e.g., it knows that “I know” is more likely than “I no” or “eye know.”) Speech recognition is used with almost all medical dictation today, even if a human checks it for accuracy.
Natural language has moved more slowly than speech technology, and is just passing the tipping point. And knowledge representation, translating “big data” into useful answers is moving ahead now (e.g., IBM’s Watson), providing a source of answers for your inquiry to your digital assistant.
Stay tuned–you won’t want to do without your assistant in a few years.
Thank you, William. Excellent post that sheds light on a technology that’s certainly heading in the right direction. I appreciate your comments and will continue to experiment with speech systems. It will be unavoidable…I’m a technologist at heart 🙂