IBM's Watson platform is being touted as the biggest paradigm shift to hit computing since we stopped using punched cards in the 1950s. Eschewing traditional programming, Watson is a cognitive computing platform that uses artificial intelligence to essentially think for itself. The system is capable of answering questions posed in natural language and is being embraced by various industries and businesses, including Australia's ANZ Bank. By all indications, the system is incredibly proficient at answering complex questions — but what about customers who don't know what they're talking about?
This week, we caught up with John Gordon, Vice President of IBM Watson Solutions during a Watson media briefing. The chief topic was how Watson can be used by businesses to engage with customers and answer their questions far more proficiently than a human employee due to the masses of data at its digital fingertips. The system, which famously won the US quiz show Jeopardy in 2011, is capable of learning from mistakes within seconds.
When it comes to answering complicated questions, Watson has proved is can comfortably outmatch even the most knowledgeable human employee. However, what happens when a customer asks a question that is just plain wrong or dumb?
As anyone who has worked in customer service can attest, consumers aren't always articulate when it comes to explaining what they need or want from a service — especially when technology is involved. One of the key strengths of a human service provider is that they are capable of steering a customer in the right direction to tease out what they're actually asking for. Can an artificially intelligent computer do likewise?
We posed this question to Gordon, who acknowledged it was one of the main challenges that Watson needed to overcome. One of the ways it combats this is by listing alternative low-confidence answers in addition to the response it thinks is the best match.
"That's one of the reasons I asked the [Watson] team to make sure we show alternative answers [to customer queries]. We've set this up so that when you ask a question you can open up other possible answers in one tap," Gordon said.
"The goal is simplicity of the experience: when customers see some of the other succinct short answers, it helps to shape the consumer's mind about what they really meant to ask and that helps them to move forward. So we basically help to provide continuous guidance."
In addition to dealing with varying levels of articulation, Watson also needs to identify and discard dubious information when pulling data from online sources.
"One of the things we've found is that answers to problems often show up on internet forums — but these aren't always helpful," Gordon explained. "For example, if someone asks Watson "how can I fix my iPhone?" it might find lots of answers on forums that say "throw it out and buy an Android!"
"The problem is that this is clearly an answer to the question, but that's obviously not what we want the system to tell you to do. So part of this understanding is being able to identify when it's a joke so we can start to diagnose what information is useful."
The Watson computing platform was official rolled out by ANZ Bank this week, which is one of the first companies in the world to take up the system.
"From a customer's perspective, I think it has the potential to make their lives easier, to do what they want to do from a bank perspective, and do it faster," Gordon said. Including inarticulate Luddites.