“Hey, Watson! Is Santa real?” – Why IBM Watson is an innocent 6-year old…
I love the technology behind “IBM Watson“. I think it’s been a long time coming and I don’t doubt that in a matter of only a few years, we will see phenomenal applications for it.
Craig Rhinehart explored some of the possibilities of using Watson to analyse social media in his blog “Watson and the future of ECM”. He also set out a great comparison of “Humans vs. Watson”, in the context of a trivia quiz. However, I believe that there is a lot more to it…
Watson is a knowledgeable fool. A 6-year old kid, that can’t tell fact from fiction.
When Watson played Jeopardy!, it ranked its possible answers against each other and the confidence that it understood the questions correctly. Watson did not for a moment question the trustworthiness of its knowledge domain.
Watson is excellent at analysing a finite, trusted knowledge base. But the internet and social media are neither finite, nor trusted.
What if Watson’s knowledge base is not factual?
Primary school children are taught to use Wikipedia for research, but not to trust it, as it’s not always right. They have to cross-reference multiple research sources before they accept the most likely answer. Can Watson detect facts from opinions, hearsay and rumours? Can it detect irony and sarcasm? Can it distinguish factual news from political propaganda and tabloid hype?
If we want to make Watson’s intelligence as “human-like” and reliable as possible, and to use it to drive decisions based on internet or social media content, its “engine” requires at least another dimension: Source reliability ranking. It has to learn when to trust a source and when to discredit it. It has to have a “learning” mechanism that re-evaluates the reliability of its sources as well as its own decision making process, based on the accuracy of its outcome. And since its knowledge base will be constantly growing, it also needs to re-assess previous decisions on new evidence. (i.e. a “belief revision” system).
Today, Watson is a knowledge regurgitating engine (albeit a very fast and sophisticated one). The full potential of Watson, will only be explored when it becomes a learning engine. Only then can we start talking about real decision intelligence.