• IBM’s recent ads for Watson paint a picture of an analytics system that behaves more like a person than a machine. But, is this truly how Watson works?
• The answer is that Watson and its associated products like Watson Analytics are part man and part machine, blending human heuristics and domain expertise with AI and machine learning.
Over the past week, IBM has been blasting out a series of intriguing but puzzling commercials featuring its non-human, cognitive computing mastermind (literally), IBM Watson, hosting some intimate chit-chat sessions with Bob Dylan and Ken Jennings (the Jeopardy master himself), among others. Within these short question-and-answer exchanges, Watson expounds upon the meaning of life, recognizing faces, cancer treatments, and whether or not machines can understand sarcasm. Spoiler alert: According to Bob Dylan as interpreted by Watson, time passes and love fades.
That sobering thought aside, I was a bit frustrated by these ads, because they portrayed Watson as a truly sentient AI, something akin to the fictional smart machines of my youth, such as Star Trek’s M-5 or Stanley Kubrick’s HAL 9000. I do think that if plopped down into the fictional world inhabited by those machines, Watson would actually do pretty well. That is most evident when you consider one of Watson’s associated products, Watson Analytics (a cloud-based data discovery and visualization tool). With IBM’s growing IoT capabilities and its impending addition of time-series predictions to Watson Analytics, I do believe Watson could predict the failure of an AE-35 unit. But, it wouldn’t initiate an interesting conversation; and it certainly wouldn’t try to kill we humans (thankfully).
Rather, what Watson and Watson Analytics do is answer questions using natural language. ‘How’d our sales team do this quarter?’ It can recommend the best way to view that data. And it can even suggest potentially interesting correlations and previously hidden data points based not just upon raw statistical models, but instead using domain expertise ‘ human domain expertise. That is Watson Analytics’ secret sauce. And that is what I like about IBM’s newly introduced Watson Analytics Expert Storybooks service.
Expert Storybooks is slated for release early next year and built not for end users, but instead for partners, those well versed in their own domains like Deloitte, the Weather Service, or Twitter. The new service promises to embody human expertise within a data-driven narrative and empower those partners to sell that expertise.
My favorite example of this comes from IBM partner AriBall, which is building Expert Storybooks using the massive realm of data known as professional baseball. Led by the fellow who practically invented sabermetrics, which is widely used to evaluate pitchers, AriBall’s Expert Storybooks will help fantasy league participants, for example, see what’s changed recently with a given pitcher. Did his velocity drop the last few games? If so, why? Maybe he’s getting fatigued. Maybe he’s changed his approach for a given rival team. Or maybe cooler weather is the cause. How can a machine know?
Being able to pick out meaningful patterns and correlations like this takes a human with the ability to see the difference between those supposed reasons for the velocity drop. We humans are after all heuristic machines. When those heuristics and domain expertise are then coupled with AI and machine learning, the outcome might actually look a bit like a sentient machine. That’s the idea with Watson Analytics. Just don’t ask it how it’s feeling or to sing its favorite song. That’s the stuff of fiction and commercials