Remember Watson, the IBM supercomputer that beat lowly humans on Jeopardy in 2011? Fast forward a couple years and IBM is trying to turn Watson into a business. In fact, they're putting $1 billion behind it and giving it its own headquarters. Not bad for a computer. But how does Watson work and what can it really do?
Watson is a cognitive system, which means it's a computer that's designed to think like a human, only much faster. The key to Watson's uniqueness is its ability to combine three capabilities: process natural language questions (think of a more powerful Siri), generate and evaluate multiple possible answers, and learn as it goes from its successes and failures. Essentially, to think and answer like a human.
Using Watson's Jeopardy performance as an example, Watson first interprets the question, looking for key words and even puns that could mislead it. It then formulates multiple hypothetical meanings, searches its database for answers and grades the probability for each that it has an accurate interpretation and answer. If a given answer meets a specific probability threshold, it answers the question. It's big data analysis made efficient and incredibly fast.
Big data is all the rage currently, but data is messy. It's stored in myriad formats, various databases and incompatible structures. There are a lot of companies that let you do clever things with your data (like Tableau), but they're limited in both what data can be read and what can be done with it. Taking more extensive advantage of the data available to you (like with a Hadoop cluster) typically involves extensive data cleaning and programming.
If Watson is to live up to its promise, it needs to have more power than a Hadoop cluster but be as easy to query as a Google search. You could feed it a hodgepodge of data and then ask questions as you normally would, from simple to complex. Who are my top 20 customers this week? What else is someone buying when they put my product in their cart? Given these symptoms, what's the proper medical diagnosis? What soybean crop yields should we expect this year?
The problem is that Watson is so flexible and so powerful, IBM is obviously having difficulty communicating what it can do. They talk like enterprise-product engineers who want to play it safe and be as broad as possible in their description:
In an attempt to rectify this and generate some concrete implementations, they're trying something very un-IBM-like: letting outside developers leverage Watson's capabilities via a challenge. Submit your idea, make the cut and get access to the Watson API to develop your app. Pretty cool. But insufficient.
Fostering the development of a couple dozen Watson-based apps is not going to capture the public's or developers' imaginations, nor is it likely to generate the revenue that justifies a $1 billion investment. They need to tap the creativity of more than a few dozen developers. They need to encourage use case experimentation. The greater the access to Watson's API, the sooner they can show real-world implementations.
IBM needs to create a WaaS platform: Watson as a Service. Open up the API to everybody, provide a free development tier and then offer metered production tiers. You cannot hand someone a black box and ask them to build a business on it, while paying for the privilege. If you're intent on keeping it a black box, let everyone experiment with the box, discover what it can do.
The Watson team can also make the technology much more approachable. Instead of talking so much about Watson, let people take it for a spin and see what it is. Create a "Stump Watson" page and let people ask natural-language questions. Show the answer, along with a graphic illustrating how Watson arrived at the answer, like they did on Jeopardy. The buzz alone will draw developers to experiment with the API.
The flaw in the Watson Challenge is that IBM is only going halfway with the crowdsourcing of implementations. IBM's developers and product managers will be vetting the mobile app proposals and picking winners. If they want to be surprised and delighted by how Watson can be used, they should let developers experiment and not presume they know the best ways to utilize it.