First of all, you can't solve big problems until you know that what are, source the data to analyze them and, most importantly, have the mechanisms in place to react to findings as quickly as you can analyze them. Potential applications mentioned include merging weather and financial data, for example, or other big problems like weather forecasting or optimization of worldwide distribution. A review in the New York Times suggested:

This is where I have a problem. If a government agency or a university or a massive data-driven enterprise have the time and desire to tackle problems like this, there is some merit to this. But an individual hospital? The phrase "if all goes according to plan" is telling. Let's face it, how many organizations have both the skill and organizational alignment to implement something so complex and controversial? In this particular case, medical data is a total mess and I, for one, would never accept the decision of a computer for medication, knowing that the records are often incomplete, conflicting and usually buried in proprietary systems. Even though the software can likely deal with outliers and errors and draw conclusions based on a large set of data, what if you're treating an outlier? There is zero tolerance for those kinds of mistakes in a hospital.

I don't like to use sports analogies, but I think one applies here. When you watch professional sports, you engage at one level, an extreme one. But when you play sports, what you try to do is hit the golf ball straight, not serve into the net or make a free throw about 30% of the time. The spectacle of professional sports drives our adrenaline to go do these things, but we never can approach the level of performance of the real pros. And that is the problem with the reporting of advances like System S. The vendor projects the image of Tiger Woods to a bunch of duffers, metaphorically speaking.

I applaud IBM for developing this capability, but I have some reservations about trying to sell it to the enterprise. James Taylor and I were clear in our book that automated decisions should be reserved for high-volume, "little" decisions that only matter to an organization in the aggregate.

E-MAIL  |