A more serious version of this trade-off came up when an ethanol plant went out of business, and houses nearby found their basements full of water. It turned out the plant was such a big water user that it artificially depressed the water table—something the home builders did not take into account when deciding how deep to dig the basements. Now both the ethanol plant and the home builders were gone, and a bunch of homeowners were flooding the council chamber demanding to know what I would do to fix their problem. The city was not technically involved, but these homeowners had done nothing wrong, and it seemed we needed to do something to help.
Ultimately, we decided to re-create the effect of the depressed water table by pumping water into a ditch, to get the homeowners some relief. In the end I got lucky: a new operator took on the ethanol plant and began pumping again. But in the meantime, it seemed that we were faced with a problem that no amount of data could solve: Do we undertake a deliberate waste of water and energy costing every city resident a few cents, or let a handful of homeowners lose thousands of dollars of value on their homes through no fault of their own? Again, there was no technical answer to this problem; it was a question of who would suffer and how much.
The question of suffering brings me to one last concern around the use of data-driven techniques to bring about better government: the importance of exceptions, otherwise known as mercy. Efficiency, almost by definition, has to do with following rules and patterns; if there is an inefficiency within a rule, it can be ironed out by making a sub-rule. But sometimes our moral intuition just tells us that making an exception is the right thing to do, even if we can’t explain or defend the precedent.
I knew of a council member once who got a call from an elderly constituent asking him to help deal with a dead raccoon. But the animal was in the person’s yard, not quite in the street, where the city would be obligated to take care of it. The rule here was clear: too bad for the homeowner. But that didn’t seem like the right answer for someone who couldn’t handle it on her own. So the council member went to the home, discreetly dragged the carcass into the street, and then called it in so it would be taken care of.
Obviously this can’t be endorsed as a way of dealing with these problems. But without exceptions to rules, the world would be a colder and probably worse place to live.
ULTIMATELY, THE RISE OF MORE DATA and technology presents tremendous opportunity for cities to be smarter and more efficient in their operations—and therefore become healthier, safer, better places to live. But after taking office, just as quickly as I learned the power of data, I also learned to be mindful of its limitations, and aware of the problems it will not solve. And I learned to maintain some level of respect for the role of intuition.
Good intuition, backed by years of experience, can make it possible to sense things—like whether a trash customer is telling the truth when he calls to say you missed a pickup at his house or whether he just forgot to take the trash out—often with remarkable accuracy and sometimes without even being able to explain how it is sensed. There is great power in human pattern recognition, which actually resembles big data analytics in its most important characteristic: the ability to know things without knowing exactly how we know them.
Often, discussions of performance management gloss over this crucial difference between data analysis in general, and “big data” used with artificial intelligence. Using data in general is nothing new; it is simply the application of factual knowledge to make decisions. As an approach to government, it came as naturally to Alexander of Macedonia as it did to Robert McNamara. For the purposes of using data, the only thing to change with the introduction of computers is that we can gather and apply it more quickly and precisely.
“Big data” is different. It has the potential to change government, along with the rest of our society and economy, in categorically different ways than the use of data in general. Not everyone may share my definition, but to me the difference is this: Using data means gathering information, understanding it, and applying it. Using big data means analyzing information to find and apply patterns so complex that we may never grasp them.4
COMPUTERS CAN NOW CRUNCH sets of numbers so vast that the patterns that emerge from them are beyond the reach of the human mind—and yet the patterns can be used. Utilities like our waterworks are beginning to tap into computing capabilities that can accurately predict points of failure in water systems without our truly understanding how the prediction was made—only that it works.
Ironically, what makes this kind of predictive technology most interesting is that it so closely resembles human intuition. More often than we realize, humans rely heavily on knowing things that we can sense, but not explain. One of the reasons it was impossible, till recently, to program a car to drive itself is that many of the mental processes we use to drive a vehicle cannot easily be defined or described (and therefore cannot be programmed). Like the parks maintenance supervisor or road foreman going through countless decisions in the back of his head, anyone driving a vehicle relies constantly on subconscious pattern recognition. I can’t explain to you how I know the moment when a snowy road has become too slick for normal braking, or whether I am a safe distance from the centerline, or whether I can beat that yellow light. I just know. If I wanted a machine to gain this capacity, I would have to do one of two things: master the precise basis of this knowledge so it could be programmed, or construct a machine capable of learning it the same way I did.
The latter is becoming possible, and this is the true fascination of artificial intelligence. For government, there are extraordinary implications to a program that could anticipate road failures five years in advance, or predict asthma attacks using linked data sets on hospitalizations, weather, and car emissions to sense patterns so exquisitely complex that we will never understand them.
Now even explicitly social functions, like gauging how much anger a certain policy proposal will cause, can increasingly be achieved through social media analysis that might eventually outmatch even the finest-tuned human political antennae. The ultimate lie detector won’t be designed, like current ones, by programming known patterns in heart rate and perspiration. It will be designed by machine learning, scanning millions of recordings of people saying true or false things, and using this to make predictions based on combinations of indicators beyond our comprehension.
The algorithms advance every year. This is why Netflix has a good sense of what films you would like, perhaps even doing a better job of predicting your preferences than you do, if all you have to go by is a trailer and a description. But these capabilities are also still in their infancy. My Internet TV device still sometimes shows me commercials clearly intended for a middle-aged homemaker or a teenage video-game enthusiast.
And no matter how sophisticated the programs, they will never fully learn our sense of mercy—the rule not to be applied, the efficiency not to be captured. Capable of something resembling intuition but nothing quite like morality, the computers and their programs can only imperfectly replicate the human function we call judgment. Knowing when one valid claim must give way to another, or when a rule must be relaxed in order to do the right thing, is not programmable, if only because it is not completely rational. That’s why, even as reason has partly replaced divine intervention for explaining our world, it will not replace human leadership when it comes to managing it. A person aided by data can make smarter and fairer decisions, but only a person can sense when an unexplainable factor ought to come into play—when, for lack of a better expression, “something is up.” And that, as John Voorde might remind me, has been the job of elected officials all along.
4 I am aware of taking a liberty with the term “big data,” which is usually defined simply to mean data that cannot be processed without the aid of powerful computers. Here, I am talking about something more distinctive—the subset of big data analytics that involves the discovery and use of patterns beyond our comprehension.
12
Brushfire on the Silicon Prairie