Data Decisions: You Can’t Be Held Responsible (Or Can You?)

In the wake of the 2009 earthquake in L’Áquila, Italy, seven Italian scientists were convicted of manslaughter charges for not adequately communicating the potential threat.

Reading Time: 3 min 

Topics

In the wake of the 2009 earthquake in L’Áquila, Italy, seven Italian scientists were convicted of manslaughter charges for not adequately communicating the potential threat.

Image courtesy of Flickr user Darkroom Daze.

Back in 1633, Italian mathematician, physicist and astronomer Galileo was sentenced, as part of the Roman Inquisition, to lifelong house arrest. His crime? Suggesting that the sun, and not the earth, was the center of the universe.

Fast forward nearly 400 years and scientific opinions are again embroiled in controversy.

In October, a group of seven Italian scientists were convicted of manslaughter charges for not adequately communicating the potential threat of something they couldn’t scientifically predict: the deadly 2009 L’Áquila, Italy, earthquake.

The fact that this group of highly regarded scientists has been sentenced to prison for providing what ABC news called “inaccurate, incomplete and contradictory information” — for giving people a false sense of security — raises the question: Could there be implications for organizations that utilize data science to guide decision making?

In other words, what happens when a strategy goes awry, a product flops or investors lose money? Can data analysts be held accountable?

Data mining newsletter KDNuggets asked its readers the question: Should data scientists/data miners be responsible for their predictions? Its results:

  • 45% said no, data scientists/data miners should not be held responsible for their predictions;
  • 37% said data scientists/data miners can be held financially responsible, if they also benefit from correct predictions;
  • 5% said data scientists/data miners can be held criminally responsible for wrong predictions; and
  • 13% weren’t sure which direction the wind was blowing

Experts that I spoke with agree on two points. First, that basing a decision on data is no guarantee that a strategy is going to work, though it will typically mean fewer failures. And second, that basing a decision on data, where data is available, beats intuition alone.

Andrew McAfee, principal research scientist at MIT’s Center for Digital Business, points out that analytics-based decision makers shouldn’t be called out for a bad decision any more than their counterparts who do not utilize analytics. At the same time, says McAfee, “as more data becomes available, decision making should become more rigorous. If it does, there will be fewer bad ones.”

“A scientific approach is not a guarantee of success,” he continued. “But it is demonstrably better, in every sense of the word that we care about, than relying on human experts and human intuition.”

Anthony Goldbloom, CEO of Kaggle, uses a poker analogy to explain his thoughts on data scientist culpability. Kaggle hosts big data competitions in which thousands of data scientists from around the globe compete to build the best algorithm — everything from predicting customer churn to predicting molecular activity.

“You can play a very good hand and lose or you could play a very bad hand and win,” says Goldbloom. “Over the long run, if you’re a good poker player and you play clever poker, you’ll win more often than you’ll lose. But you are going to lose some hands. I think the analogy applies to data-driven decision-making, as well. If you’re basing your decision on hard data rather than intuition and gut feel, you should win and make good decisions more often.”

But here’s the other thing to keep in mind: A small team of data scientists can mean big bucks for an organization — win or lose.

“Because there’s so much leverage in some of these algorithms, the difference between a great data scientist working on an algorithm and a poor data scientist working on an algorithm can be in the order of hundreds of millions of dollars in ROI,” says Goldbloom. “So a great data scientist with a data set and an important problem can generate enormous value, whereas a poor data scientist with a data set trying to build a predictive modeling algorithm can cost a lot of money.”

As dollar amounts tied to analytics get bigger and the stakes become higher, will judgment revert to the middle ages?

Topics

More Like This

Add a comment

You must to post a comment.

First time here? Sign up for a free account: Comment on articles and get access to many more articles.

Comments (2)
somkdsr
It depends on the position of the person doing the prediction. If we place a smoke alarm and we expect it to work when there is smoke. A sprinkler is expected to work when there is a fire. A pilot is responsible for the safe flying of the aircraft. The aircraft designer, the manufacturer, and the maintenance people have responsibilities for their portions of the enterprise. By following that kind of a logic, one would first ask if the scientist is simply calculating and presenting the results along with the assumptions and the process followed, he is merely a technician and his responsibility is minimal. If he is guaranteeing his results, he is like a doctor prescribing a medicine, and his responsibility is higher. If he is in-charge of safety, he is very responsible.
dbhalbur
I certainly agree with the "great vs poor" data scientist comparison.

One of the challenges of working with data is clearly communicating the results of the analysis.  It would surprise me if the Italians that are now in jail gave advice that they communicated as being absolutely correct.

Any meaningful analysis and conclusion comes with a margin of error and communicating this is a challenge.  The audience may take phrases such as "reasonably sure" to mean the outcome is guaranteed - when of course it's not.

While data scientists can get themselves in trouble by not adequately emphasizing the level of risk, they should not be held liable to the extent their analysis has not be skewed in any way.