Bring on the 'algorithmists'

Tools

Viktor Mayer-Schonberger and Kenneth Cukier earlier this month published a book together called "Big Data: A Revolution That Will Transform How We Live, Work, and Think" that examines the drivers, trajectory and impact of big data analytics.

Mayer-Schonberger is a professor at the Oxford Internet Institute at Oxford University, and is on the advisory boards of Microsoft (NASDAQ: MSFT) and the World Economic Forum. Cukier is the data editor of The Economist and a prominent commentator on developments in big data.

They shared an excerpt of their book this week in Quartz, about how the algorithms used in big data are creating artificial intelligences that no human can understand, making it quite the challenge to retrace the steps that led to a determination.

Software code is decipherable. Programmers can trace their way back step by step to find an error, and then fix it. And logs generated by network elements can help network operators determine why something may have routed incorrectly. But data scientists can easily paint themselves into a corner.

Using Google's (NASDAQ: GOOG) experiment in identifying a flu breakouts as an example--not that the experiment was all that successful--the authors pointed out that even that program relied on the testing 450 million mathematical models. It would be very difficult to retrace the relationships to identify how and why a conclusion was reached. They said "explainability" is important for humans who tend to want to know why, not just what. If a model automatically generated, say, 600 predictors from a given algorithm, most of which were weighted low individually but together would improve the model's accuracy, "the basis for any prediction might be staggeringly complex."

Answering the question, "On what basis?" could become so complex that "big-data predictions, and the algorithms and datasets behind them, will become black boxes that offer us no accountability, traceability, or confidence," the authors said. This, they added, will create the need for a new kind of expertise in monitoring and transparency. "These new players will provide support in areas where society needs to scrutinize big-data predictions and enable people who feel wronged by them to seek redress," they said.

In other words, big data analytics may have to be certified. The authors suggest the new players required for this purpose be called "algorithmists" and monitor the process from both internal and external perspectives, like having in-house accountants as well as outside auditors. They said algorithmists would provide a market oriented approach to problems and perhaps forestall the imposition of intrusive regulations around big data.

For more:
- see the Quartz excerpt

Related Articles:
Google's bad case of the flu
Causation vs. correlation: Which is it?
Non-scientific Internet trend analysis can mislead