Are robots taking over? Vasant Dhar, a data scientist and professor at CDS, focuses his research on the balance between automation and humans and has recently published his work in Harvard Business Review.
Given the growing number of industries where computers are able to make decisions, a need has risen for a systematic way to gauge if a machine should be used or not. In response to this need, Dhar has created a “risk-oriented framework” to help guide decisions about what tasks should be performed by humans and what tasks are better performed by robots. This stems from his extensive background applying predictive systems to fields like finance, creating a machine that essentially learns from data and thereby makes investment decisions automatically.
“The issue of trust is central here. Are you willing to turn over control in terms of strategies and decisions to a machine?” said Dhar when explaining his framework.
The framework categorizes human vs. robot problems by predictability and cost per error. Both classifications express risk. The more predictable a problem, the less risk associated with it and vice versa. Likewise, the higher the cost per error, the more risk associated with the problem.
The spectrum of predictability ranges from “random” to “sure thing,” with long-term trading, for instance, being completely “random” and driver-less cars being a “sure thing” considering the technology and physics involved. This facet of the framework highlights where automation is worth pursuing and where it needs more work, if any. The second facet of the framework is calculating the cost per error of the same problems.
Once the two elements of each problem have been analyzed, Dhar plots them on a Decision Automation Map (DA-MAP), plotting predictability against cost per error for each problem like driverless cars and long-term trading. Both axes quantify risk; the x-axis exhibits how frequently things go wrong and the y-axis exhibits how bad is it when they do.
The plotted points are not set in stone, however. Technological advancements, societal changes, increased data, and tightening or loosening of regulations can all cause shifts on the map.
Viewing automation decisions with these two classifications weighs the costs and benefits. For instance, deciding to fully embrace driver-less cars is tricky because of the serious repercussions of an accident. For early education support, though, the consequences are minimal as it’s an extra tutoring tool for children that simply gauges their academic level and questions them accordingly. Simply put, “A higher cost per error requires a higher level of predictability for automation.”
The framework, which is less than a year old, and the consequent DA-MAP are useful for decision makers like managers, regulators, investors and others in need of data-driven guidance about automation decisions.
Businesses, primarily consulting companies, have so far shown interest in it as they seek data science strategies for their clients. In the long-term, it is a fascinating way to for executives to prioritize initiatives from a risk based perspective.
by Nayla Al-Mamlouk