That humans will be woefully inadequate against smart data science machines, in at least some areas of inquiry and analysis, is already a foregone conclusion. But the way in which machines are outperforming humans changes every day and is progressing rapidly. Perhaps in the future, when machines become smarter than humans, it will be impossible to understand the calculations they make and why they made them.
Massachusetts Institute of Technology reported last week that a group of researchers from the Computer Science and Artificial Laboratory (CSAIL) has developed a prototype Data Science Machine that beat 615 out of 908 human teams in three data science competitions. The machine was developed from a thesis written by MIT student Max Kanter, who believes the device is a natural complement to human intelligence.
The challenge was to find predictive patterns in unfamiliar data sets using what is normally described as human intuition. In current analysis of big data, automation and the use of algorithms are now used routinely but data scientists are still critical players in determining which data is important as “actionable intelligence.”
In the competition to analyze unfamiliar data, human teams needed months to make progress in figuring out meaningful patterns in data while the Data Science Machine took just 12 hours. Researchers at CSAIL noted that using a supercomputer to choose the feature set used to identify predictive patterns in big data may soon replace the need for humans and human intuition in these types of analyses.
