Translate to multiple languages

Subscribe to my Email updates
If you enjoyed these post, make sure you subscribe to my Email Updates

Tuesday, January 03, 2017

What’s Inside the “Black Box” of Machine Learning? | RTInsights

Machine learning can optimize business decisions, but the decision reached by an algorithm often isn’t transparent.

"How far can machine learning take us? The list of possibilities is endless.  Machine learning applications “can provide customer service, manage logistics, analyze medical records, or even write news stories,” a recent report by McKinsey Global Institute explains." according to Joe McKendrick, RTINsights' Industry Insights Editor in charge of contributed case studies.

Photo: RTInsights (press release) (blog)

The McKinsey report identified 120 potential use cases and interviewed 600 industry experts on the potential impact of machine learning. As machines take on routinized decision-making processes, “the value potential is everywhere, even in industries that have been slow to digitize,” the report’s authors explain. At the same time, machine learning faces challenges as it gains traction across enterprises.

The report’s authors, led by Nicolaus Henke, global leader of McKinsey Analytics, observe that “recent advances in machine learning can be used to solve a tremendous variety of problems — and deep learning is pushing the boundaries even further.”

Traditional software is programmed to do one single task repetitively. Machine learning, however, is based on “algorithms that ‘learn’ from data without being explicitly programmed,” Henke and his team explain. “The concept underpinning machine learning is to give the algorithm a massive number of ‘experiences’ — training data — and a generalized strategy for learning, then let it identify patterns, associations, and insights from the data. In short, these systems are trained rather than programmed.”...

Machine learning applications 
Problems machine learning can address problems from “keeping race cars running at peak performance to ferreting out fraud.” For example, Formula One teams “recently turned to machine learning to hold down costs in their aerodynamics operations divisions, which typically eat up more than 80 percent of development resources. Building on years of diverse project data—including CAD logs, human resources data, and employee communications—they looked for patterns that influenced the efficiency of an individual project. They discovered, for example, that too many engineers or long stoppages typically increased labor hours on a given project by five to six percent, while team use of the documentation system improved productivity by more than four percent. Overall, this application reduced the budget by 12 to 18 percent, saving millions of dollars.”

Challenges with machine learning 
As with everything that seems too good to be true, there are gotchas. The McKinsey report addressed the challenges machine learning still faces:

Deep learning models are opaque.

Call it the black box effect. “As of today, it is difficult to decipher how deep neural networks reach insights and conclusions, making their use challenging in cases where transparency of decision making may be needed for regulatory purposes,” the report observes.

Machine intelligence has ethical issues. 
Whoever programs or feeds data into machine learning algorithms may have undue influence on results delivered. “One set of ethical concerns relates to real-world biases that might be embedded into training data,” Henke explains. “Another question involves deciding whose ethical guidelines will be encoded in the decision making of intelligence and who is responsible for the algorithm’s conclusions. Leading artificial intelligence experts, through OpenAI, the Foundation for Responsible Robotics, and other efforts, have begun tackling these questions.”

Source: RTInsights (press release) (blog)