Women account for only an estimated 12 percent of AI researchers.

SwissCognitive

Women account for about half the world’s population and nearly 47 percent of the U.S. workforce. But when it comes to the people researching and creating artificial intelligence—a technology with growing influence in everyday life—women account for only an estimated 12 percent of AI researchers.

That’s why many leaders in the field warn that gender bias can creep into AI algorithms. To change that dynamic, they are engaged in a variety of efforts to raise awareness and encourage more women—and members of other under-represented groups—to join the profession. Lisa Amini: “We’re seeing that AI driven by machine learning is transforming how we live, how we work and how we socialize with others.” The problems of gender bias in AI are all too evident. State financial regulators in New York, for example, recently began investigating Goldman Sachs Group and allegations of gender discrimination in the Apple Card credit card issued by the bank. Algorithms help determine who gets approved for a card, at what credit limit. A Bloomberg news article reported that one male Apple Card customer received 20 times the credit limit offered to his wife, even though she has a better credit score.

News reports have also told of AI-based job recruitment systems that give short shrift to female and minority applicants.

“We’re seeing that AI driven by machine learning is transforming how we live, how we work and how we socialize with others,” said Lisa Amini, director of IBM Research’s lab in Cambridge, Mass. “If women aren’t participating in that transformation, we risk both shortcomings in the technology and in women’s ability to also benefit from those changes,’’ Amini said. She also oversees the company’s AI Horizons Network , through which IBM researchers collaborate with university faculty and students worldwide.

Bias can find its way into an AI system in various ways, including the use of limited or mislabeled data in machine-learning algorithms that train the system and the algorithmic parameters that weight some data points as more relevant than others.

The biggest problem, though, isn’t the data or the algorithms. It’s the blind spots created by a lack of diversity—of experience, education and thought—within teams developing AI that make it difficult for them to anticipate bias and its potential impact.

Cultivating Diverse Perspectives

“Computer science is one of the largest growing fields with an important impact on day-to-day life for everybody,” says Carla Brodley, dean of Northeastern University’s

Khoury College of Computer Sciences. “That’s why it’s so important for everybody—particularly under-represented groups—to have an invitation to the table when it comes to machine learning, AI and computer science more generally.”

Northeastern offers a program, Align, that enables students without a computer science undergraduate degree to earn a master’s in that field.

“The current pipeline of computer science students is insufficient to meet the demand for tech-trained people in the workplace, worldwide,” Brodley says, adding that programs similar to Align are expected to launch next year at Columbia University, Georgia Tech, and the University of Illinois at Urbana-Champaign. […]