Lack of diversity in data science perpetuates AI bias
Data privacy measures such as the General Data Protection Regulation and the California Consumer Privacy Act expand the definition and protection of private sensitive data. Anonymization efforts, while valiant, can only go so far.
“YYou can only manage what you measure, right?said Hannah Sperling (pictured), business process intelligence, academic and research alliances at SAP SE. “But if everyone is afraid to touch sensitive data, we might not get where we want to be. I entered data anonymization procedures, because if we could make more data on the usable workforce, especially when it comes to increasing diversity in STEM or in technological jobs, we really should let the data do the talking.“
Sperling spoke with Lisa Martin, host of theCUBE, SiliconANGLE Media’s livestreaming studio, at the Women in Data Science (WiDS) Event. They discussed data anonymization and the inherent bias in human-generated analysis.
Complete objectivity is logically impossible
Removing the human factor from analysis is not only idealistic, it’s the wrong way, according to Sperling. Since the analysis is by nature a retrospective effort, she believes recognizing and adjusting for these biases is the model to follow.
“I am sometimes amazed at the number of people always seem to think data can be unbiased,” Sperling said. “Tit’s sooner than we realize that certain biases must be taken into accountthe closer we will get to something that better represents reality and could also help us change reality for the better.
The lack of diversity in data science has perpetuated biases in AI decisions, from soap dispensers that only recognize fair skin to hiring decisions, financial demands and parole approvals.
“There is a big trend around explainability, interpretability in AI in the world because awareness around these topics is increasing,” Sperling explained. “That will show you the blind spots that you might have, no matter how much you think about the context. OWe need to improve including everyone; otherwise you will always have some selection bias.”
Here’s the full video interview, which is part of SiliconANGLE and theCUBE’s coverage of the Women in Data Science (WiDS) event: