E-Handbook:
Machine learning and bias concerns weigh on data scientists
Article 1 of 4
Bias is the silent killer of well-intentioned algorithms
Bias in machine learning algorithms has come under scrutiny in countless applications, underscoring how much "silent AI" has permeated so many aspects of everyday life and influenced far-reaching, sea change decisions in business and society.
The issue of machine learning and bias has especially intensified with COVID-19 lockdowns and hospital bed count predictions, police and prison reforms, credit ratings, upcoming elections and even the use of language. Facebook CEO Mark Zuckerberg, AI Trends reported, "has promised that machine learning and AI will enable the company to combat the spread of hate speech, terrorist propaganda, and political misinformation across its platforms." First Amendment watchers are taking notice.
"Regulating AI is challenging, but industries will need to create standardization around development and certifications, as well as a set of professional standards for ethical AI usage," noted Gartner. The pressure is on companies to balance ethics with their need to fully realize machine learning's competitive advantages. "By 2024," IDC predicted, "the productivity gap will double between enterprises that deploy ML-enabled data management, integration and analysis to automate IT and analytics-related tasks and those that do not."
Well-intentioned data science warriors are working to address, correct, neutralize and eliminate bias. Attributed to a combination of human factors, bad data or a lack thereof, bias can taint predictive models and perpetuate social stereotypes by race, gender, age, location, economic status and ethnicity.
"There are a number of different kinds of bias, including sample bias, observer bias, exclusion bias, etc. These can creep in at various stages of the machine learning process," said Fern Halper, senior director of research, advanced analytics, at TDWI. "There is research going on regarding tools and algorithms to help with this." IBM's AI Fairness 360 open source toolkit, for example, includes a set of metrics for data sets and models that test for bias as well as algorithms that mitigate it.
This handbook examines the many forms of machine learning bias that affect the lives and livelihoods of individuals and businesses. We also take an inside look at the nuances of bias in predictive models and the methods to combat it as seen through the eyes of a veteran data scientist who deals with machine learning and bias every day. And responsible AI and human-centric machine learning may just be an answer to ensuring fair, balanced and realistic algorithms that can benefit everyone.
