While appreciation that algorithms and machine-learning programs are not immune to bias is increasingly mainstream, ongoing plans to correct for bias in said programs among businesses that use them are not.
That’s the conclusion of a recent study commissioned by Progress Software Corp., a business software provider, on business leaders’ awareness of data bias. Conducted by independent research firm Insight Avenue, the study was based on 640 interviews with IT professionals in 12 countries, including the United States.
“Data bias” is a term used in the study to refer to the effects of algorithms run on biased data. While “artificial intelligence” and machine learning programs are technically incapable of making human errors, these programs are “trained” based on data entered by ordinary, fallible human beings. If the data set used is flawed or biased, software programs can recreate problems that typically only come from people, like racial discrimination or predictions with human-like errors.
The study found that while 66% of interviewees expect their organizations to rely more on programs and algorithms to make decisions in coming years, 65% of respondents believed bias was already present in systems used by their company, and only 13% said that anything was being done to prevent it.
The results of biased data leading programs to make biased decisions carry serious problems for companies looking to acquire talent and avoid lawsuits. The survey notes that a large, unnamed retail company found one of its hiring algorithms had become biased against considering qualified women candidates for tech jobs at the company.
In a statement accompanying the report, John Ainsworth, GM for Progress’s application and data platform, alluded to the effects of machine conclusions based on biased data.
“Every day, bias can negatively impact business operations and decision making—from governance and lost customer trust to financial implications and potential legal and ethical exposure,” Ainsworth said.
According to Progress’ analysis, the best ways to counter data bias are by using a transparent platform with an effective system for tracking changes to data as well as by having a diverse company and HR culture free of bias.
Editor's note: A previous version of this story said the study was conducted by Progress Software Corp. The study was commissioned by Progress Software Corp. and conducted by Insight Avenue, an independent research firm. We regret the error.
