Microsoft is no doubt one of the leading technology companies that have taken a deep interest in artificial intelligence. Recently, it started creating a tool for automatically identifying bias in an array of different artificial intelligence (AI) algorithms.
This effort marks the boldest move yet for automating the detection of unfairness, which may set into machine learning. This impressive move could assist enterprises in making use of artificial intelligence (AI) without unintentionally discriminating against certain individuals.
Big technology companies appear to be in the race to sell their off-the-shelf machine learning technology, especially those that can be accessed through the cloud. As additional customers utilize these algorithms in an effort of automating vital decisions and judgments, the issue of bias will become critical.
Since bias can easily creep in, machine learning models, ways of automating the detection of unfairness could, in turn, prove to be an essential part of artificial intelligence toolkit.
Rich Caruna, Microsoft’s senior researcher, working on the unfairness-detection dashboard, said that intelligibility, explanation and transparency are new to the field. However, he said that only a few of the company’s experts have adequate experience in knowing everything that they ought to look for and all the various ways unfairness might creep into their machine learning models.
Algorithmic unfairness of bias serves as a growing concern for numerous technology experts and researchers. While algorithms are utilized in automating vital decisions, there is a considerable risk that bias could be deployed at scale, become automated and increasingly hard for the victims to detect.
According to Caruna, Microsoft ’s unfairness-detection product promises to aid artificial intelligence researchers in identifying more instances of bias, even though not all. He emphasized this point by saying that they cannot expect perfection from the product since there will always be traces of undetected bias, which cannot be eliminated. However, the goal for the researchers and experts at Microsoft is to do the best they can.
According to Caruna, one of the most vital things that companies can currently do is educating their workforce in a bid to ensure that they are well-aware of the myriad ways in which bias cannot only arise but also manifest itself. In turn, this effort will allow them to come up with tools aimed at making bias easier to spot and models easier to understand.
Microsoft is not the only company that is pursuing this line of technology. In fact, Facebook unveiled its own tool for spotting unfairness during its yearly developer conference, which was held on May 2, 2018.
Dubbed Fairness Flow, this Facebook tool automatically alerts if an algorithm makes an unfair or a biased judgment regarding someone, for instance, based on his or her age, race or gender. Facebook said that it required the tool since more individuals at the company now are utilizing artificial intelligence (AI) in making critical decisions.
Bin Yu, UC Berkeley’s professor, said that the tools from Microsoft and Facebook appear to be a step towards the right direction, although they may not be sufficient. For this reason, she suggested that big entities or companies ought to have external experts to audit their algorithms to prove that they are indeed unbiased.
Source TechnologyReview
