The Time is Now: Let's Eliminate Bias in Emerging Tech

Based on the research I have overseen, there are three areas where emerging technology solutions are at the most risk of perpetuating bias: skewed datasets, subjective hardware, and humanization of technology. 

Skewed Datasets 

We all know that when it comes to AI, data sets are critical. Since most AI today learns from the information we provide, past biases (whether we are aware of them or not) are embedded in those solutions. They are being replicated and normalized.

Solutions influenced by these biases can be found everywhere:

• A ProPublica study found that unaware of bias, judges have relied on a racially-biased AI model to determine risk of recidivism, resulting in longer sentences for African American offenders for the same crimes as others. 

• In April 2015, the first female results for a Google Image search for "CEO" required scrolling past multiple rows to a Barbie CEO. Research from the University of Washington showed image search results can influence people's perspectives of who is in a role, exaggerating stereotypes that women are not CEOs. 

The good news—in this case—is that Google is doing something about it. John Giannandrea, who leads AI at Google, was recently quoted as saying:

"It's important that we be transparent about the training data that we are using, and are looking for hidden biases in it, otherwise we are building biased systems… If someone is trying to sell you a black box system for medical decision support, and you don't know how it works or what data was used to train it, then I wouldn't trust it."

 Subjective Hardware

Bias in hardware design can also work against female users. Surgical robotics that fit male surgeons' hands better, for example, act as barriers for women who are at a disadvantage when using the same technology. This 'fit' issue for women could mean they are perceived as being less effective at their jobs, which can have a negative impact on their careers, potentially prevent women from being accepted or respected in the field, and thus discourage other women from entering the field. 

The embedded software in devices can also make it harder for select populations to use devices. The most glaring example comes from the latest virtual reality headsets. As first identified by Danah Boyd—a Principal Researcher at Microsoft Research and founder of Data & Society—there are biological preferences for specific types of depth perception cues, and the solutions selected to provide depth perception in most virtual reality headsets make women feel more nauseous than men. 

These embedded biases can affect women in a range of personal and professional scenarios, from job training, to learning how to fly a plane, to training for parents on CPR for newborns. 

Humanization of Technology

Giving technology products human personas in audio or physical forms also creates risk of perpetuating stereotypes and normalizing biases. Virtual assistants like Siri and Alexa were all launched with a female only voice option, furthering the stereotype that women should assist you with tasks. After significant media attention, these products now offer the option of a male voice (though the default is still female).

When they first came out, these digital assistants were also programmed to respond in a subservient manner to sexual comments like "you're sexy". After pushback that normalizing verbal sexualization of digital assistants would normalize real-life behavior, now digital assistants actively acknowledge inappropriate behavior by saying "I'm not going to respond to that." 

So, Who is Responsible Here?

As rapid innovation with emerging technology continues, I see a real opportunity for consultants in being able to identify and eliminate biases. Do companies need to take responsibility for the fact that end users think the curvy robot should get them coffee? As consultants, what is our role in educating our clients who are building products on bias identification and inclusive design?

Ultimately, the responsibility for leading the charge to incorporate inclusive and thoughtful design into products lies with our clients.  It is my goal that with some education and insights, we can educate them on the benefits of eliminating bias in its early phases. By adding bias assessment into the product development lifecycle, developers and data scientists will be better able to catch it before it becomes legacy and more difficult to change. 

Companies are in a powerful position to increase the value of designing products for everyone and showcase the business value of doing so. Many have already done so in cases where media has brought attention to the issue, but that's not enough; it needs to be intertwined with corporate values. 

Thank You, Professor

Second, there's also an opportunity to enable academia. Increased funding for academics to identify more use cases with biases will draw more attention to the risks and can lead to potential solutions for reducing biases. We need to equip emerging creators—data scientists, product managers, developers—with the framework and tools to deal with issues such as identifying biases in datasets.

Conclusion

Let's not wait until biases are spread across too many new solutions before our group of professionals addresses the issue. The time is now for emerging technology leaders to step up and enable their teams to be thoughtful and successful in building unbiased products.

Megha Mathur is a technology business strategy consultant who is passionate about using technology and data to empower organizations and people. In her client-facing work, she brings deep insights into an array of areas for Fortune 500 companies developing and using emerging technologies such as AI/ML. She also heads Keystone Strategy's Diversity Initiative.