Preventing AI Bias Starts at the Top – Just Ask These Female Chief AI Officers

Imagine you’re staffing up – you’ve put up the job postings and begun screening candidates using AI, but you’ve noticed there’s a lack of diversity in the candidate pool. Wondering why? A chief AI officer can tell you: Chances are the AI is biased.

Examples of biased AI have been cropping up for years, like the hiring tool Amazon developed and then scrapped in 2018 after it was found to discriminate against women. Another firm’s resume-screening tool had a fondness for the name Jared and players of high-school lacrosse, two random factors it found indicative of job performance. In another recent case of AI bias, where computer-science researchers asked ChatGPT to write recommendation letters for hypothetical employees, it described male candidates as “respectful,” “reputable,” and “authentic,” and females as “stunning,” “warm” and “emotional.”

At a time when 70% of companies report using automated and AI-enabled tools for their hiring processes, it’s critical for AI teams to be able to identify and correct biases – not just gender bias but all kinds – in their AI tools and algorithms to protect not only the business’s reputation but also its bottom line. Diversity on the human teams plays a key role in this, and more businesses are recognizing the value of women in AI leadership positions.

Diversity impacts the workplace in myriad ways. So does the Digital Employee Experience (DEX). Here’s how to monitor, manage, and improve yours in real time from one platform.

A recent IBM survey showed 33% of businesses have a woman in charge of AI strategy decision-making. Still, more than half (56%) of business executives and mid-level managers report that there “aren’t enough women leading the conversation about generative AI” and 73% of business leaders note that having female leaders in AI is critical for reducing gender bias.

“We have to curb the deficit and [ensure that] more than one qualified woman is sitting at the table,” notes Janet George, corporate vp, data center and AI group at Intel Corporation. “There’s often just one.”

This has been a tough year for women in AI. In March, UNESCO released a Generative AI study revealing gender biases and racial stereotyping in large language models (LLM). In April, the 13-year old nonprofit Women Who Code closed. In May, the question of AI regulations in identity protection came under scrutiny when Scarlett Johansson alleged that OpenAI used her voice without consent for its “Sky” ChatGPT. Despite such challenges, there are new opportunities for women-owned AI businesses on the horizon. Google and the Female Foundry are providing access to a Visionaries AI Incubator for select female-founded companies.

With the rise of AI driving a critical need for diversity of thought and perspective in the executive suite, Focal Point turned to three female chief AI officers to get their take on the hurdles and horizon before them. They discussed the importance of mitigating bias, bringing more voices to the table, and boosting the business’s bottom line, which – as they emphasized – are all related.

Diversity for the bottom line

Some sectors and companies have seen a backlash to diversity efforts, but the business case is clear: According to McKinsey’s 2023 report titled Diversity Matters Even More, businesses “in the top quartile of ethnic representation” have a 39% increased likelihood of outperformance.

We have to [ensure that] more than one qualified woman is sitting at the table. There’s often just one.

Katia Walsh thinks that applies especially to leadership. “The world needs to change… opportunities for underrepresented leadership needs to change,” notes Walsh, chief digital officer at Harvard Business School and former chief AI officer at Levi Strauss & Co. and Vodafone Group. That “monumental” task, she notes, requires leaders in all industries to work toward greater diversity. “We have the data, but we need to act on it,” she says. “The more diverse the executive team is, the better the business decisions will be, that will represent the real world better.”

At Michelin, the powerhouse French tire manufacturer, Ambica Rajagopal, group chief data and AI officer, cites collective intelligence as a critical enabler for a high-performing team, noting that equity and diversity in all facets (of background, personality type, gender, and beyond) are all equally important. Rajagopal is responsible for building AI technology teams and for identifying the AI use cases to bring value to the business. Besides just tires, Michelin is developing AI-fueled tech for a variety of “mobility solutions,” from microchips that can predict tire wear and tear to cultural taste-prediction tools to make their world-renowned Michelin hotel and restaurant guides more personalized and customizable.

[Read also: Chief AI officers are in demand and identifying qualified candidates can be tough – here’s where to find yours]

“When you think about building customer-focused solutions that really have an empathy with the customer base, having a reflection of the broader society in the team helps to build that as well,” Rajagopal says.

Lifting underrepresented voices

Forward-thinking enterprise leaders can take a number of actions to boost underrepresented voices, which can help ensure less-biased AI deployment. One controversial option that many believe will bear dividends: adopting regulations that encourage diversity in leadership roles at your company. For example, Nasdaq’s listing rule requires companies have a minimum number of diverse board directors. The ruling hasn’t been easily accepted, and experts believe the law will be quashed by an appeals court that heard oral arguments in May.

We have the data, but we need to act on it. The more diverse the executive team is, the better the business decisions will be.

Still, experts like Intel’s George believe enterprises should start following this rule, whether it remains law or not. “We have to champion laws,” she says. “We need dramatic progress milestones in the composition of the C-suite. When a law is in place, people will follow it.” George advises that every C-suite team should look at its own makeup and ask: Is our C-team composition 50-50 [men-women]? “Equity starts at the top of the leadership food chain,” says George.

Another way to bring more voices to the table is to review barriers in job descriptions to make sure you’re not impeding applicants. When building her team, Walsh removed the education requirement.

“Really assess [job descriptions] with an objective eye,” advises Walsh. “In some cases, you can use generative AI in job descriptions, and flag certain requirements that are impeding.”

Having programs in place to encourage underrepresented voices to enter the field can also help. Walsh’s former employer instituted a reskilling program, an eight-week intensive training in machine learning. The program had a screening process to test applicants for problem-solving skills, analytical skills, and perseverance – qualities needed for the program – but did not require a coding background.

[Read also: Enhancing employees’ ability to work with technology pays massive dividends for individuals and the overall digital employee experience (DEX) – learn more]

“It’s about giving people the opportunity [to learn] and you see business outcomes,” says Walsh. “Some changed jobs, but most people we trained went back to their old roles, and instead of using Excel, they were able to automate their tasks using code in Python. Now they have the technical skills, updated their jobs, and delivered better business outcomes.”

Looking to be a chief AI officer? Find a mentor first

While enterprise leaders examine such systemic ways to improve AI diversity, those looking to move up in the AI world can do their part, too. Peer-to-peer mentorship can be helpful in solidifying your niche in the field of AI. Rajagopal encourages connecting with professionals in the industry who share your specific interests.

Understand the aspects of AI that really matter to you, and find those people in the community at conferences and LinkedIn who share your same concerns.

“Understand the aspects of AI that really matter to you, and find those people in the community at conferences and LinkedIn who share your same concerns and thought processes,” says Rajagopal. “For me, it was the human impact of AI both from the ethical and change management perspective, and the deep technical aspects, people who were technically gifted in AI.”

She also notes the prevalence of the “leaky pipeline” effect that impacts women in tech. “Mid-career we see the percentage of women dropping… In the case of bringing women into AI, it’s about building role models and talking about our experiences and our journey.”

How chief AI officers mitigate bias

When it comes to implementing AI, having diversity of data, diversity of teams, and diversity of tools are all key to mitigating bias.

Being alert to potential biases in the existing data, for example, should be everyone’s responsibility, from the C-suite that approves and is ultimately responsible for AI implementation to the departmental employees that work with these algorithms. When everyone feels part of the cybersecurity team, the chances of catching bias are greater. Not doing so can be costly. Last August, the EEOC settled its first AI-related discrimination lawsuit, in which a tutoring consortium agreed to pay $365,000 after its AI hiring tool automatically rejected female applicants over 55 and males over 60.

Choosing the right tools can also help reduce biases. Walsh recommends using tools that incorporate representative data. “I have a preference for open-source tools because by definition you have the whole world working on that, from different backgrounds, educations, locations, and that gives you greater chances [for diversity],” she says. “It’s not eliminating [voices]; it’s maximizing the chance to have less bias.”

[Read also: AI is not only transforming business, it’s protecting it – here’s your ultimate guide to AI cybersecurity]

Rajagopal emphasizes the importance of establishing enterprise-wide AI guidelines, and likes tools that educate data scientists on potential biases. Her team is currently exploring options from companies such as Holistic AI and Fiddler AI, which offer tooling solutions that provide machine learning and LLM-model monitoring and AI safeguards. For example, Fiddler lets users assess intersectional unfairness by reviewing dataset metrics.

The goal is to reflect the range of human experience and perspective, with AI as “a mirror that we hold to life,” says Walsh. “The mirror reflects what’s there. Until we change what’s there, the mirror will continue to reflect a very flawed world.”

TO LEARN MORE

For those looking to build their network and knowledge, consider these resources:

• Chief AI Officer Summit – The CDO Club’s second annual CAIO summit will be held October 2 (a day after its CDAO Summit) in Washington, D.C. Speakers will include the U.S. Treasury Department’s deputy CAIO, Brian Peretti; senior advisor of digital talent at the DoD’s Chief Digital and Artificial Intelligence Office, Angela Cough; and chief AI officers from Avanade, Fox Rothschild, and the U.S. Department of Housing and Urban Development, among others.
• Chief AI Officer (CAIO) Talent Map – The first-of-its-kind report produced by the CDO Club covering key chief AI officer demographic data, compensation, region, career path, reporting structures, and open CAIO jobs. Due out in mid-July.