Improving the representation of Black students and other people of color in science, technology, engineering, and mathematics careers has been a long-term goal in the U.S. education sector. There has been notable progress; for example, while comprising 11% of the U.S. workforce as a whole, Black workers represent 9% of STEM jobs.
However, Black Americans remain underrepresented in the STEM workforce relative to their 14.4% share of the US population. This is particularly evident in the data science field, where only 4.2% of data scientists are Black, underscoring questions about how lack of diversity can contribute to biases in AI algorithms.
Partnering with HBCUs
In 2023, FICO created an analytic challenge relevant to aspiring Black data scientists that would give them a taste of the real-world problems we work to solve in the financial services industry. Working with historically black colleges and universities (HBCUs), we created analytic problems that students could complete in about eight weeks, while concurrently learning about the principles and practices of “Responsible AI.”
In Year 1, the data science problem challenged participants to identify and solve for bias in housing data sets that factor into lenders’ mortgage decisions. Enlisting key members of my data science team as practitioner mentors, we met and worked with more than 100 students, faculty and staff at HBCUs.
In Year 2, we expanded the program to include two new HBCUs and a two-year technical college. This year, the students built real-time analytic models to detect fraudulent credit card transactions. Throughout this process, my team and I have enjoyed igniting students’ curiosity and intellectual passion, attracting the next generation of talent to any profession.
Many data science programs at HBCUs are nascent but eager to grow, leveraging resources like the HBCU Data Science Consortium. HBCU educators were uniformly excited to expose their students to real-world exercises and have them work with industry data science practitioners. At each school, we found scores of aspiring data scientists wanting to effect positive change in their communities, supported enthusiastically by faculty and staff.
Across Years 1 and 2 of the Educational Analytics Challenge, FICO has partnered with Alabama Agricultural & Mining University, Bowie State University, Delaware State University, Morehouse College, North Carolina Central University, Fayetteville State University and J.F. Drake State Community & Technical College.
The impact of new perspective
Throughout the program, my team and I have consistently heard from students that what they liked best was having the unique opportunity to engage in rigorous, real-life analytic exercises beyond the classroom. For the first time, they could get their hands on real data—big, unwieldy, messy data—and grapple with real-world issues such as lending bias, which persists despite the landmark Fair Housing Law in 1968.
The reality is that data science is not a clean, deterministic system; there is no single right answer. Instead, there are myriad decisions for data scientists to make, each of which incrementally impacts the quality or bias of the resulting model.
For example, deep in the trenches of the Year 1 work period, my team and I were particularly impressed by how one of the student groups was thinking about tract housing data. They pointed out that large amounts of tract housing on any given land parcel can be a proxy for race, and perhaps that data shouldn’t be used in an analytic loan originations model because it could impute bias. Other data scientists ultimately may have come to the same conclusion of imputed bias, but the eventual conversation would likely be missing that honest, passionate introspection.
As a corollary, given the collective complexity of these data science decisions, each one needs to be carefully tracked, no matter how infinitesimal it may seem. Analytic models are the eventual products of multiple paths that were explored during the development process and provide important pointers as to how data scientists can drive better outcomes from ethics and discrimination perspectives.
Teaching the students how to track their work, and ensure that key decisions are accountable and auditable, served as a primer on one of the fundamental practices of Responsible AI, and one of my personal passions, Auditable AI.
Lasting power of mentorship
The analytics challenge introduced students to another real-world concept: the value of working with seasoned practitioners as mentors, to gain a true understanding of how data scientists build and train analytic models within industry.
In addition to my lectures and speaking with students in person at all the schools, senior data scientists on my team met with students weekly, helping them to work through the challenge exercises. These ‘office hours’ gave the students a sense of what a data science career could look like if they continue this professional direction. Best of all, the students and mentors forged genuine, mutually enriching connections that can last a lifetime.
To successfully drive change, companies of all kinds need to act now to prepare upcoming data scientists and business leaders. Doing so requires more than providing funding or internships. Being personally involved with students, working through problems together and jointly experiencing the joy of learning is the most successful way to inspire young people. It is also the most gratifying.
