Suggested time: 5 minutes
Algorithmic transparency is the principle that information about decisions made by algorithms should be visible to those who use, regulate, and are affected by the systems that employ those algorithms. Algorithmic transparency should enable the monitoring, checking, criticism, or intervention by interested parties. In this course, we use the following broad definition:
Algorithmic transparency refers to the extent to which a human can understand an algorithms (Bell, Nov, Stoyanovich 2023).
There is no single there is no single definition for algorithmic transparency. The definition we provide here is an amalgamation from Policy Briefs on algorithmic transparency and academic work, and it is intentionally very broad to represent multiple meanings.
For example, in one situation, algorithmic transparency may mean that people are able to anticipate what decision an algorithmic system would make, and in another context it could mean knowing the list of factors that are taken into account by the algorithmic system.
As an analogy, consider that there are many ways to understand how a television set works. You can understand aspects like,
It’s important to note that data scientists, researchers, managers, and even policymakers use a variety of definitions (and often interchange them) to speak about transparency. For example, researchers and data scientists most commonly use the term explainability to mean transparency, but other terms include interpretability, understandability, intelligibility, comprehensibility, accountability, traceability, and legibility.
Algorithms have completely profileated the public and private sector, and ultimately our lives. Algorithms determine which movies are recommended to us on Netflix, what pages are at seen at the top of your Google search result, and even makes recommendations on what word we should type next in a text message. While many of these algorithms remain opaque, here are 3 real-world examples of algorithmic systems that are transparent (to varying degrees):
Hiring: It is increasingly common for companies to use algorithmic hiring tools that automatically process and screen resumes that they receive from online job applicants. These tools are meant to “read” through resumes and select the best candidates for interviews. In New York City, it is now law that companies must inform job applicants that their resume is being processed by an algorithm rather than a human (Zejnilovic et al. 2021).
Finance: many banks uses algorithms to determine if an applicant should be accepted or denied for a loan. By law, if an algorithm recommends to reject an applicant, that algorithm must provide reason codes that are given to the applicant, such as “too many recently opened bank accounts with balances.”
Education: Many schools across the US have begun using algorithms to identify students that are risk of failing out of school. These systems are called Early Warning Systems, and provide a score to school administrators and teachers for how likely a student is to fail. To support transparency, the algorithms behind the Early Warning Systems are made extremely simple, and normally involve adding up aspects like “how many times has the student been absent?”, “how many classes have they failed in the past”, and “how many times has the student been sent to detention?” In this way, one can have a deep understanding of how these algorithms are making decisions.
Excercise: What are some examples of algorithms that you encounter in your daily life, or at work? Are they transparent in any way?