A black box algorithm is one whose internal decision-making process cannot be easily understood or examined by users, affected people, or sometimes its developers. In IB Digital Society, this is an algorithmic dilemma within topic 3.2 Algorithms, shared by SL and HL.
An algorithm receives inputs, follows defined steps, and produces outputs. In a black box system, people can observe inputs and outputs but cannot clearly explain how variables produced a result. Opacity may arise from proprietary code, technical complexity, or machine-learning models with many interacting parameters.
| Feature | Transparent algorithm | Black box algorithm |
|---|---|---|
| Process | Rules and decision steps can be inspected | Decision process is inaccessible or difficult to interpret |
| Accountability | Errors can be traced more easily | Responsibility for errors may be unclear |
| Main dilemma | Transparency may expose sensitive code or enable manipulation | Opacity can conceal bias, unfairness, or unreliable reasoning |
A common misconception is that a black box algorithm must be deliberately secret. Secrecy is one cause, but complexity can also make a model difficult to interpret. Algorithms are not neutral simply because they use mathematics: their objectives, data, variables, and thresholds reflect human choices and may reproduce inequalities.
System: COMPAS is a proprietary algorithm generating recidivism risk scores for some US courts. Example specifics: ProPublica reported in 2016 that error patterns differed between Black and white defendants, while the developer disputed its analysis using another fairness measure. Impacts and implications: Scores could influence decisions affecting defendants; opacity created risks for fairness and accountability. Concepts: The case shows unequal power, competing definitions within values and ethics, and interactions between technical and social systems.