Can a computer predict a crime, and should it? For PBS NOVA's Computers v. Crime, Pixeldust created the graphics and animation that let viewers see how machine learning works, and how the data behind it shapes what it predicts.
Computers v. Crime examines the use of artificial intelligence in policing and the justice system, and the questions it raises about fairness. Machine learning happens out of sight, so the documentary needed visuals that made it understandable without flattening the debate.
Pixeldust’s sequences connect the story’s human subjects with the data and computational processes behind these systems. The finished work brings live-action footage and animated digital environments into one visual style, with facial tracking and network imagery giving viewers concrete points of reference as the story moves between people and computers. The related NOVA segment, How Computer Algorithms Make Predictions, is also part of a PBS LearningMedia resource.
Science storytelling for PBS NOVA
Machine learning operates largely out of sight. The challenge was to give those processes a clear visual form for a general audience. The graphics had to support the documentary’s explanation of pattern recognition while preserving its central question: how does the data used to train an algorithm shape the predictions it makes?
The segment examines how predictive policing can reinforce patterns already present in historical records. It describes a feedback loop: more police presence generates more recorded incidents, which can then influence future predictions.
That context gives the graphics their purpose, helping viewers follow the technology while considering how its inputs affect its conclusions.
Pixeldust built a consistent visual language around blue and violet light, layered interfaces and moving fields of information. Animated code and binary digits suggest the scale of computer processing, while city imagery anchors the subject in recognizable places. Changes in camera perspective carry viewers through the digital environment and keep the technical material engaging.
Tracking boxes placed over people in a crowd make automated facial analysis visible within live-action footage. Other sequences move into a digital interface, where a face is framed and overlaid with a network of points and connecting lines.
The neural network animation shifts the perspective again: connected nodes and pulses of light represent information moving through a computational network. Together, these treatments give the audience distinct visual cues for facial analysis and the processing behind it.
