Bias in Facial Recognition with Reverse Correlation Noise Algori

AI Tool(s) Used:

• Reverse Correlation Noise Algorithms
• Facial Recognition AI
• Randomized Image Processing

Description of Result:

The installation explores how facial recognition systems and reverse correlation noise algorithms can reflect cultural biases. The participants in Los Angeles engage in deciding which selfie best fits random criteria—such as “cheerful” or “dictator”—and then have one of their own images manipulated using AI to present two slightly different versions, questioning how machine learning systems interpret human faces and emotions.

Step-by-Step Breakdown:

1. Setting Randomized Criteria: The installation assigns random emotional or character-based criteria (e.g., “cheerful” or “dictator”) to images within a set of selfies.
2. Participant Decision-Making: Participants are asked to choose which image best fits the criteria, reflecting how AI might be trained to make similar judgments about facial features.
3. Reverse Correlation Algorithm: The AI uses reverse correlation noise algorithms to generate altered images. This technique involves applying noise patterns to a photo and generating new versions that emphasize or distort specific features.
4. AI-Manipulated Output: The participant’s photo is manipulated, producing two images with subtle differences. This step highlights how the AI’s perception of identity and emotion can vary based on the algorithms applied.
5. Cultural Bias Reflection: The piece shows how facial recognition technologies can carry biases, as the AI might associate certain facial features with inaccurate or culturally biased judgments.

Tips & Tricks:

• Experimenting with Noise Algorithms: When applying reverse correlation noise, small changes can drastically affect the final image. Experiment with different noise levels to see how facial characteristics shift.
• Cultural Awareness in AI Design: AI tools must be built with cultural sensitivity. Ensure that training data for facial recognition models is diverse to minimize bias.
• Participant Interaction: Including participants in decision-making (e.g., selecting images based on criteria) enhances engagement and allows for real-time exploration of AI’s decision-making process.

Annotation:

This artwork exposes the inherent biases in facial recognition systems by inviting participants to take part in selecting images based on arbitrary characteristics. By doing so, it raises important questions about how AI models are trained to recognize and interpret human faces. The use of reverse correlation noise algorithms reflects how subtle distortions in images can change perceptions, highlighting the potential flaws in how AI analyzes and categorizes people.

The manipulation of participants’ selfies to offer two slightly altered versions underscores the unpredictability of AI’s interpretation of facial data. The artwork pushes viewers to think critically about the accuracy of AI in recognizing human emotions and characteristics, especially in contexts like security and surveillance, where such judgments can have serious consequences. The piece also illustrates the dangers of cultural biases being baked into AI systems, prompting conversations about ethics in machine learning and the future of facial recognition technology.

Instagram Post – Facial Recognition Bias