48 x 76 x 5 cm
AI-equipped interactive mirror that only allows you to
see yourself when you express happiness


Algorithms in Reflection is a series of interactive mirrors with embodied AI that responds to facial expressions. The work initially began as a single mirror that responded exclusively to happiness. Using a camera and a convolutional neural network trained on the FER-2013 dataset, it detects when a viewer smiles and reveals their reflection only when they appear happy. The AI assesses the authenticity of the smile and generates a “confidence score,” which measures its certainty about the smile’s genuineness. The mirror interacts with viewers by raising or lowering a blind. To see their reflection, they must keep smiling. If they stop, the blind falls and hides their reflection. This creates an immediate and immersive exchange between the viewer and the AI system. The artwork highlights the growing influence of AI in our everyday life. This mirror explores what it means to express emotions to a non-human entity and have them acknowledged by it. We are already accustomed to smiling and laughing in social interactions to appear happy, whether genuine or not. But does it feel different when we must convince AI? In the end, this mirror shows us that as we interact more with AI, we are also reflecting on ourselves, reminding us of our own nature as social beings.

The series has since expanded to include sadness, creating a clear contrast with the happiness mirror. These two opposing emotions highlight how differently we are trained to perform them: smiling is socially encouraged, while sadness is often concealed. Experiencing both mirrors allows audiences to feel how distinct it is to sustain or feign each emotion, and how differently each is interpreted by an AI system

ISEA 2025

Technical Specification

Hardware

• The mirror is a hangable artwork with all the electronics embedded inside a sealed structure.
• It uses a stepper motor to roll the blind up and down. The system is controlled by single board comptuer and mcirocontorller connected to Camera and a driver.
• The hardware operates in low DC voltage of 5v and 12v
• The hardware system is stable and it was rigorously tested for four months in a real exhibition environment without interruption.

Software

• AI control is powered by a Convolutional Neural Network (CNN) trained on the FER-2013 dataset, which comprises approximately 30,000 faces.
• Program only detects the largest face in the camera’s view, prioritising that person’s expression as the main focus, even when other individuals are present in the background.
• The working distance of the mirror is approximately between 40 cm to 110 cm
• Operates entirely offline in real-time, independent of any external connections, and does not record or store any data.
• The software system is stable and it was rigorously tested for four months in a real exhibition environment without interruption.