The future of our memories

What It Covers

Technology transforms memory preservation through AI-generated synthetic memories, chatbots recreating deceased relatives, and three-dimensional digital scans of heritage sites. Pau Alekum Garcia creates visual memories for refugees and dementia patients, Amy Kurzweil builds FredBot from her grandfather's archives, and Chance Kochenour digitally preserves war-damaged cultural artifacts using photogrammetry.

Key Questions Answered

• •Synthetic Memory Creation: AI generates painterly, deliberately blurred images from oral descriptions to help refugees and trauma survivors visualize lost memories. The intentional lack of photorealistic detail prevents fixation on factual inaccuracies and better captures emotional truth. Models trained on local cultural archives produce more authentic results than generic American algorithms when reconstructing specific regional contexts like 1930s Barcelona homes.
• •Reminiscence Therapy Application: Synthetic memories integrate into dementia treatment through reminiscence therapy in nursing homes and hospitals. The methodology uses AI-generated images alongside music from patients' pasts to trigger visceral memories, proven to reduce anxiety and depression. The approach works because imperfect, dreamlike images mirror how actual human memory functions, creating stronger emotional connections than hyperrealistic reconstructions.
• •Chatbot Memory Archives: Creating conversational AI from deceased relatives' writings requires collecting 600-plus pages of letters, essays, and documents. The selective chatbot responds using only actual sentences the person wrote, creating a poetry-like exchange rather than simulated conversation. This approach works best when users embrace the temporal gap, allowing questions to transport them into specific past moments rather than pretending the person exists in present time.
• •Photogrammetry Preservation: Three-dimensional digital models reconstruct destroyed artifacts using multiple tourist photographs taken from different angles. Algorithms detect similar features across images to build complete models detailed enough to show ancient inscriptions. The same technology now operates on mobile devices and drones, enabling rapid documentation of 200-plus Ukrainian heritage sites threatened by ongoing conflict.
• •Cultural Archive Bias: AI models trained primarily on American imagery fail to accurately reconstruct non-Western cultural contexts. Fine-tuning models with municipal archive photographs from specific time periods and locations dramatically improves authenticity. This reveals how default AI systems embed cultural assumptions that distort memory reconstruction for non-dominant populations, requiring intentional dataset curation to preserve accurate historical representation.