Reconstructing Memories Using Generative AI

Memory loss due to trauma, aging, or neurological disorders affects millions globally. Traditional memory rehabilitation methods lack personalisation and immersion. Generative AI (e.g., diffusion models, GANs) combined with Virtual Reality (VR) can recreate environments from partial cues (photos, sounds, videos, text), potentially aiding memory recall and emotional wellbeing.

This PhD project aims to investigate the viability and benefits of generative AI use for memory reconstruction in VR.

It focuses on employing generative AI to recreate lost or fragmented memories in VR based on cues (photos, voice/video recordings, etc) with view to addressing key challenges like how representative and ethical AI-driven memory reconstruction is and whether this innovative approach can help trauma therapy or nostalgia-based interventions. As such, it examines answering the following research questions:

1. How accurately can generative AI reconstruct personal memories from incomplete data and support reminiscence activities?
2. Does immersive VR enhance emotional engagement and memory retrieval compared to non-immersive methods?
3. What ethical implications arise from AI-generated memory content?

To address these research questions, the project objectives are set to:

• Developing a pipeline that uses multimodal inputs (images, audio, video, text) to generate VR scenes.
• Evaluating the impact of reconstructed VR environments on memory recall and emotional response.
• Establishing ethical guidelines for AI-driven memory reconstruction.

The first phase of the methodology to pursue in the project will be about data collection for gathering anonymised personal media (photos, voice notes, etc) from volunteers. This will drive the AI modelling phase, where a generative model (e.g., Stable Diffusion + LLM for context) will be trained to create realistic VR scenes. Subsequently, the VR integration phase will lead to implementing scenes in Unity or Unreal Engine for immersive experiences.

Final phase of the project will be dedicated to evaluation of the proposed solution by conducting a series of lab-based VR experiments and participant response surveys pre- and post-exposure. The expected impact-driven research contributions from this PhD project are a novel framework for memory reconstruction using AI and VR, supporting reminiscence activities; insights into psychological and emotional impacts of AI-generated memories; and ethical framework for responsible use of the proposed technological solution.