A recent study documented a participant using an AI model to simulate a conversation with her late grandmother. According to her report, the experience offered a sense of closure that she had not experienced before.
Researchers at the University of Colorado Boulder call these AI agents ‘generative ghosts.’ These systems use data such as writing, voice recordings, or photographs from the deceased to create simulated conversations. Some platforms, such as Project December and Séance AI, generate text-based versions of journal entries and messages. Others, such as HereAfterAI, include voice and images. A few companies have also developed virtual reality experiences that let users interact with holographic representations of loved ones.
Jack Manning, the PhD candidate in Information Science who led the study, expected the sessions to feel unsettling. “We originally thought it might feel very Black Mirror creepy to people and make them uncomfortable,” he said. “I ended up being completely wrong. People thought it was amazing.”
How To Build a Ghost
The study included 16 participants, aged 22 to 50, each of whom joined a Zoom interview about a deceased relative or friend. During the session, a facilitator led the conversation while another researcher entered details into a large language model, generating an AI representation of the deceased in real time.
Each participant interacted with two versions of the AI model for about twenty minutes each. One version spoke as the deceased, using the first person, while the other described the person in the third person. A facilitator stayed present to offer support if the conversation became emotionally difficult.
The results indicated a clear preference for the version that spoke as the deceased individual, rather than about them. Researchers described this as a preference for ‘reincarnation’ over ‘representation.’
What People Will and Won’t Forgive
Participants tended to overlook factual mistakes or made-up details from the AI. What mattered more was whether the AI captured the right tone and language. If the model used words or phrases that felt out of character, people were more likely to disconnect from the experience. The study found that getting the voice and manner right was more important than perfect factual accuracy.
How the AI communicated also shaped the experience. Participants preferred short, to-the-point sentences and even the use of emojis, rather than the longer, more formal responses that AI models often produce. This suggests that future versions of these systems should pay close attention to conversational style.
A Technology Both Wanted, and Feared
All sixteen participants indicated that they would be willing to use the technology again. However, most also expressed concerns about the potential risks of individuals in grief accessing these AI systems without supervision.
Lead researcher Jack Manning said his own experiences with loss have shaped his view of the risks and benefits of AI grief tools. He acknowledged that negative outcomes are possible but also noted that the technology could help some users find a sense of closure and peace.
Jed Brubaker, the study’s senior author and an associate professor of Information Science, said the lab took on the work precisely because the field had outpaced the research. “To our knowledge, we are conducting the first user experience studies of simulated AI ghosts,” he said.
The research team intends to involve mental health professionals in future studies. The next phase will evaluate the psychological risks and benefits of prolonged use of AI grief tools to ensure that empirical research guides product development.
Generative ghost technology is already available through commercial platforms, and people are actively using these systems. The study highlights that voice, tone, and pacing are often more important to users than factual accuracy. As the technology develops, researchers are working to address the challenges of providing appropriate safeguards and responsible design.
Austin Burgess is a writer and researcher with a background in sales, marketing, and data analytics. He holds an MBA, a Bachelor of Science in Business Administration, and a data analytics certification. His work focuses on breaking scientific developments, with an emphasis on emerging biology, cognitive neuroscience, and archaeological discoveries.
A new 3D image projection system from researchers at the UCLA Samueli School of Engineering and the California NanoSystems Institute (CNSI) marks a major step toward overcoming a longstanding problem for holographic technology.
Professor Aydogan Ozcan led the work reported in a recent paper published in Light Science and Applications, which describes the team’s new snapshot 3D image projection system, solving one of the technology’s most daunting issues.
“These results establish the diffractive 3D display system as a compact and scalable framework for depth-resolved snapshot 3D image projection, with potential applications in holographic displays, AR/VR interfaces, and volumetric optical computing,” the authors write.
3D Image Challenges
While 3D films viewed wearing special glasses were popular a decade ago, this type of 3D image projection is something more akin to the holodeck on Star Trek, projecting sophisticated, real-time images. While advances have been made in bringing such technology to life, many obstacles remain to creating convincing three-dimensional projections.
The snapshot 3D image projection system described in the paper combines a digital encoder with a passive optical decoder and uses a deep learning algorithm that operates end-to-end to overcome one of holography’s greatest challenges.
Interplane cross-talk is one of the main issues associated with 3D image projection. As 3D objects are sliced up and images projected at various depths, elements from one image can leak through to the image in front of it, destroying the illusion of a solid object existing in three-dimensional space.
Improving Holographic Technology
In the new system, an image first enters the encoder, which uses a Fourier-based neural network to separate spatial and frequency features into a stack of images. It then analyzes the positions where the images will be projected to simultaneously generate a 3D visual. Then the images are decoded and projected onto structurally optimized surfaces, where the depth-dependent field programming occurs as the light propagates.
The deep learning algorithm, building on recent advances in combining machine learning and 3D projection, optically routes the image slice to its correct depth while adjusting for in-plane leakage.
Together, this system creates a compact, high-fidelity display in a single shot, combining closely spaced axial planes to create the illusion of a three-dimensional object from projected light. These improvements provide for a more natural depth perception and increase visual comfort for the viewer.
The Future of 3D Image Technology
Advancing 3D imaging technology is essential for a variety of emerging applications, including augmented reality, virtual reality, holography, and immersive visualization. The rapid degradation of image fidelity caused by light bleed has been one of the primary issues impeding progress in these fields.
So far, the researchers have demonstrated their multi-plane snapshot image projection with a model of volumetric scenes comprising 28 axial slices. They also constructed a simpler real-world prototype to validate the model, made of only two planes. Experiments with the prototype corroborated the model’s results, producing outputs that matched the simulated numbers and achieved a similar image quality. Additionally, when the team ran their prototype against a projection without a diffractive decoder, their device exhibited considerably superior performance.
Between the simulations and laboratory tests on a physical prototype, the team has demonstrated that their work is scalable yet compact, enabling a 3D display with high axial resolution. In their current form, the researchers see near-eye displays, microscopy, and volumetric optical computing as the primary use cases, but continued development could open possibilities for multi-perspective holography and energy-efficient 3D systems.
Ryan Whalen covers science and technology for The Debrief. He holds an MA in History and a Master of Library and Information Science with a certificate in Data Science. He can be contacted at ryan@thedebrief.org, and follow him on Twitter @mdntwvlf.
There’s still no word on the final launch price, however. | Image by Xreal
The Project Aura glasses collaboration between Xreal and Google is now one step closer to being something you can buy. Reservations for the second Android XR device, now dubbed the Xreal Aura, are available for $99 starting today, with a full launch in the US, UK, Japan, Canada, and South Korea expected sometime this Fall.
The Xreal Aura follows Google's first Android-powered XR device, the Samsung Galaxy XR headset, which launched at $1,799 in October 2025. We still don't have a full price for the Aura - which Google has previously described as "a headset masquerading as glasses" - but Best Buy will be the first in-store retail partner whe …
Login
