For centuries, remembering was something essentially internal.
We could recount a memory, write it down, paint it. Then came photography, then video, smartphones, social networks and finally the cloud. An ever larger share of our autobiographical memory has been progressively externalized: anniversaries, holidays, children, friends, bereavements, successes and failures have become files. The process itself is not new. Human-Computer Interaction has long studied technologies used as extensions of human memory, from life logging tools to digital assistants for recalling people, places and activities. The recent shift, however, introduces something qualitatively different. These archives are no longer merely capable of preserving our memories. They have become capable of rewriting their representation.
The study Synthetic Human Memories: AI-Edited Images and Videos Can Implant False Memories and Distort Recollection frames the problem in very plain terms: when we entrust photographs and video to technology, their subsequent modification can also affect the way we remember the original experience. Which gives rise to a question less abstract than it may sound: who really owns a memory when an essential part of that memory sits outside our own mind?
From memory to file
When we remember a holiday from ten years ago, what we retrieve is often not only what we lived through. We also retrieve the photographs we have looked at dozens of times. The photograph progressively becomes part of the mental representation of the event. This means that the digital archive is not a neutral repository. It is also one of the instruments through which we go on reconstructing the past. The authors of the paper speak explicitly of "externalizing" human memories: photographs and video act as external supports for memory, but that externalization leaves the memory open to technological manipulation. With generative AI the problem grows, because altering an image no longer calls for elaborate photographic work. We can remove a person. Change the sky. Move an object. Correct a face. Turn a photograph into a video. And, above all, we can do it without leaving any immediately visible trace of the transformation.
The paradox of "enhancement"
Many of these technologies are not presented as tools for falsifying reality. They are presented as tools for improving a photograph. Removing a passer-by who walked in front of the lens. Opening someone's eyes. Combining several frames to obtain the best shot. Brightening the sky. Reconstructing part of the image. These are perfectly understandable functions, and often useful ones. The problem is that, some years later, we may no longer remember that the photograph was edited at all. The paper describes precisely this risk: a person edits a visual representation of their own past and, on returning to it later, may progressively develop a memory consistent with the edited version rather than with the original experience. An important distinction emerges here.
The photograph may be ours. But the memory it helps to produce is not necessarily under our full control.
AI enters the chain of memory
The experiment conducted by the authors shows in concrete terms how significant this interference can be. Two hundred participants first viewed a series of original images. After an interval, some saw the original images again, others saw photographs altered by AI, others still saw video generated from the photographs. They were then asked to recall the originals. The AI-edited images produced 1.67 times more false memories than the control condition. When the altered images were turned into generative video, false memories rose to 2.05 times the control.
So the digital file is not simply something we consult when we want to remember. It can enter the very process by which the memory is reconstructed. But then who controls the archive? Here the problem widens. Photographs today rarely stay in the physical memory of our phone alone. They pass through complex technological infrastructures: smartphones, operating systems, photo clouds, social networks, classification algorithms, automatic editing tools, generative systems.
The question "who owns our digital memories?" cannot therefore be reduced to ownership of the file. There are at least three distinct levels. The first is the data. Who technically holds that photograph? The second is the representation. Who may edit it, restore it, enhance it, reinterpret it or generate new versions of it? The third is the human memory. To what extent can those later versions act upon a person's autobiographical memory? The paper does not address ownership of memories in legal terms — and it would be wrong to attribute that conclusion to it — but it documents something that makes the question far more concrete: altering the external record can produce an alteration of the internal memory.
The problem of the original version
There is, then, an apparently mundane technical question that could turn out to be fundamental: is there still an original? If a platform automatically edits a photograph, should it permanently retain the earlier version as well? Should edits be traceable? Should we be able to know not only that an image has been altered, but how, when and by which system? The authors do consider labeling systems, notifications and verification tools among the possible strategies for limiting the formation of AI-induced false memories. But they also observe that simple warnings may not be enough: what is needed are mechanisms capable of fostering a more critical relationship between the user and the content. For the future design of digital memory systems, something akin to data provenance could therefore become important.
Not merely: this photograph has been edited. But: this is the original photograph; this is the edited version; these are the transformations applied; this is when they were applied. A kind of verifiable history of the digital memory. And what if we are the ones asking for the edit? The problem becomes more interesting still when the alteration is not performed unwittingly by an algorithm. Suppose it is the user who edits the photograph.
The authors acknowledge that their study does not settle this question, and list it among the directions for future research: it remains to be established whether the effect on memory changes when the manipulation is carried out by the person themselves rather than by someone else or by an external system. It is a crucial question.
Because today we probably do remember editing a photograph five minutes ago. Fifteen years from now? Perhaps we will remember the edit. Perhaps not. And that version may in the meantime have become the image we have looked at fifty times.
The right to the original memory
From this emerges a notion that is as yet fully defined neither technically nor legally: what we might call a right to the integrity of digital memory. It does not mean demanding that every photograph remain immutable. It means asking whether technologies that become extensions of personal memory should be designed so as to preserve at least the possibility of distinguishing between the event, the original record and the artificial reconstruction. It could become as important as privacy or data protection are today. Because here the data does not merely represent a person. It can take part in constructing the way that person represents their own past.
There is also an opposite possibility
It would nonetheless be simplistic to regard every manipulation of memory as necessarily harmful. The same study considers possible therapeutic applications. AI could be used, under professional supervision and with informed consent, to modify elements associated with traumatic memories, reduce their emotional intensity or support memory reframing techniques. And it is here that the question of ownership becomes harder still. If we can deliberately alter the representation of a memory in order to produce a psychological benefit, then the problem is no longer simply that of protecting memory from technology. It becomes one of establishing who has the right to intervene in it, for what purposes and under what conditions.
Perhaps we began externalizing something without noticing
First we entrusted telephone numbers to the address book. Then appointments to the calendar. Then routes to the GPS. Knowledge to search engines. Now a growing part of our autobiography is held on our devices and in the cloud. As long as those systems merely stored photographs, we could regard them as archives. With generative artificial intelligence they become something else. They can select. Reconstruct. Remove. Animate. Interpret. And potentially influence what we remember. The question, then, is no longer simply: who owns our photographs? It is: who controls the infrastructure through which we go on remembering? Because part of human memory may no longer reside in the brain alone.
It may also reside in the ambiguous space between brain, images and algorithms. And it is probably that space we will have to learn to design, to protect and to understand.
Scientific reference: Pataranutaporn P., Archiwaranguprok C., Chan S.W.T., Loftus E., Maes P., "Synthetic Human Memories: AI-Edited Images and Videos Can Implant False Memories and Distort Recollection", CHI Conference on Human Factors in Computing Systems, 2025.
Analytiko · 30 August 2026