There is something seductive about asking AI, “What did we decide?”, “What did I say about this before?”, or even, “Remind me what I was thinking.” Perhaps you have already done it. A meeting blurs, an old exchange disappears into a message thread, or you remember reaching a decision but not why. Instead of searching through the evidence, or your own memory, you ask the machine. Seconds later, it hands back a neat account. As AI becomes more personalized, it may know the trail of our conversations and choices better than we consciously recall it. That feels like better memory: The machine remembers so we do not have to. Yet AI may do more than remember for us. Through what it selects, connects, omits, and reconstructs, it can begin to influence what we remember.
That reaches beyond cognitive independence into something more personal: identity. Think of the stories you tell about yourself: why you left a job, how a friendship changed, what you learned from a failure. Such memories help explain, to ourselves as much as to others, who we are. What we remember about what happened, and why, becomes part of the person we understand ourselves to be.
We are witnessing the rise of our hybrid alter ego
In Latin, alter means “other” and ego means “I” or “self”: literally, an “other self.” A diary preserves what we experience and care to remember; a photograph freezes a fleeting moment as a lasting snapshot. AI does more. It can connect scattered records to return a coherent account of what happened and what it meant. Imagine an assistant that has accompanied you for a long time and read thousands of your messages. When you ask, “Why did I make that choice?”, its answer may sound remarkably like you. Over time this other self becomes ever better at reconstructing us, and the better it becomes, the more we rely on it. It is a relationship that begets the question whether we will eventually forget where its version of our past ends and our own memory begins.
Our memory reconstructs experience, which is not the same as replaying an untouched recording. Siblings remember the same childhood event differently; colleagues leave the same meeting with different accounts. Later information and retelling can alter what feels like recollection. Generative AI adds a fluent new source of post-event information, filling gaps in language coherent enough to become familiar. In simulated witness interviews, generative chatbots induced more than three times as many immediate false memories as control conditions. Experiments with AI-edited media likewise found increased false recollections and confidence in them, while another controlled study found that misleading information introduced through chatbot dialogue increased false recollection in humans. This does not mean everyday AI is rewriting our internal autobiographies, yet. The implication is subtler: something supplied by an artificial interlocutor can later feel like something we remember.
When the source becomes blurry
This creates a problem: Do you know where your thoughts came from? Do you actually remember that detail? Or did AI suggest it during one of the many conversations you no longer recall having? Different humans remember different versions of the same event. The so-called Mandela effect has been intensified with 24/7 exposure to (fake) information. AI changes the conditions further, because beyond visual influence its contribution arrives through dialogue, often in language tailored to our history and way of speaking. After enough repetition, a sentence generated for us may begin to feel like a sentence generated by us. Picture an old disagreement with someone close to you. Years later, you ask AI, “What actually happened between us?” It retrieves old messages and notes. Much may be accurate; some may be inference. Yet its version is cleaner than the fragments in your head. Eventually that polished account becomes easier to retrieve than your original memory. Are you remembering the argument—or the story the machine assembled about it?
The stakes reach into cognitive ownership: when we delegate parts of thinking to AI, do we still experience the result as genuinely ours? Memory makes the question more consequential. Our remembered past influences whom we trust, what we regret and which choices feel consistent with “who I am.” An artificial alter ego that increasingly retrieves and interprets that past enters a space deeper than information management. It becomes one of the voices involved in constructing our identity.
An alter ego should remain the other self
None of this makes AI the enemy. The interaction with generative AI can help us recover a forgotten idea, reconstruct a project chronology or help us navigate years of digital clutter. The important distinction is between retrieving evidence and authoring meaning. Asking AI, “Find the messages from that period so I can work out what happened” differs from asking, “Tell me what happened.” One helps us recover the record. The other invites the machine to co-author the story.
That distinction will become harder to protect as artificial memory improves. AI's summaries may become cleaner and easier to access than our partial recollections. Yet those untidy fragments are not defective data awaiting optimization. Hesitation, disagreement and the effort of reconsidering what we once thought are part of how humans make meaning. Outsource too much of that work and the result may be a beautifully coherent story of ourselves whose authorship has become difficult to trace.
Protect your EGO
We do not need to stop using AI as external memory. We need habits that keep us in the loop. A simple EGO discipline can help.
E — Experience before outsourcing. After an important meeting, conversation or decision, form your own impression before asking AI for its version.
G — Guard the source. Distinguish what you remember from what you inferred, what the evidence shows and what AI suggested. For consequential memories, return to primary evidence.
O — Own the meaning. Let AI retrieve, compare and challenge while keeping the final interpretation meaningfully yours. You should still be able to say: this happened; this part is uncertain; this was suggested to me; and this is what I think it means.
As our “other self” is getting better at remembering us, our task is to ensure that convenience does not gradually turn retrieval into interpretation, and interpretation into authorship. Our memories have always been imperfect. They are also among the materials from which our self arises.