I remember phone books.
Not metaphorically. Actual phone books.
Every few years, an enormous printed directory containing the names, telephone numbers, and home addresses of almost everyone in your city would simply arrive on your doorstep. Nobody had opted in. Nobody seemed particularly alarmed. If you wanted to know where someone lived, you looked them up.
The remarkable part is that almost nobody thought of this as a privacy violation. Because the social architecture around the information was different.
The information was public, but frictional.
Someone could find you, but they had to want to find you. They needed your approximate location. They needed to know your name. They had to get the right phone book, look you up, write the information down, and then actually do something with it.
There was no search history. No aggregation. No automated correlation. No behavioral profile silently accumulating behind the act of looking someone up.
Today, the idea seems almost absurd.
My Substack - Some Assembly Required - has always been where I work things out in public. It’s where the ideas are longest, sharpest, and least edited for palatability.
We have become extraordinarily sensitive to what information about us is captured, stored and made accessible. We debate cookies and location tracking. We cover laptop cameras. We argue about facial recognition, data brokers and whether our phones are listening to us.
Privacy has become one of the defining anxieties of the digital age.
But I wonder whether we have been worrying about the wrong thing. Or at least, only half of it. The technological revolution underway isn’t simply creating a world in which more information about us is captured.
It is creating a world in which less of our lives will ever be forgotten. And I’m not sure we have begun to understand what that might do to us. Because forgetting isn’t simply the failure of memory.
It may be one of the things that makes us who we are.
We Are Not What Happened to Us
An assumption is buried in the way we talk about identity: that we are the sum of our experiences. The older I get, the less aligned I am with this.
I am not the sum of my experiences. I am the sum of the experiences I remember, and the interpretation of those memories.
An enormous amount has happened to you that you cannot remember. Thousands of breakfasts. Conversations. Arguments. Commutes. Rooms. Faces. Feelings. Entire ordinary Tuesdays have disappeared.
Even the things you do remember aren’t stored as pristine recordings. Of my past relationships, I retain a reasonably generous edit: two good people who, in the end, found themselves in difficult circumstances.
The other person may carry a different edit. And what actually happened, if it could somehow be reconstructed in full, might flatter neither of us. We may not have been as good—or as bad—as each of us remembers today.
Memory edits.
It compresses. It exaggerates. It combines. It changes emphasis. It removes context. It reconstructs events using information acquired years after they happened.
A relationship that lasted ten years becomes five or six scenes. A childhood becomes a collection of fragments. A career becomes a handful of triumphs, failures and people who mattered.
Sometimes the editing is generous to us. Sometimes it isn’t. But slowly, almost invisibly, those selected fragments become a narrative. And that narrative becomes us.
Ask someone who they are, and they won’t give you a database. They’ll tell you a story. Where they came from. What happened to them. Who loved them. Who hurt them. The chances they took. The mistakes they made. The things they overcame.
Identity is autobiographical. And autobiography requires editing. Until now, forgetting has been the most ruthless editor we’ve ever had.
The Great Human Compression Algorithm
We tend to describe forgetting as a weakness. I forgot his name. I can’t remember where we went. The details are fuzzy. She remembers it better than I do.
The language itself suggests malfunction. Memory would ideally be complete; unfortunately, the biological hardware isn’t quite good enough.
We have spent thousands of years trying to fix the problem. Writing externalized memory. Painting preserved faces. Photography preserved moments. Film preserved movement. Audio preserved voices. Computers preserved information. Cloud storage made preservation effectively limitless.
And now AI is making it all retrievable.
That last development changes the equation.
An archive containing a billion pieces of information isn’t really memory if nobody can find anything inside it. Until recently, this provided a strange kind of protection. We were already capturing extraordinary amounts of our lives, but most of it disappeared into digital attics.
Twenty thousand photographs. Fifteen years of email. WhatsApp conversations with people we no longer speak to. Calendar entries. Search histories. Documents. Voice notes. Videos. Location records.
The information existed, but practically speaking, much of it was forgotten.
AI changes that because it doesn’t merely increase our capacity to store the past. It increases our capacity to retrieve and interpret it. Suddenly the attic has someone living inside it, and they have unpacked and cataloged everything.
The Witness Who Never Forgets
Imagine having an AI companion for forty years. Not today’s chatbot that vaguely remembers a preference you mentioned three conversations ago.
Something persistent.
It has read your messages, with permission. Seen your photographs. Managed your calendar. Helped write your documents. Listened to meetings. Watched your health data. Organised your finances. Travelled inside your phone. Perhaps eventually seen much of what you see and heard much of what you hear.
At 70, you might have an intelligence beside you that has effectively witnessed your adult life.
And unlike your spouse, children, friends or colleagues, it hasn’t forgotten.
You say: “I’ve always been good under pressure.” It replies: “That’s interesting. Your behaviour suggests otherwise.”You say: “I was always there for my children.” It can retrieve the calendars.You say: “That was the happiest period of my life.” It has the messages you wrote while you were living it.You tell a favorite story about something that happened in 2034. It says: “That’s not quite how you described it at the time.”At first this sounds useful. Then it starts to sound disturbing. Because what happens when your memory and the record disagree?
Who Gets to Decide Which Version of You Is True?
Imagine a father who genuinely remembers himself as present and loving during his daughter’s childhood. His daughter remembers something different.
Today, there is no definitive resolution. His memory contains one childhood. Hers contains another. Both have been edited through emotion, perspective, and time.
But imagine an AI capable of reconstructing those years. It knows how many evenings he arrived home after she went to bed. How often he cancelled weekends. How frequently he called when traveling. It can search photographs, calendars and messages. It can identify patterns neither person remembers.
Perhaps it concludes:
“Between the ages of eight and twelve, your father canceled 31 percent of the weekends you were scheduled to spend together.”Something extraordinary has happened. The argument is no longer between two memories.
There is now a third witness. The archive. And the archive has receipts. But even then, has it discovered the truth?
Thirty-one percent is a fact. It isn’t necessarily a childhood.
Perhaps the weekends he was there were extraordinary. Perhaps the daughter experienced his absence differently at different ages. Perhaps his work prevented financial catastrophe. Perhaps one conversation mattered more than a hundred missed dinners.
Data can reconstruct events with extraordinary precision while still failing to reconstruct meaning. Which leaves us with a deeply uncomfortable possibility. The machine may eventually remember our lives more accurately than we do while understanding them less completely. And yet we may still defer to it because it has evidence.
Memory Has Never Been a Courtroom
This is where perfect memory collides with something fundamentally human. We don’t simply remember events. We reinterpret them.
Something that humiliated you at 25 becomes funny at 45. A relationship you considered disastrous becomes, in retrospect, the experience that taught you what love required. A career failure becomes the moment everything changed. Someone you once hated becomes someone you understand.
Memory isn’t static because we aren’t static. The meaning of the past changes as the person remembering it changes. That may be one reason memory works the way it does. Its instability allows the past to remain psychologically negotiable.
We can forgive. We can contextualize. We can soften. We can occasionally deceive ourselves. And we can move on.
A permanent searchable archive introduces something humans have rarely had to contend with: the past refusing to change with us.
The Right to Become Someone Else
Think about adolescence.
Almost everyone who grew up before social media benefited from a remarkable privilege we rarely recognized at the time.
Most of our stupidity disappeared. The terrible haircut survived in three photographs. The offensive joke disappeared when the sound waves dissipated. The drunken argument became an unreliable story among friends. The ridiculous opinion you passionately defended at seventeen vanished unless somebody happened to write it down.
You were allowed to become someone else partly because there wasn’t enough evidence to continuously prosecute you for who you had been.
That condition is disappearing. A child born today may accumulate a searchable digital history before they can speak. By adolescence, there may be thousands of hours of photographs and video. Messages. Posts. Searches. School records. Location data. AI conversations. By middle age, there could be enough material to reconstruct an astonishing percentage of a life.
We usually frame this as a privacy problem. But perhaps it is also an identity problem.
What happens to reinvention when every previous version of you remains available? What happens to forgiveness when every offence can be replayed? What happens to growth when contradiction can always be exposed?
“You say you believe this now. But here’s what you said twenty-three years ago.”We tend to celebrate consistency as a virtue. But a human being who never contradicts their younger self isn’t necessarily principled. They may simply have stopped evolving. The ability to leave previous versions of ourselves behind may be essential to becoming someone new.
From Voluntary Recall to Algorithmic Recall
Another shift worries me even more. Historically, remembering has largely been initiated by humans.
A smell reminds you of your grandmother. A song transports you back to university. You open an old photograph. A friend says, “Do you remember when…?”
Memory arrives through association, accident or intention. AI creates the possibility of something different: memory initiated by someone — or something — other than you.
Your AI assistant notices a pattern.
“This isn’t the first time you’ve reacted this way.”It retrieves an argument from fourteen years earlier.
Your relationship AI notices that you and your partner are repeating a conflict from 2037. Your health system identifies that your behavior resembles the months preceding a previous depressive period. Your professional AI reminds you that the strategy you’re advocating contradicts something you argued in a meeting six years ago.
These could be incredibly valuable interventions. They could also fundamentally alter our relationship with the past. Because now memory isn’t merely something we possess. It is something that can be done to us.
An algorithm can decide which version of our past is relevant to our present. That is an extraordinary amount of power.
Privacy May Not Be the Biggest Question
For decades, we’ve asked who should be allowed to collect our data. That’s still an important question. But the age of AI introduces another:
Who should be allowed to remember us? And for how long?
Should your AI remember everything you’ve ever told it? Should you be able to ask it to forget? Should some memories decay automatically? Should an AI distinguish between factual memory and emotional memory? Should your teenage conversations survive into middle age? Should your employer’s AI be permitted to remember mistakes after the humans involved have forgotten them? Should a relationship have a shared memory? Should a family?
And what happens when two people’s AI memories disagree?
These sound like technology governance questions. I suspect they will eventually become philosophical ones. Because beneath them sits something we have barely had to articulate before:
Is forgetting a human right?
Perhaps Forgetting Was Never the Bug
There is something wonderfully human about the fact that memory is incomplete.
Two old friends can sit at a table and argue about something that happened forty years ago.
“No, no. That’s not what happened.”
“Of course it was.”
“You weren’t even there.”
“I absolutely was.”
Eventually they laugh.
There is no server to consult. No transcript. No definitive replay. The past remains partly theirs. There is ambiguity in it. And ambiguity creates room for interpretation, forgiveness, mythology and meaning.
I’m beginning to wonder whether we have misunderstood forgetting because we have evaluated it like engineers. We saw missing information and assumed failure. If complete memory were an uncomplicated evolutionary advantage, we might expect humans to remember more than they do.
Instead, in the end, humans forget the majority of the things that happen to them. Perhaps that limitation carries benefits we have never needed to name because we have never seriously threatened it before.
Forgetting allows emotional wounds to soften. It allows insignificant information to disappear. It allows experiences to become stories. It allows stories to become meaning. And perhaps most importantly, it allows identity to remain editable.
The Last Thing We May Need AI to Learn
There is a wonderful irony here.
Human civilization can be understood, in part, as a war against forgetting. We carved stories into stone because storytellers died. We invented writing because memory failed. We painted portraits because faces disappeared. We invented photography because moments vanished. We built databases because organizations forgot. We built the internet because information was difficult to retrieve.
And now we are building artificial intelligence capable of making almost everything searchable, interpretable and persistent.
We may finally be approaching the thing we’ve been pursuing for thousands of years: something close to perfect memory. And when we get there, we may discover that perfect memory was never the objective.
The objective was meaning. Those are very different things.
The great challenge of artificial memory may therefore not be teaching machines how to remember more.
It may be teaching them what humans have always known instinctively: what deserves to be remembered, what should be allowed to fade, and what must eventually be forgotten.
Because our identities were never constructed from everything that happened to us. They were constructed from what remained.
And if forgetting becomes optional, we may discover that the most important question of the AI age isn’t whether machines can remember us.
It is whether they will allow us to forget.
