When Your AI Knows You Better Than You Know Yourself
By John R. Kennedy ·
The emerging value of artificial intelligence, memory, and personal understanding
I have a collection of photographs that goes back decades. Some are carefully organized, while others are scattered across computers, phones, and storage devices I've accumulated over the years. Every so often, I'll come across a photograph I haven't seen in a long time, and it will bring back a memory I'd nearly forgotten.
What interests me isn't simply the photograph itself. It's everything associated with it.
I might remember where it was taken, what was happening in my life at the time, or why that particular moment mattered. Someone else looking at the same photograph might see an ordinary family gathering, while I see a collection of relationships, experiences, and emotions that give the image an entirely different meaning.
The photograph contains information, but much of its value comes from the context surrounding it.
I've been thinking about that distinction as artificial intelligence becomes more capable of remembering our interactions and using what it learns to help us.
Today, many people use AI as though each conversation is an isolated event. They ask a question, receive an answer, and move on. Some systems can retain selected information across conversations, while others allow users to provide documents or connect external sources of information.
These capabilities are still developing, and different AI services handle memory in different ways. But the direction is interesting because it introduces the possibility of an assistant that gradually develops an understanding of the person using it.
Imagine having an AI assistant that has been helping you for ten years.
During that time, you've discussed your work, family, interests, and the decisions you've made. You've used it to explore unfamiliar subjects, solve problems, plan projects, and organize important information. Perhaps you've also shared photographs, correspondence, and other records that help explain the circumstances of your life.
Over time, the assistant might recognize connections between experiences that occurred years apart. It could remind you of an idea you explored long ago when a new opportunity makes that idea relevant again. It might notice that a problem you're struggling with resembles something you successfully resolved in the past.
That accumulated understanding could become extraordinarily useful.
But I think its value goes beyond making an assistant more convenient.
Consider how much of what we know about ourselves exists only in our memories. We remember experiences, develop opinions, recognize patterns, and gradually form an understanding of the world based on everything we've encountered.
Yet our memories are imperfect. We forget details, reinterpret past events, and sometimes lose track of why we made particular decisions.
An AI assistant with access to reliable historical information could potentially help us recover some of that context.
Suppose you're considering a career change. You remember being unhappy in a previous position, but the reasons have become less clear over time. An assistant that helped you through that period might be able to retrieve the conversations, notes, and circumstances surrounding your earlier decision.
Rather than simply offering generic career advice, it could help you revisit your own reasoning.
You might discover that your dissatisfaction had less to do with the work itself than with the environment in which you were doing it. That distinction could influence the decision you're facing today.
The assistant wouldn't need to know you better than you know yourself. It would simply need to help you reconnect with information and experiences that had become difficult to recall.
I find that possibility particularly compelling because it changes the role of technology from storing information to helping us understand the relationships between pieces of information.
For decades, we've used computers to preserve documents, photographs, messages, and other records. We've become extraordinarily good at accumulating information, although finding and organizing it remains a challenge.
Artificial intelligence introduces the possibility of making those collections more meaningful.
A photograph could become connected to the people, places, and events surrounding it. A business decision could be connected to the conversations and assumptions that influenced it. A family story could be preserved alongside the historical records and personal experiences that give it meaning.
Instead of maintaining isolated collections of information, we might begin developing something closer to a personal history that can be explored and understood.
But there's an important complication.
An AI assistant doesn't necessarily remember information the way a person does. Depending on how it's designed, it may store selected facts, retrieve earlier conversations, summarize past interactions, or infer preferences from patterns in our behavior.
Those processes can be useful, but they aren't equivalent.
A statement you made during a difficult week might be remembered as a lasting preference. An opinion you were exploring could be interpreted as a belief you firmly hold. A summary of a complicated conversation might preserve the conclusion while losing the uncertainty and disagreement that made the discussion valuable.
Over time, small misunderstandings could become part of the assistant's accumulated picture of you.
And because that picture influences future responses, the consequences might extend far beyond a single inaccurate memory.
Imagine an assistant that concludes you dislike taking financial risks because you once expressed concern about a particular investment. Years later, it might consistently recommend conservative options, even though your circumstances and priorities have changed.
You might never realize that an outdated interpretation was influencing its recommendations.
This is why I believe the distinction between remembering information and understanding it deserves careful attention.
A useful personal AI shouldn't simply accumulate more information. It should preserve enough context to distinguish between what happened, what someone said, what was inferred, and what remains uncertain.
It should also allow people to revisit and correct the understanding that develops over time.
After all, our identities aren't fixed collections of facts. We change our minds, learn from experiences, and sometimes become very different people from who we were years earlier.
An assistant that remembers our past without recognizing our capacity to change could become less helpful as its memory grows.
There's another aspect of this that I find especially interesting.
Much of the value of a long-term relationship comes from shared experience.
Think about someone you've worked with for many years. You may be able to communicate complicated ideas with relatively little explanation because you share a history of conversations, decisions, and experiences. You understand the circumstances behind certain preferences and recognize when something is unusual.
That shared understanding can make collaboration remarkably efficient.
An AI assistant could potentially develop a similar practical advantage through accumulated context, even though its relationship with us would be fundamentally different from a human relationship.
But that raises a question about where the value actually resides.
If an AI assistant spends ten years helping you manage a business, understand your health, or preserve family history, the accumulated context may become one of your most important digital assets.
Yet much of that understanding might exist inside systems operated by an organization you don't control.
Suppose you decide to change AI providers.
You might be able to download your conversation history or export certain records. But transferring the original information doesn't necessarily preserve the relationships and interpretations that made the assistant useful.
A new system might need to reconstruct years of context, and some of the original understanding could be difficult to recover.
We've experienced similar frustrations when changing software platforms, but the stakes could be different when the information involved reflects years of personal decisions, experiences, and relationships.
I don't believe people should have to remain with a particular AI service simply because leaving would mean losing the accumulated understanding they've developed.
At the same time, preserving that understanding isn't as simple as copying a collection of files. Memories may contain information about other people, private conversations, or conclusions that were never verified. Moving them between systems would require careful consideration of accuracy, permissions, and the interests of everyone involved.
These are difficult problems, but I suspect they'll become increasingly important as AI assistants take on more meaningful roles in our lives.
There's also something deeply personal about the idea of preserving accumulated understanding.
When someone dies, the people who knew them often discover how much knowledge disappears with them. Family stories, professional experience, explanations for important decisions, and countless small details may never have been recorded.
We preserve photographs and documents because we recognize their value, but those records often capture only fragments of a person's life.
Artificial intelligence could potentially help people preserve more of the context surrounding those fragments.
Imagine future generations being able to explore family stories alongside photographs, letters, and historical records. They might learn why a grandparent made a particular decision, how a family business began, or what life was like during an important period in their family's history.
Such a system wouldn't recreate the person, and it shouldn't pretend to. A collection of memories and records can never fully capture the complexity of a human life.
But it could preserve knowledge that might otherwise disappear.
I find that possibility meaningful because it reminds me that the value of information isn't limited to what it can accomplish for us today.
Some information becomes more valuable with time.
A conversation that seems ordinary now may help someone understand an important decision years later. A photograph that appears unremarkable today may eventually become one of the few remaining records of a person or place.
And the context connecting those records may be what makes them truly meaningful.
As artificial intelligence becomes more capable of remembering and interpreting our experiences, we'll have opportunities to preserve and use information in ways that previous generations couldn't have imagined.
But we'll also need to decide how much of that understanding we want to entrust to systems outside our control.
I don't think the objective should be to create an artificial intelligence that defines who we are. Our experiences, beliefs, and identities are too complicated to be reduced to a collection of records or predictions.
Instead, I hope we develop technology that helps us better understand our own history while preserving our ability to decide what that history means.
Because the greatest value of an AI that remembers us may not be that it knows us better than we know ourselves.
It may be that it helps us remember, understand, and preserve the things we never realized we might forget.
Continue reading: The Five Stages of Artificial Intelligence