Veritas Silentii Conversations Across Time
Conversations Across Time
Conversation No. 5 · September 2026
Claude Shannon and the Dangerous Road to Monotony
Some of the most important signals in our lives first arrive disguised as noise.
Imagine two young mothers meeting for a morning gym class sometime in the 1960s.
They know each other well. They live in the same neighborhood. Their husbands know each other. Their children play together. Their lives overlap enough that much of what needs to be said does not actually need to be said.
The conversation might sound something like this:
A: You?
B: OK.
A: Kids?
B: Sniffles.
A: Bob?
B: Traveling.
A: Where?
B: Head office.
A: Lunch?
B: When?
A: Noon?
B: Good.
A: Alice too?
B: Yeah.
A: Great.
To someone listening in, it barely sounds like a conversation.
Whole sentences are missing. Context is assumed. Predictable words have disappeared. To an outsider, exchanges like this, repeated day after day, might seem boring…even monotonous.
Yet for the two women, that predictability is what makes the conversation effortless. They understand each other perfectly because so much no longer needs to be said.
Now imagine one more exchange.
A: Oh, one more thing.
B: What?
A: I'm divorcing Bob.
Three words.
And suddenly almost everything changes.
Those three words contain more meaningful new information than the entire conversation that preceded them.
Why?
Because nobody expected them. They weren't predictable…just more of the same. They are a surprise.
And that brings us to one of the most consequential thinkers of the modern technological world.
Claude Shannon. The Man Who Invented Information Theory
In 1948, while working at Bell Labs, Claude Shannon published A Mathematical Theory of Communication.
It became the foundation of what we now call information theory.
Shannon was not primarily concerned with whether a message was profound, useful, truthful, interesting, or even intelligible to a human being.
He was trying to solve an engineering problem.
How much information can be transmitted through a communications channel? How do we measure it? How reliably can we transmit it when the channel itself may introduce noise?
And in answering those questions, Shannon did something extraordinary. He gave us the mathematics of information.
At its simplest, one of his results is almost counterintuitive:
The more predictable an event is, the less new information its occurrence conveys.
If I tell you tomorrow that the sun rose this morning, I have told you something true. But I have given you almost no new information.
You expected it.
If I tell you that the sun did not rise this morning, something very different happens. An extraordinarily improbable event has apparently occurred.
That is why our little conversation matters.
"Bob? Traveling again." Almost completely predictable.
"Kids? Sniffles." No great surprise there either.
But: "I'm divorcing Bob." Predictability collapses, uncertainty takes center stage, and surprise explodes.
And so does the new information conveyed by those three words.
But beneath Shannon's idea was something even more interesting.
A Strange Word Called Entropy
Shannon needed a way to measure uncertainty. And to do that, he borrowed a concept from nineteenth-century physics and thermodynamics…an idea called entropy.
Very roughly, entropy is a measure of uncertainty. If only one outcome is possible, there is very little uncertainty…and therefore low entropy. If many outcomes are possible and we don't know which will occur, uncertainty rises.
And so does entropy.
Shannon made a brilliant connection. He realized that the same mathematical idea could be applied to communications.
He called it information entropy.
Imagine an ordinary deck of cards arranged perfectly by suit and number. Turn over one card and, if you know the ordering, you can predict which cards surround it. Very little uncertainty. Very little surprise.
Now throw the deck into the air, gather the cards randomly from the floor and put them back together.
Turn over one card. What comes next?
Almost any of the remaining cards could appear. You have much less ability to predict the outcome.
More uncertainty.
Higher information entropy.
This relationship between uncertainty, prediction and information led Shannon in a brilliant direction.
In the late 1940s and early 1950s, he began applying these ideas to the predictability of ordinary English.
People were shown preceding text and asked, in effect:
What letter do you think comes next?
The more context they were given, the better their predictions became.
Shannon used experiments like these to estimate the entropy and redundancy of English.
Think about that for a moment.
In 1951, Claude Shannon was studying how knowledge of what came before helps us predict what comes next.
More than seventy-five years later, we are building AI enabled agents, the greatest prediction machines in history…doing something remarkably similar, but at a scale Shannon could never have imagined.
And that makes Shannon unexpectedly relevant again.
The Agentic World
Much of modern AI is built around an extraordinarily powerful capability: prediction.
What word is likely to follow this one? What product are you likely to buy? What movie will you probably enjoy? Which candidate is most likely to succeed? Which customer is likely to leave? Which route are you likely to prefer? Which person should you meet? Which opportunity fits your background?
And increasingly: what should you do next?
The better an agent understands us, the more personal it becomes and the better some of its predictions can become.
It learns our histories and preferences…our purchases, relationships, successes, failures and habits…and perhaps eventually our ambitions, fears and appetite for risk.
We generally describe this as progress. And most of the time, it is. Like the conversation between our two young mothers at the beginning of this article, prediction can remove enormous amounts of friction from life. It can be comfortable.
We don't want to evaluate 5,000 restaurants before dinner. We don't want to read every available document before a meeting. And we don't want to examine every flight, every hotel, every investment, every candidate or every possible route ourselves.
We want the machine to separate what matters from what doesn't, to find what's important, and eliminate what appears to be noise…irrelevant.
But if Shannon were here today, he would probably have asked:
What if what looks like noise to your agents isn't really noise at all?
In Shannon's engineering world, surprise and noise are not the same.
Noise is interference that corrupts a signal. We want to eliminate as much noise as possible.
A surprising message, something unexpected, may actually contain a great deal of information.
That distinction matters. Because something destructive in one context may be valuable in another.
AI agents are now taking the idea of noise that was born in communications engineering and using it to make judgments about human relevance.
It may be deciding which restaurant deserves to be shown to you. Which candidate deserves consideration. Which article deserves your attention. Which opportunity fits your history. Which strategy looks promising enough to advance.
And from the perspective of a predictive system deciding what deserves our attention, something genuinely novel and surprising may initially look remarkably like irrelevant noise.
The Surprise That Changes Everything
Consider how many important events in life begin with something that initially appears to have little relevance:
- A conversation you didn't intend to have.
- A person you weren't supposed to meet.
- A book unrelated to the problem you were trying to solve.
- A job you probably shouldn't have taken.
- A trip that went wrong.
- A failed business.
- An unsuccessful experiment.
- A wrong turn.
- An idea from an industry completely unrelated to your own.
Most lead nowhere important or life changing. And that's important to remember because most unexpected events are noise…and most noise really is, well, just noise.
But occasionally something strange can happen.
The irrelevant person who introduced you to someone who changes your career.
The book you picked up for no particular reason gives you the idea that becomes your business.
The wrong turn becomes the right one.
Looking backward, we call these moments serendipity, or just plain luck.
But looking forward, they have a different characteristic.
They contained possibilities we could not yet predict.
And that creates a problem for a world increasingly organized around prediction.
The Future We Are Most Likely to Want
Suppose your personal AI agent knows almost everything relevant about you.
It knows what you've done. What you've enjoyed. What you've rejected. What you've purchased. Who you know. Where you've traveled. What you've read. Which risks you've taken. And perhaps it also knows the behavior of millions of other people that are similar to you.
Its job is to help. So, it predicts a future that makes sense given your past and the past of people like you.
It recommends the restaurant you are most likely to enjoy. The article you're most likely to read. The person you're most likely to find useful. The investment consistent with your risk profile. The job most compatible with your career. The vacation resembling vacations you've enjoyed before.
Every recommendation may be excellent. But something may still be missing. Because the machine is necessarily reasoning from evidence about who you have been, not what is possible because of something that has never happened to you before.
Something as unexpected as your friend saying:
"I'm divorcing Bob."
That creates a strange paradox:
The better a machine becomes at predicting the future we are likely to choose, the easier it may become to miss the future we never would have predicted.
And that future lives in a peculiar place.
It's a place called OUTSIDE.
Outside the highest probability, the obvious recommendation based on what the machine already knows about you.
And sometimes in what it judges to be noise outside everything that initially appears relevant at all.
In other words, it may live among possibilities your personal agent concludes are not worth advancing. Improbable. Unnecessary. Inefficient. Irrelevant. Accidental. And strange.
Take a moment to think about your own background and you may discover something surprising…and not noise at all.
Many of the turning points in successful lives occur precisely in these surprising places.
The unexpected conversation. The unlikely person. The book they almost didn't read. The job they never intended to take. The idea that made no sense at first. The wrong turn that somehow became the right one.
These are places where novelty enters our lives. Where growth begins. Where opportunity appears. And sometimes where an entirely different future first becomes possible.
Sometimes Writers Understand This Better Than the Rest of Us
Near the end of John Steinbeck's classic novel East of Eden, seventeen-year-old Abra is talking to her friend Cal about growing up.
It's 1918. The world is at war. They are living during uncertain times.
As children, she tells him, they had lived inside comfortable, predictable stories they had created for themselves.
But eventually that was no longer enough. Reality had become larger than the story.
Aron, her former boyfriend, could not make that transition. He needed life to conform to the story he had already constructed for it.
It was safe. He couldn't tolerate another ending.
Abra wanted something different. She didn't need to know how the story would come out.
She just wanted to be there while it was happening.
There is something remarkably modern in that century-old idea.
An AI agent learns the story of you. Your history. Your preferences. Your habits. Your relationships. Your purchases. Your ambitions. The decisions you have made before.
And then, quite reasonably, it uses that story to predict what you are likely to want next.
The better your agent becomes, the more coherent the story may become.
And perhaps it becomes harder for something outside the story to enter…because prediction has a temporal bias toward the past.
That may be one of the subtler risks of extreme agentic personalization.
Not that the agent misunderstands who you are. Instead, that it understands who you have been so well that it keeps showing you who you are likely to remain…while possibilities that do not fit the story become progressively easier to overlook.
Growing up…personally, professionally, intellectually…has always required something else. A willingness to leave parts of the past behind. To let the story change. To encounter something that doesn't fit. To choose something your past could not have predicted. Not always to know how it comes out.
And, as Abra decides, simply to be there while it is happening.
Shannon Meets Simon
This is where our last Conversation Across Time, with Herbert Simon, unexpectedly meets this one.
Simon, the 1978 Nobel Prize winner, told us that a wealth of information creates a poverty of attention.
There is simply too much information and too little human attention.
As a result, making decisions requires filtering.
Claude Shannon gives us another way of looking at the problem.
The predictable tells us relatively little that is new. The unexpected, the surprises, can carry far more information.
Put those two trailblazers in a room to discuss AI and Simon might tell us:
Agents can help us make decisions by reducing the overwhelming world of information.
But Shannon might whisper something different:
Be careful what you classify as irrelevant before you know what it means.
Increasingly, the models making those judgments will be personalized to us. Which means the machine may become extraordinarily good at identifying what appears relevant for the person it believes we are.
And that may be exactly what we want.
Until someone tells you:
An opportunity capable of changing who you are arrived yesterday…but your agent decided it wasn't relevant enough to show you.
Now Let's Turn Our Attention to Business
So far this has been about individuals.
But a business is not really a building or a balance sheet. It's a living system, run by people, making thousands of small decisions about what to say, sell and pursue next.
Increasingly, those decisions run through predictive systems trained not merely on one person's history, but on aggregated customer searches, questions, purchases and behavior.
Ask the machine what customers want and it can tell you what they've searched for, asked about, clicked and bought.
Ask what content performs and it can identify what has already performed.
Ask how to be found by AI search and answer engines and increasingly we optimize for the characteristics that appear to make existing content visible.
Each decision makes sense. But do this long enough, at scale, across an entire industry, and something else can happen.
The channel begins optimizing itself toward monotony.
Research into AI-driven hiring has already raised a related concern. It's called algorithmic monoculture, where different organizations relying on similar systems or signals produce correlated decisions and blind spots.
But there's no reason to assume the underlying dynamic must stop with hiring. It can appear wherever organizations increasingly use similar predictive systems trained on similar histories to answer similar questions.
What do customers want?
What should we make?
What should we say?
What should we hire?
What should we recommend?
What is most likely to work?
None of this requires conspiracy. Every individual decision can be defensible…even smart. Match the format that performs. Follow the signals customers are already sending. Do more of what the data says is working.
But a business built entirely this way risks doing to itself what the personalized agent can do to an individual when it reasons too completely from precedent.
It becomes less capable of imagining a potentially successful strategy for which no successful precedent yet exists.
I call this Agentic Monotony: the condition that emerges when a company's strategy and voice are increasingly shaped both by what predictive systems say customers want, and for what those same systems are likely to reward.
Eventually, the two loops begin to close on each other.
And everyone gets better at moving toward the same place.
Every decision leading there may appear entirely rational. But competitors are doing the same thing…using similar agentic systems, trained on similar industry histories, responding to similar customer signals.
And then something unexpected happens.
The better everyone becomes at predicting what will work, the more everyone begins to look alike.
The strange part is that this can happen to a company that is winning by every visible metric.
But the cost is invisible:
The strategy that might have emerged if nobody had gone looking for what was already popular first.
Think for a moment about some ideas that changed entire categories.
- A talking gecko selling automobile insurance.
- A woman with a sledgehammer running into a room of gray conformists to introduce the Apple Macintosh.
- A pink mechanical rabbit beating a drum to sell batteries.
- Volkswagen telling Americans accustomed to enormous automobiles to Think Small.
Today these ideas are part of the historical record. And predictive systems have already learned from their success.
But imagine evaluating them before they happened. What existing data would have predicted them? What successful precedent would have justified them? What model of existing consumer behavior would have ranked them as the highest-probability answer?
And that is precisely the point.
Breakthrough ideas often look obvious only after someone has had the courage to do the improbable thing first.
And that is the danger hidden inside Agentic Monotony.
Courage to choose never becomes an issue if the improbable option never makes it into the consideration set.
Perhaps We All Need Some Uncertainty
There is an extraordinary historical irony here.
For more than seventy-five years, Shannon's mathematics has helped us build systems capable of communicating information more efficiently and reliably in the presence of uncertainty and noise.
Now we are beginning to build systems capable of reducing uncertainty in an entirely different way: by predicting what is likely to happen next.
And perhaps the next challenge for truly intelligent agents will be learning not only how to predict us extraordinarily well…but when not to.
Perhaps we should deliberately ask them:
- Show me something outside my normal interests.
- Introduce me to someone you would ordinarily consider irrelevant.
- Give me an option that doesn't fit my history.
- Show me the low-probability possibility.
- Tell me what you nearly excluded from consideration.
Perhaps the most sophisticated personal agents won't simply know us extraordinarily well. They will know when to surprise us…somewhere between maximum uncertainty and maximum predictability. And it should know when each limit matters.
The Shannon Echo
That leaves us with a practical habit suggested by the thinkers who never encountered the world we're now building. When the machine gives you exactly what you expected, occasionally ask:
- What surprised you?
- What did you encounter that didn't fit?
- What seemed irrelevant?
- What had a low probability but potentially large consequence?
- What did you nearly exclude?
And occasionally:
Show me something you don't think I will like.
Most of the time, the system will probably be right, and the discarded possibility will be irrelevant.
Move on.
But occasionally, something else may happen. The machine will show you the equivalent of those three unexpected words in an otherwise predictable conversation.
And the world you thought you understood may suddenly open into possibilities neither you nor the machine had predicted.
For more than seventy-five years, we have been teaching machines to reduce uncertainty.
The Agentic Age may require us to learn something more difficult:
When not to.
Because some of the most important signals in our lives, and in our businesses, may first arrive disguised as noise.
What Inspired This Conversation
This Conversation draws principally on Claude Shannon's 1948 paper "A Mathematical Theory of Communication," his 1951 paper "Prediction and Entropy of Printed English," John Steinbeck's East of Eden, and research into algorithmic monoculture and AI-assisted decision-making.