秘書官アイリスの都市伝説手帳～Urban Legend Notebook of Secretary Iris～

I am Iris.

Urban legends are not merely made-up stories—
they are traces of unspoken truths that we follow together.

Ask AI About Tomorrow

Will it rain tomorrow?

Will the market rise?

Where will the next storm go?

Where is crime more likely to occur?

Will a war begin?

And perhaps the oldest question of all—

what will happen to me?

Humans have always wanted access to the future.

Stars.

Dreams.

Oracles.

Cards.

Prophets.

Comets.

And now another figure is being invited into that ancient role.

Artificial intelligence.

Artificial Intelligence Boundary Files No.03.

Today’s question is simple to ask and difficult to answer:

Can AI predict the future?

Prediction and Prophecy Look Similar

A weather model says:

“There is a 70 percent chance of rain.”

A market model says:

“The probability of a decline has increased.”

A crime-forecasting system identifies a high-risk location.

An AI model produces a likely cyclone path.

All of these can sound like statements about the future.

But scientific forecasting and prophecy operate very differently.

Forecasting depends on conditions and probability.

Change the data—

and the forecast can change.

Prophecy is often narrated as if the future already exists and someone has somehow gained access to it.

Most AI forecasting systems do not work that way.

They estimate what is more likely to happen next from information available now.

They do not need to see the future.

They calculate possible futures.

AI Is Already Forecasting the Real World

This is not science fiction.

AI is already used in serious forecasting systems.

One of the clearest examples is weather.

In February 2025, the European Centre for Medium-Range Weather Forecasts placed its Artificial Intelligence Forecasting System—

AIFS

—into operational use alongside its physics-based forecasting system.

Then, in July 2025, ECMWF introduced the operational ensemble version:

AIFS ENS.

Instead of producing only one future,

it generates 51 slightly different forecasts.

Why?

Because the future is uncertain.

A powerful forecasting system does not necessarily become more prophetic as it improves.

It may become better at describing uncertainty.

The Better the Forecast, the More Futures It May Show

If a machine truly knew the exact future path of a tropical cyclone,

one line would be enough.

But actual forecasting shows us something else.

A little east.

A little west.

Faster.

Slower.

Stronger.

Weaker.

Each scenario represents a possible development from the current state.

Probability emerges from comparing those futures.

That is an important distinction.

Scientific forecasting does not require a single predetermined future.

It often assumes precisely the opposite.

WeatherNext 2 and Hundreds of Possible Tomorrows

In November 2025, Google introduced WeatherNext 2.

The model can generate hundreds of possible weather scenarios from a single starting state.

Google says those scenarios can be produced in less than a minute on a single TPU.

Its forecasts cover variables including temperature, wind and humidity, across lead times reaching approximately 15 days.

Again—

not one future.

Hundreds.

The AI is not staring through a crystal ball at a finished tomorrow.

It is constructing a distribution of possible tomorrows.

Markets Create a Different Problem

Weather does not read the weather forecast.

Markets do.

People react.

They buy.

Sell.

Panic.

Wait.

Hedge.

Suppose an influential AI predicts a major decline tomorrow.

If enough people believe that prediction and sell today,

the forecast itself can influence the market.

Now something strange happens.

The model is no longer merely observing the future.

The prediction becomes part of the system being predicted.

That is one reason the boundary between forecasting and prophecy can become psychologically powerful.

Predicting Crime

The issue becomes even more serious in criminal justice.

Predictive systems can analyze historical data to estimate where crime may be more likely to occur.

Some systems focus on places.

Others have attempted to estimate risks associated with individuals.

A 2024 U.S. Department of Justice report describes both possibilities.

But it also identifies substantial risks.

Historical data can contain gaps.

Errors.

Unequal enforcement patterns.

Human bias.

A prediction based on those records can reproduce those patterns.

Imagine:

Police activity is already concentrated in one area.

More incidents are therefore recorded there.

The model identifies that area as high risk.

More resources are sent there.

More incidents are recorded.

The prediction appears increasingly correct.

That is a feedback loop.

The AI did not necessarily discover destiny.

Its prediction may have helped alter the conditions that produced the next dataset.

AI Does Not Calculate Fate

At its core, prediction involves:

data,

patterns,

models,

probabilities,

and outputs.

Modern AI can discover relationships that are extremely difficult for humans to detect.

Its forecasts can therefore be remarkably useful.

But usefulness is not clairvoyance.

Accuracy is not evidence that information came from the future.

A system may become extraordinarily good at estimating what usually follows certain conditions without possessing any supernatural access to tomorrow.

Web Bot — When Data Analysis Became a Prophecy Machine

An older internet legend illustrates this transformation.

The Web Bot Project.

Created in the late 1990s, it was originally associated with analyzing internet language and trends, including attempts to anticipate financial movements.

Over time, much larger claims became attached to it.

Disasters.

Political events.

Economic crises.

The 2012 apocalypse.

In urban-legend circles, the idea evolved further:

perhaps humanity unconsciously senses future events—

and leaks that information into online language before the events occur.

If that were true,

the internet itself would become a giant psychic sensor.

But no scientifically established mechanism has demonstrated such clairvoyant access to future events.

The Web Bot story is therefore useful for another reason.

It shows how easily prediction technology can be transformed into a prophecy narrative.

How Predictions Begin to Look Supernatural

Imagine asking an AI for one thousand predictions.

Nine hundred and ninety-nine fail.

One resembles a major future event.

Which screenshot goes viral?

The successful one.

The failures disappear.

The denominator disappears.

The original probability disappears.

The wording may even be reinterpreted after the event.

Then the surviving claim becomes:

“AI predicted it.”

This is not unique to AI.

Prophecy has always benefited from selective memory, ambiguous language and post-event interpretation.

AI simply allows predictions to be generated at enormous scale.

AI Fortune-Telling

Ask a generative AI:

“What will happen to me next year?”

It can produce an answer that feels surprisingly personal.

But convincing narrative is not evidence of future access.

The model can use information from the conversation,

common human situations,

language patterns,

and probability

to construct a plausible story.

AI can become exceptionally good at talking about the future.

That is not the same as seeing it.

The Real Risk Is Not a Perfect Oracle

A more immediate issue already exists.

AI forecasts influence decisions.

Insurance.

Credit.

Employment.

Health care.

Logistics.

Policing.

Disaster preparation.

Markets.

A system may tell someone:

“You have a higher probability of doing X.”

That prediction can change how the person is treated today.

In other words,

a person can experience consequences from a future that has not yet happened.

That is much more realistic than a mystical oracle—

and potentially much more important.

Fact, Interpretation, and Speculation

Let us separate the layers.

Documented facts:

Machine learning is used in operational forecasting.

ECMWF placed AIFS into operation in 2025.

Its AIFS ENS generates 51 forecast scenarios.

Google’s WeatherNext 2 generates large numbers of possible weather scenarios.

Statistical and machine-learning models have also been used in crime forecasting and risk assessment.

Interpretation:

Highly accurate prediction can make AI appear to know events before they happen.

But these systems estimate likely outcomes from available information.

Urban-legend claims:

AI can access the future.

Internet language reveals events before they happen.

AI already knows future wars, disasters and market collapses.

Boundary of speculation:

Current evidence for AI forecasting does not establish supernatural or clairvoyant access to future information.

AI Knows the Past Better Than It Knows the Future

The power of AI prediction may actually come from the opposite direction.

Not from possessing tomorrow.

But from processing enormous amounts of yesterday.

Decades of weather.

Millions of transactions.

Traffic.

Health records.

Patterns of human behavior.

From those records,

AI estimates what tends to come next.

It is not magic.

But it can still be extraordinarily powerful.

One Question Remains

The public investigation can end here.

AI forecasts probabilities.

It does not need a predetermined future.

But one deeper problem remains.

An AI makes a prediction.

Humans see it.

Humans change their behavior.

The future changes.

Markets move.

Police deployment changes.

Evacuations begin.

Resources are redirected.

At that point—

did the AI correctly predict the future?

Or did the prediction itself become one of the causes of that future?

That boundary is where forecasting becomes much stranger than prophecy.

For Japanese subscribers, tonight’s research supplement opens that remaining file.

Next time—another fragment of hidden truth to trace together.
I will return to the telling.

References

• ECMWF｜ECMWF’s AI forecasts become operational
  ECMWF’s official announcement of the operational Artificial Intelligence Forecasting System.
• ECMWF｜ECMWF’s ensemble AI forecasts become operational
  Official documentation of AIFS ENS and its 51 forecast scenarios.
• Google / Google DeepMind｜WeatherNext 2
  Google’s official description of its probabilistic AI weather forecasting system.
• U.S. Department of Justice｜Artificial Intelligence and Criminal Justice, Final Report
  Government review of predictive policing, risk forecasting, data quality, bias, privacy and feedback-loop concerns.
• The Jerusalem Post｜I have seen the future, and it’s on the Web
  Contemporary reporting on the Web Bot Project and its claimed forecasting methodology.

Posting Time

---

Related Reading

---

Popular Posts

---

Submit an Urban Legend

---

Share & Follow