The Polar Bear Does Not Need Ice

TRANSCRIPT

The polar bear does not need ice.

This is important to establish at the outset.

He lives on it. Sleeps on it. Hunts from it. The polar bear has, for the entirety of his life and the lives of every polar bear before him, managed to acquire ice without the assistance of a third party.

Then one morning, his phone buzzes.

ICEAZON — Ice, delivered.

The proposition seems ridiculous.

He is, at that exact moment, standing atop several million tons of the stuff.

He dismisses the notification.

Later, he sees an ad.

The ice available naturally in his area, Iceazon informs him, may not be the right ice.

Iceazon has options.

Clear ice. Crushed ice. Artisanal ice. Locally sourced ice. Fast-freezing ice. Slow-melting ice. Premium ice harvested from glaciers with excellent customer reviews.

Even a dupe of the cylindrical, satisfyingly crunchy ice otherwise available at Sonic—conveniently located only about 200 road miles south of the Arctic Circle.

There’s also a personalized recommendation attached:

Based on your activity, we think you’ll love Arctic Select™.

The bear is skeptical.

He has always simply walked outside.

But then a second service appears.

Ubear Ice.

Ubear Ice doesn’t bother claiming the bear can’t find ice himself. That would be absurd.

Instead, it asks a more sophisticated question:

Why should he have to?

Ice acquisition takes time. Time has value. Walking is friction. Choice is good. Convenience is progress.

And for just $9.99 a month, Ubear Ice+ members receive unlimited ice delivery, reduced service fees, priority access during peak freezing events and exclusive discounts from participating glaciers.

Cancel anytime.

The bear begins with a free trial.

Just to see.

His first order arrives in eleven minutes.

The bag is pleasantly cold, and designed to mimic something that might have fallen off a 17th-century Dutch whaling ship.

He gives the courier five stars.

A week later, the app sends another notification:

Looks like you might be running low.

He isn’t.

But the system knows his previous order size, local weather patterns, recent browsing behavior and the average ice-consumption patterns of bears like him.

Perhaps it knows something he doesn’t.

He orders again.

Soon the bear develops preferences.

He discovers that he likes one supplier more than another. The ice from Glacier Direct seems somehow cleaner. Ubear Ice labels it a Top Pick.

He starts checking prices in the morning.

Iceazon has a sale.

Ubear Ice has faster delivery.

A new startup promises AI-personalized ice selection based on biometric data.

One morning, Glacier Direct appears with a warning:

ONLY 3 BAGS LEFT AT THIS PRICE.

Beneath it:

14 bears are viewing this glacier.

The bear orders quickly.

The bear also begins reading reviews.

And one day, he notices the ice beneath his feet and wonders whether it is any good.

There are no ratings.

No origin information.

No verified-purchase badge.

He cannot determine its melting profile.

There is no customer service number. No help button.

And slowly, almost imperceptibly, the absurd proposition becomes ordinary:

The polar bear no longer believes that ice is something he encounters.

Ice is something he accesses.

Through a platform.

Through a marketplace.

Through a recommendation system.

Through a company that knows which ice is best.

The system has accomplished something much more impressive than selling ice to a polar bear.

It has changed the bear’s relationship to ice.

We Are The Polar Bear

We are the polar bear.

Not literally, obviously. Most of us do not spend our mornings comparing same-day glacier delivery options.

But increasingly, modern life is composed of systems offering to insert themselves between us and things we once encountered more directly.

We used to hail a cab. Now an algorithm decides which driver appears.

We used to choose music. Now a system continuously predicts what we will want to hear next.

We used to visit websites. Search engines organized them. Social platforms ranked them. And now AI systems increasingly read them on our behalf, returning an answer without requiring us to see—or verify—the source at all.

We date through matching systems.

Apply for jobs through screening systems.

Navigate cities through routing systems.

Discover restaurants through ranking systems.

Encounter news through recommendation systems.

Buy products through marketplaces whose interfaces determine which products are visible in the first place.

Increasingly, even knowledge itself arrives through an intermediary.

Ask the machine.

It will read.

It will summarize.

It will interpret.

It will decide what matters.

The promise is extraordinary: a bottomless well of online information from which the exact thing we are seeking can somehow be surfaced from the depths on demand.

Convenience is not a scam simply because it is convenient.

Navigation software is better than getting lost. Search engines are more useful than personally indexing the internet. Anyone who remembers the Yellow Pages can attest to this. A system capable of reading a 200-page document and helping someone locate a relevant paragraph can save hours of tedious work.

The point is not that mediation is inherently bad.

The question is what happens when mediation becomes dependence, and dependence quietly becomes authority.

Because eventually, the intermediary does more than transport.

It begins to shape the encounter itself.

The polar bear begins by allowing an app to deliver ice.

He ends by asking the app what counts as good ice.

That distinction matters.

The Hostess

Perhaps the future will not belong entirely to the people who build the systems.

Perhaps it will belong to whoever stands between the system and everyone trying to use it.

Sex and the City understood this years ago.

In one early episode, Carrie declares that the most powerful woman in New York isn’t a politician, an editor or a celebrity. It’s the hostess at Balzac, which had, overnight, according to Carrie, “become the only restaurant that mattered.”

The hostess doesn’t own Balzac. She doesn’t cook the food. She didn’t make the restaurant fashionable.

She controls the door.

She knows which tables are available, which reservations matter, who can wait at the bar and who gets told the restaurant is fully booked. Everyone standing on the other side of the podium experiences Balzac through her.

That’s a peculiar kind of power.

You don’t have to own the thing if you control the conditions under which everyone else encounters it.

Carrie can look around the restaurant. She can see an empty table. She can insist that surely there must be somewhere to sit.

What she cannot see is the reservation book.

She cannot independently verify what the hostess knows, what rules are being applied, or whether “fully booked” means fully booked or fully booked for her.

She receives the conclusion.

Her practical options are to trust it, challenge it or leave.

Increasingly, technology occupies the hostess stand.

The platform does not need to own the restaurant, write the article, make the music, manufacture the product or drive the car in order to exercise enormous influence over how we encounter it.

It only needs to stand in between.

And increasingly, standing in between is enough.

The Intermediary

The intermediary itself is not new.

We have always relied on systems to help us navigate a world too complicated for any one person to encounter directly. Search engines have stood between us and the web for decades. Recommendation systems stand between us and music, movies, products and news. Navigation software stands between us and the street.

What is changing is how much we ask the intermediary to do.

At first, the system helped us find the thing.

Then it helped us choose the thing.

Now it can increasingly interpret the thing for us.

And the next generation of systems promises to act on the thing without us being there at all.

We once searched for a restaurant, looked through the options and called to make a reservation.

Then an app showed us which restaurants it thought we might like, ranked the options and handled the booking.

Now an AI assistant can be asked to find somewhere good for dinner, compare the choices, account for our preferences, select what it thinks is the best option and make the reservation.

Each step removes friction.

Each step can make life easier.

Each step also moves another decision from the person to the intermediary.

And that distinction matters, because the intermediary does not simply deliver information.

It filters.

Ranks.

Summarizes.

Recommends.

Excludes.

Predicts.

Interprets.

And, eventually, it acts.

The danger isn’t that we use systems to navigate reality. Humans have always done that.

It’s that we can gradually lose sight of where the system ends and reality begins.

The polar bear does not lose the ice.

He loses his direct relationship with the ice.

Carrie does not lose the restaurant.

She experiences it through the hostess.

The driver does not lose the road.

The navigation system increasingly determines how the road is experienced.

The reader does not necessarily lose the website.

But an AI system may increasingly determine which parts of that website ever reach the reader at all.

Every added layer can make life easier.

Every added layer can also move us slightly farther from the thing itself.

Until, eventually, people can become outsourced from their own relationship to reality.

Not because reality disappeared.

Because increasingly, someone—or something—else is describing it to us.

Legibility

For a long time, complicated technology could still feel strangely legible.

Consider the car.

A modern automobile was never truly simple. Thousands of parts worked together through combustion, hydraulics, electronics and mechanics to turn explosions inside a metal block into a trip to the grocery store.

Most people were never automotive engineers.

But there was once a cultural assumption that the machine was, at least in principle, understandable.

You could open the hood.

Someone in your family probably knew how to change the oil.

A mechanically inclined teenager could spend a Saturday replacing a belt.

There were manuals. Dense manuals.

Parts were visible.

Problems could be heard.

A strange knock from the engine meant something.

You might not know exactly what was wrong, but you understood that something under there was producing the knock.

Increasingly, the car looks—and feels—less like a machine and more like an iPhone with wheels.

The hood still opens, technically.

But more and more of what makes the vehicle behave as it does lives inside software, sensors, firmware, proprietary diagnostic systems and tightly integrated components increasingly inaccessible to the person sitting behind the wheel.

The transformation is often presented as progress.

And frequently it is.

Cars are safer, smarter and enormously more capable than they once were.

But something else changed alongside the capability.

Our relationship with the machine moved from:

I don’t understand everything happening here

toward:

I am not really expected to understand what is happening here.

The second condition is different.

Complexity requires expertise. Opacity requires trust.

And modern technology asks us for an extraordinary amount of it.

The problem becomes stranger when the machine stops merely operating and begins interpreting.

A transmission can shift gears without explaining itself.

But an artificial intelligence system can give you what sounds like an explanation for why it believes something. That is not necessarily the same thing as seeing what actually produced the answer.

It can assess a job application.

Summarize a medical record.

Detect suspicious behavior.

Recommend a financial product.

Determine which advertisement you are likely to respond to.

Tell a company which customers appear likely to leave.

Tell a teacher whether a student’s essay looks artificially generated.

Tell you what happened in the news.

Tell you what a document means.

Tell you what you mean.

And these systems increasingly encounter us through their conclusions rather than their machinery.

We see the answer. We rarely see everything that produced it.

What information was considered?

What was ignored?

What assumptions were embedded into the model?

How confident was it?

What did the system infer about us?

Which alternatives were discarded before we ever saw them?

Who decided what constituted a successful output?

When the system is wrong, what exactly are we appealing?

And to whom?

This creates an odd inversion.

Technology is becoming less legible at precisely the moment it is becoming more consequential.

We are building systems with astonishing power to classify, recommend, predict and interpret while simultaneously accepting that their internal operations may be inaccessible to nearly everyone affected by them.

And then we place an interface on top.

Clean.

Friendly.

Helpful.

A blinking cursor in an empty box.

What can I help you with?

Who Controls the Future?

And there’s another question hiding beneath all of this.

Who controls the future?

The engineers?

The companies?

The people writing the models and assembling the servers and designing the interfaces?

Certainly they control part of it.

But increasingly, even the people building these systems operate inside systems larger than themselves.

A company responds to markets.

Markets respond to investors.

Products respond to user behavior.

Algorithms optimize for metrics.

Workers respond to algorithms.

Competitors respond to each other.

Governments respond to companies.

Companies respond to governments.

Infrastructure responds to demand.

Demand responds to products designed to create more demand.

The systems connect.

And connect.

And connect.

Eventually we arrive at something more difficult to describe than a machine or a company.

A system of systems.

One capable of exerting enormous influence even when no individual sits at its center deciding precisely where it should go.

The engineers may build the road.

The executives may choose the vehicle.

The market may supply the fuel.

The algorithm may optimize the route.

And suddenly everyone is moving very quickly in a direction nobody remembers choosing.

But distributed agency cannot become an excuse to diffuse accountability.

The system did not build itself.

Someone still owns the servers.

Someone still chooses the metric.

Someone decides what gets optimized, what gets measured and what gets ignored.

Someone designs the interface.

A system can become larger than any one participant without suddenly becoming nobody’s responsibility.

This is where the question of control becomes slippery.

Power does not have to sit neatly in one place.

It can be distributed among the people who build the machinery, the companies that deploy it, the markets that reward particular outcomes and the interfaces that determine how the rest of us encounter what all of them have produced.

Sometimes controlling the future means building the machine.

Sometimes it means owning the infrastructure beneath it.

And sometimes it means standing at the door.

This Sounds Made Up

That sounds dramatic.

Maybe it is.

These are dramatic systems.

It is difficult to discuss artificial intelligence, ubiquitous automation, algorithmic decision-making and the enormous technological infrastructure growing around them without occasionally sounding like you have wandered into science fiction.

That is partly because a great deal of contemporary life already sounds made up.

Computers generate photorealistic video from sentences.

Algorithms predict what strangers will buy.

Cars drive themselves through cities.

Alone.

Machines converse in natural language.

And can sound like anyone.

Companies construct facilities requiring enough electricity to reshape regional power planning just to train and run those machines.

People form emotional attachments to software.

Software attends meetings on behalf of people.

Search engines increasingly answer questions rather than sending users to the humans who wrote the answers.

And all of this is happening while most of us are attempting to answer emails, pay bills, plan for the weather and decide what to make for dinner.

It is unprecedented territory.

It is also weird.

And hard to process.

At times, the existence of these systems — and the endless stream of outputs, promises, predictions, press releases and consequences they produce — can be difficult to believe.

Hence the name.

Artificial Incredulous.

Incredulous does not mean anti-technology.

It does not mean assuming every new machine is fraudulent, every engineer is dangerous or every algorithm is secretly conspiring against us.

It means retaining the ability to look at an extraordinary claim and say:

Hold on.

Show me.

How does this work?

Who benefits?

Who suffers?

Who decides?

What does the system know?

What does it only think it knows?

What did we gain?

What did we give up?

Who can inspect it?

Who can challenge it?

And what happens when it’s wrong?

Follow the Interface

Artificial Incredulous will follow the interface down.

From the answer to the system that produced it.

From the system to the company that deployed it.

From the company to the incentives shaping its decisions.

From the software to the servers, power, labor and infrastructure underneath it.

And from all of those systems back to the people expected to live with their consequences.

Sometimes that will mean reporting on artificial intelligence.

Sometimes it will mean following an algorithm into a workplace, a power grid, a search engine, a car, a government agency, a dating app or a grocery store.

Sometimes it may mean discovering that the technology works exactly as promised.

Sometimes the promise itself will be the interesting part.

The goal is not to approach technology assuming the answer.

It is to retain enough curiosity to keep asking the question.

Technology increasingly arrives accompanied by the promise that we need not concern ourselves with the machinery.

The system will handle it.

The recommendation will appear.

The route will be calculated.

The answer will be generated.

The ice will arrive in eleven minutes.

Maybe this really is the better way.

But when increasingly consequential systems ask us to surrender understanding in exchange for convenience, trusting them should not require surrendering our curiosity too.

Artificial Incredulous exists for the moment after someone says:

Don’t worry about how it works.

And before we answer:

Okay.