AI Understands Nothing… So Why Is It Already Making Decisions for You?

I want to start with a strange idea.

An artificial intelligence system can influence your life without understanding your life.

It does not need to know who you are. It does not need to care about your goals. It does not need to understand your fears, your plans, your relationships, your work, your future, or the weirdly specific video you watched at 1:37 a.m. and immediately pretended never happened.

It only needs something much simpler.

Signals.

Clicks.

Watch time.

Searches.

Patterns.

Predictions.

And once enough of those signals are collected, the system can start doing something extremely powerful:

it can arrange the world before you arrive.

Not the whole world.

But the version of the world that reaches you.

The posts you see.

The news that feels important.

The products that appear at the perfect moment.

The opportunities that pass through automatic filters.

The creators, businesses, and ideas that get shown to people.

Or do not.

That is the uncomfortable part.

The most influential AI in your life may not look like a robot.

It may not speak in a human voice.

It may not feel futuristic at all.

It may look like a feed.

A search result.

A recommendation.

A ranking.

A score.

A quiet filter you never notice.

The rule of this article

Before we go further, we need one rule.

This is not an article about AI becoming conscious.

That is a different conversation.

This is not about a machine waking up, developing feelings, and deciding to ruin your afternoon because your playlist had too much confidence.

The point is simpler.

And honestly, more useful.

AI does not need consciousness to matter.

It does not need emotions to affect you.

It does not need intentions to change what you see.

It only needs to be placed in the middle of important systems.

Between you and information.

Between you and opportunities.

Between you and other people.

Between you and your own decisions.

In this article, I am using the word AI broadly.

Not only chatbots.

Not only image generators.

Not only the tools that answer questions or write emails.

I am also talking about recommendation systems, predictive models, ranking algorithms, automated filters, and decision systems that sort, score, prioritize, and classify things every day.

Some of these systems are advanced.

Some are simpler.

Some are not what people imagine when they hear the word “AI.”

But they all share one important role:

they help decide what moves forward and what gets left behind.

And that is enough to matter.

Why the name confuses everything

The phrase “artificial intelligence” is useful.

It is also dangerous.

Because the moment we hear the word “intelligence,” we imagine something mind-like.

Something that understands.

Something that reasons.

Something that knows what it is doing.

But a lot of AI does not work like that.

It works more like prediction.

A system receives data.

It finds patterns.

It estimates what is likely.

Then it produces an output, a ranking, a recommendation, a classification, or a decision.

That can be useful.

Very useful.

But usefulness is not the same as understanding.

A navigation app can guide you through a city without knowing what a city means.

A recommendation system can predict which video you might watch next without knowing why you are tired.

A language model can produce a fluent explanation without experiencing the meaning behind the words.

That distinction matters because humans often see fluent output and imagine understanding behind it. We see a confident answer and assume knowledge. We see a clean ranking and assume fairness. We see a score and assume precision.

But a system can produce a result that looks intelligent without understanding the human situation behind the result.

Think of a very well-trained parrot.

It can say the right phrase at the right time.

It can sound surprisingly appropriate.

It can make you pause for a second and wonder whether something deeper is happening.

But the parrot is not understanding your life.

It has learned a pattern.

AI takes that idea and scales it dramatically.

Not a parrot with five phrases.

A pattern machine with enormous data, endless speed, and the confidence of someone who has never had to defend a group project in front of a tired professor.

The mistake: treating prediction like judgment

Prediction is not automatically a problem.

Predicting traffic can help you get home faster.

Predicting fraud can protect people.

Predicting what you may want to watch can help you find something interesting.

The problem begins when prediction starts acting like judgment.

A prediction says:

“Based on the pattern, this is likely.”

A judgment says:

“This is what should happen.”

Those are not the same thing.

But in real systems, they can become dangerously close.

A model predicts which candidate looks promising.

A platform predicts which post will keep people watching.

A feed predicts which topic will make you react.

A search or recommendation system predicts which result deserves attention.

And then the prediction shapes the environment.

The candidate is ranked lower.

The post is not shown.

The topic appears again.

The result moves to the top.

Nobody had to say:

“This person does not matter.”

The system only had to decide that someone else was more likely to produce the desired signal.

More clicks.

More watch time.

More conversion.

More retention.

More profit.

More engagement.

That is why the question is not only:

“Can AI think?”

A better question is:

“What happens when systems that do not understand human life begin arranging human options?”

Place 1 — The version of the world you see

Open your phone.

You are not seeing “the internet.”

You are seeing a version of it.

A personalized version.

A ranked version.

A version shaped by signals you gave, signals other people gave, and signals the platform decided were worth optimizing.

That means your feed is not a neutral window.

It is more like a menu where the order has already been chosen.

Some things are placed at eye level.

Some things are pushed down.

Some things never appear.

Some things are repeated until they feel more important than they are.

And the system does not need to hate you for that to happen.

It only needs a goal.

If the goal is attention, the system will learn what captures attention.

If outrage keeps people watching, outrage becomes useful.

If fear keeps people returning, fear becomes useful.

If conflict creates reactions, conflict becomes useful.

The algorithm does not need to believe anything. It does not need to be angry. It does not need to have an agenda in the human sense. It only needs to notice what works.

And over time, that can shape your perception.

Not by controlling your mind directly.

That would be too dramatic.

And honestly, less interesting.

It shapes perception by deciding what reaches your mind repeatedly.

That is quieter.

And much more believable.

Official explanations from platforms describe recommendation systems as ranking and personalizing content using signals such as watch history, search history, activity, interests, likes, comments, and other interactions. YouTube explains that recommendations use signals such as watch history, search history, subscriptions, and likes. TikTok describes its For You system as ranking videos from a combination of factors, including user interactions. Instagram has also described ranking as a process that uses signals to predict what people may care about most.

The comedy club and the collapse

Here is a simple way to feel the problem.

The same app can feel like a comedy club for one person and the end of the world for another.

One person opens their phone and sees jokes, music, food, pets, shopping, and light entertainment.

Another opens the same app and sees crisis, arguments, warnings, political rage, social comparison, and reasons to feel behind in life.

The app did not become two different apps.

The filter changed.

That matters because people do not only consume feeds.

They adapt to them.

A feed can train what feels normal.

It can train what feels urgent.

It can train what feels threatening.

It can train what feels desirable.

It can train what feels invisible.

Over time, the feed is not only showing you content.

It is shaping the emotional weather around your attention.

And when something surrounds your attention every day, it can start to feel like reality.

Not because it is the whole reality.

But because it is the reality most frequently placed in front of you.

Place 2 — The opportunities that reach you

Now move outside the feed.

Imagine applying for a job.

You prepare your CV.

You upload it.

You press send.

And before a person ever reads your name, an automatic system may help filter, sort, or rank your application.

The system may look for keywords, experience, formatting, previous roles, education, patterns from past applicants, or signals that suggest whether you match the position.

Sometimes this can help organizations manage huge numbers of applications.

But it also introduces a strange new layer.

You may be rejected before you feel rejected.

Not by a person looking you in the eyes.

Not by a recruiter carefully reading your story.

But by a system deciding whether your profile resembles what it has learned to treat as promising.

That does not mean every automatic filter is evil.

It does mean the filter changes the nature of opportunity.

Because opportunity is not only about whether you are capable.

It is also about whether the system recognizes your capability.

And systems do not see people.

They see data.

They see structure.

They see words.

They see categories.

They see missing information.

They see probabilities.

That difference can matter a lot.

The EEOC has published guidance and resources about the use of software, algorithms, and AI to assess job applicants and employees. The U.S. Department of Justice has also noted that employers may use computer software to score applicants’ resumes, and that many of these tools use algorithms or AI.

The statistical version of you

This idea goes beyond jobs.

Similar patterns can appear in credit, insurance, education, advertising, fraud detection, risk scoring, and other automated systems.

A model may not know you.

But it may estimate you.

Probability of paying.

Probability of leaving.

Probability of clicking.

Probability of buying.

Probability of being risky.

Probability of being profitable.

This is efficient.

It is also deeply strange.

A person becomes a profile.

A profile becomes a score.

A score becomes a decision.

And the decision may shape what that person can access next.

The Consumer Financial Protection Bureau has warned that some creditors use complex algorithms, sometimes described as “black-box” models, in credit decisions. It has also stated that legal requirements for adverse action notices still apply when lenders use artificial intelligence or complex models.

That is the quiet power of prediction.

It does not need to declare who you are.

It only needs to decide which version of you an institution gets to act on.

Maybe that version is useful.

Maybe it is incomplete.

Maybe it is unfair.

Maybe it is technically impressive and socially clumsy at the same time.

Very human, in the worst possible way.

Place 3 — Whether anyone gets to see you

There is another place where this becomes very personal.

Visibility.

A musician uploads a song.

A local business posts a new menu.

A creator publishes a video.

A researcher shares an idea.

A student builds a project.

An artist posts work that took weeks.

All of those things can exist online.

But online, existence is not the same as discovery.

Something can be published and still be functionally invisible.

Between the thing you make and the people who might care, there is usually a ranking system.

A feed.

A search engine.

A recommendation layer.

A platform deciding who sees what, when, and for how long.

That system may test your post with a small audience, measure reactions, compare performance, and decide whether to continue showing it.

And sometimes, that helps good things spread.

But not always.

The best thing does not automatically win.

The clearest thing may win.

The fastest thing may win.

The most emotionally immediate thing may win.

The thing that creates a reaction before the viewer has time to think may win.

This changes how people create.

A musician does not only ask:

“Is this song good?”

They also ask:

“Does the first second hold attention?”

A business does not only ask:

“Is the product good?”

They also ask:

“Will the image stop the scroll?”

A creator does not only ask:

“Is this idea worth sharing?”

They also ask:

“Will the platform give it air?”

That is a major shift.

Visibility used to be mostly about access to distribution.

Now it is also about surviving automated selection.

The dangerous part is not hatred

The easiest version of this story would be:

“The machines are evil.”

But that is too simple.

Most of these systems are not evil.

They are indifferent.

That is the scarier part.

A malicious system has a goal you can identify.

An indifferent system follows signals.

If the signal is watch time, it learns watch time.

If the signal is clicks, it learns clicks.

If the signal is retention, it learns retention.

If the signal is conversion, it learns conversion.

The system does not have to ask whether the result is healthy, fair, wise, or true. Unless those values are built into the design, measured carefully, and protected against trade-offs, the system may optimize whatever is easiest to measure.

And what is easiest to measure is not always what matters most.

Truth is hard to measure.

Wisdom is hard to measure.

Human dignity is hard to measure.

Context is hard to measure.

Long-term consequences are hard to measure.

Clicks are easy.

Watch time is easy.

Purchases are easy.

Engagement is easy.

That imbalance is one of the core problems of automated systems.

We often measure what is visible.

Then we optimize it.

Then we act surprised when the invisible things get damaged.

This is why AI risk frameworks matter. The NIST AI Risk Management Framework, for example, is built around identifying and managing risks in AI systems rather than assuming that usefulness automatically means safety.

Useful, fast, cheap, and wrong often enough

You do not need a conscious AI to change society.

You only need systems that are useful enough to adopt, fast enough to scale, cheap enough to deploy widely, and wrong often enough to matter.

That combination is powerful.

A tool can be useful and still make mistakes.

A ranking can be efficient and still hide important things.

A prediction can be statistically reasonable and still harm an individual.

A filter can save time and still erase context.

That is why the conversation should not be only hype or panic.

Hype says:

“AI will solve everything.”

Panic says:

“AI will destroy everything.”

Both are too easy.

The better question is more specific:

“Where are we placing AI, what signals are we asking it to optimize, and who gets affected when the system is wrong?”

That question is less dramatic.

But much more useful.

Reality is not gone. It is filtered.

This does not mean you have no agency.

It does not mean every decision is fake.

It does not mean you are a puppet controlled by a recommendation system wearing a tiny villain cape.

Reality still exists.

Your choices still matter.

Your effort still matters.

Your judgment still matters.

But your choices increasingly happen inside environments that have already been arranged.

That is the key.

You still choose what to click.

But the menu was ranked.

You still decide what to believe.

But the information was filtered.

You still apply for the job.

But the application may be screened.

You still publish the work.

But the platform decides how much air it gets.

This is not total control.

It is architecture.

And architecture matters.

A staircase does not force you to climb.

But it shapes where movement is easy.

A door does not force you to enter.

But it defines the path.

A feed does not force you to think.

But it shapes what appears worth thinking about.

The Digital Services Act in the European Union reflects this broader concern by requiring online platforms that use recommender systems to explain the main parameters used by those systems and provide options for users to modify or influence them.

How to stop being a passive user

You cannot remove every algorithm from your life.

That is not realistic.

And honestly, it is not necessary.

The better goal is not escape.

The better goal is awareness.

You can learn to notice when a system is arranging your options.

You can learn to ask better questions.

You can learn to treat automated output as useful without treating it as final.

Here are five places to start.

1. Do not let your feed become your only map

A feed is not the world.

It is a selected slice of the world.

So if all your information comes from one platform, your reality is being filtered through one set of incentives.

Read longer explanations.

Compare sources.

Follow people who do not all reward the same emotional reactions.

Look for context before you let a feed define what matters.

2. Notice what emotion the system keeps feeding

After scrolling for a while, ask:

what emotion did this app amplify?

Outrage?

Anxiety?

Desire?

Envy?

Urgency?

Entertainment?

None of those emotions are automatically bad.

But if the same emotional pattern appears every time, that is not random.

That is feedback.

3. Write for humans and filters

If a machine may evaluate your CV, portfolio, profile, or post, make the important signals clear.

Use simple structure.

Use relevant keywords honestly.

Show measurable outcomes when possible.

Do not hide your value behind vague language.

This is not selling your soul to the system.

It is making sure the system can read what a human should notice.

4. Treat AI output as a starting point

When AI gives you an answer, especially about something serious, do not confuse fluency with truth.

Health.

Money.

Law.

Work.

Education.

Public information.

In those areas, confidence is not evidence.

A clean answer is not the same as a correct answer.

Use AI to explore, organize, draft, compare, and challenge ideas.

Then verify what matters.

5. Keep authority in the right place

A tool helps you think.

An authority thinks for you.

That distinction is everything.

If AI helps you see more options, good.

If AI helps you ask better questions, good.

If AI helps you understand a complicated topic faster, good.

But if you stop checking, stop thinking, and stop deciding, the problem is no longer only the tool.

The problem is that you handed over the steering wheel.

References and further reading

• YouTube — How YouTube recommendations work
• YouTube — Algorithm-Based Recommendations on YouTube
• TikTok — How TikTok recommends videos #ForYou
• TikTok — Learn why a video is recommended For You
• Instagram — Shedding More Light on How Instagram Works
• Meta Transparency Center — Instagram Feed AI System
• EEOC — Artificial Intelligence and the ADA
• U.S. Department of Justice — DOJ and EEOC warning on algorithmic tools in employment
• CFPB — Adverse action notification requirements and complex algorithms
• CFPB — Credit denials by lenders using artificial intelligence
• NIST — AI Risk Management Framework
• EUR-Lex — Regulation (EU) 2022/2065, Digital Services Act

So, who chose the order?

Tonight, when you open your phone for five minutes, try one small experiment.

Do not only look at the posts.

Look at the order.

What appeared first?

What appeared again?

What felt urgent?

What disappeared?

What emotion did the system seem to reward?

That order is not random.

It is a design.

A prediction.

A bet.

A guess about what will keep you engaged.

And once you notice that, the screen changes.

It stops feeling like a simple window.

It starts feeling like an arranged environment.

That does not mean you should be terrified.

It means you should pay attention.

Because AI does not need to be alive to matter.

It does not need to understand you to influence you.

It does not need to be human to stand in the middle.

Between you and information.

Between you and opportunities.

Between you and visibility.

Between you and your own decisions.

And the more systems stand in the middle, the more important one question becomes:

“Did I choose this… or was this placed in front of me?”

If you want to keep exploring, you can keep reading on AtomicCurious.

And if you want the next explorations before everyone else, join the newsletter.

⚛️ AtomicCurious — Exploring science, technology & smart curiosities.