See Inside AI: Power, Risk, and Climate Costs

1. Why AI Feels Useful and Worrying

Think about a smart shopkeeper in Jauharabad who has seen thousands of buying patterns. If you ask them what customers might want next, they can often make a surprisingly good guess — not because they truly understand your life, but because they’ve learned patterns from a lot of past examples. AI works in a similar way.

That is the first key idea: AI is good at pattern recognition and prediction. It can draft, summarize, classify, recommend, translate, and sort faster than a person can. That is why it feels useful. It turns messy information into something more manageable.

But the same thing that makes it useful also makes it worrying.

AI does not “know” in the human sense. It can sound confident and still be wrong. It can reproduce bias hidden in the data it learned from. It can be used to save time — or to automate mistakes at scale. So the feeling of AI is often a mix of two truths at once:

• It helps because it is fast, flexible, and pattern-driven
• It worries us because it can be convincing without being reliable

By the end of this course, M. Shahzad, you will be able to explain why AI can seem intelligent while still being limited, why people trust it so quickly, and why that trust needs to be handled carefully. You’ll also be able to connect those benefits and risks to a bigger question: what does it cost the world to make and run these systems?

So keep this anchor in mind:

AI is not magic. It is a very powerful prediction system.

Once that idea becomes clear, the rest starts to make sense.

2. What AI Actually Is Beneath the Hype

Remember earlier when we said AI is a powerful prediction system? Let’s open that up and look inside it.

At the most basic level, AI is not a robot brain, not a thinking creature, and not a pile of magic. It is a system that turns data into patterns, and patterns into outputs. That’s the core.

Think of it like a person who has spent years sorting huge stacks of clothes in a market store. They are not memorizing every shirt. They are noticing repeating signals: fabric, color, size, texture, brand, season. Over time, they become very good at guessing where a new item belongs. AI works in a similar way: it learns what usually comes next based on what it has seen before.

That gives us the first irreducible truth:

AI does not “understand” in the human sense. It maps patterns.

This matters because a lot of confusion comes from expecting human-style thinking. Humans can explain why something matters. AI mostly estimates what is likely.

1) AI learns from examples, not from common sense

A child can see one cup fall and understand that gravity exists. AI usually needs many examples before it notices a useful pattern. It is closer to a student who learns by reviewing thousands of past exam papers than a student who learns by curiosity.

That is why AI can be excellent at tasks with lots of examples:

• recognizing faces,
• translating language,
• predicting the next word,
• detecting spam,
• recommending videos or products.

But it also explains its weakness. If the examples are incomplete, biased, or misleading, the system learns the wrong lesson. Remember earlier when we said AI can be convincing without being reliable? This is one reason why.

AI is only as good as the patterns it has absorbed.

2) AI does not store the world; it compresses it

Here is another key idea.

AI does not carry around a full copy of reality. It compresses information into a smaller internal form. It keeps the parts that seem useful for prediction and drops the rest.

Imagine packing for a trip with only one small bag. You cannot take your entire room with you, so you choose what matters most: clothes, charger, documents. AI does something similar with information. It keeps the features that help it answer or predict, but it loses detail along the way.

This is why AI can be both powerful and limited at the same time.

• Powerful, because it can reduce huge messes into useful structure.
• Limited, because it never contains the whole situation.

This is the same idea showing up again: AI is not reality itself; it is a compressed map of reality.

And like any map, it can be helpful without being complete.

3) AI outputs are guesses, not truths

A very important distinction: AI produces likely answers, not guaranteed ones.

Imagine asking a friend to guess how many people are in a wedding hall from outside the gate. They may make a good estimate, but they are still guessing from clues. AI works this way all the time. It calculates what answer best fits the patterns it has learned.

That’s why AI can write a fluent paragraph that is wrong, or give a confident answer with missing facts. It is not checking truth like a human investigator. It is choosing the most probable response.

So when AI sounds certain, that certainty is partly a performance of language. It is not the same as knowledge.

This table is the heart of the confusion. AI can imitate the surface of understanding without possessing understanding in the human sense.

4) The system matters more than the label

People say “AI” as if it is one thing. But it is really a family of systems built for different kinds of pattern matching.

Some are good at classifying. Some are good at generating. Some are good at recommending. Some are good at forecasting. The label is broad, but the behavior is specific.

That is why asking “Is AI good or bad?” is too vague. A better question is:

• What was it trained on?
• What task is it doing?
• What errors are acceptable?
• Who is relying on the output?

A calculator and a weather forecast app are both “tools,” but you would never trust them in the same way. AI is similar. The system’s design determines what it can safely do.

5) AI is shaped by tradeoffs

Now we reach the idea that connects directly to the course title.

Every gain in AI usually comes with a tradeoff.

If you want a system that is more flexible, it often needs more data and more computation. If you want it to be more useful across many tasks, it may become harder to explain. If you want it to work faster for millions of people, it may require more energy and infrastructure.

This is like choosing a motorcycle, a car, or a bus. A motorcycle is quick and efficient for one person. A bus is better for many people at once. A car gives comfort and flexibility. Each choice solves one problem while creating another.

AI works the same way:

• More capability often means more cost.
• More generality often means less transparency.
• More speed and scale often means more hidden impact.

That last tradeoff will matter later when we talk about environmental cost.

6) AI becomes powerful by being trained, not born finished

A final first principle: AI is not built with all knowledge inside it. It is trained.

Training means the system is adjusted repeatedly until its outputs improve. You can think of it like shaping a clay pot. At first it is rough and unstable. Repeated pressure and correction give it form.

This helps explain why AI systems can improve so much from large amounts of data and computation. But it also explains why they inherit the limits of their training process. They are not born wise. They are tuned.

And that tuning takes work — human work, data work, and energy work.

So if we strip away the hype, the simplest true statement is this:

AI is a trained pattern system that compresses examples into probable outputs.

That sentence is plain, but it carries the whole foundation.

The idea to carry forward

If you remember only one thing from this section, remember this:

• AI is not a mind.
• It is not truth itself.
• It is not magic.
• It is a pattern-based system that can be extremely useful, but only within the limits of what it has learned.

That is why AI feels impressive, and why it can also fail in ways that human beings do not expect.

In the next section, we’ll take this foundation and ask a harder question: if AI is just a prediction system, why does it still create such real benefits, risks, and environmental costs?

3. How Benefits, Harms, and Energy Costs Connect

Remember earlier when we said AI is a trained pattern system that compresses examples into probable outputs? That sentence is doing a lot of work. It explains not just what AI is, but why its effects in the real world are so mixed.

AI’s benefits, harms, and energy costs are not separate topics. They are all different expressions of the same underlying design choice:

AI gains power by using more data, more computation, and more scale.

That one engine creates the whole picture. When the engine gets stronger, the useful parts get better — but so do the risks, and so does the energy use.

1) Why the same system can help and harm

Think back to Section 1. We said AI feels useful because it can quickly turn messy information into something manageable. That is real. AI can help people write, search, summarize, translate, classify, and draft much faster than before.

But the reason it can do that is the same reason it can go wrong:

• it learns patterns from data,
• it generalizes from those patterns,
• and it produces likely outputs even when the situation is messy.

That means AI is not just a tool that “knows” what to do. It is more like a very fast guesser. When the environment is stable, that guess can be excellent. When the environment is ambiguous, biased, or unusual, the guess can become misleading.

So the benefit and the harm are twins.

The benefit side

AI is strong where pattern repetition is high.

• If many emails are spam, it can learn spam signals.
• If many documents follow similar structures, it can summarize them.
• If many images show familiar objects, it can identify them.

In these settings, AI saves time because it turns repeated human judgment into automation.

The harm side

The same pattern-based logic can fail when the world is not as neat as the training data.

• A face recognition system can misidentify people.
• A hiring model can inherit old bias.
• A chatbot can sound sure while being wrong.

This is not an accidental side effect. It is what happens when a system trained on the past is used to make decisions about the present.

The more AI relies on patterns, the more it struggles with novelty, unfairness, and hidden context.

That is the tradeoff at the center of the whole topic.

2) Why scale makes AI better and riskier at the same time

Now let’s connect this to scale.

Remember in Section 2 we said AI is not born finished. It is trained. Training usually means more data, more computing, and more iteration. That gives AI more capability. But capability does not come alone.

Scale changes the system in two directions at once:

1. It improves performance on many tasks.
2. It amplifies impact when the system is wrong.

That second point is easy to miss.

A small mistake made by one person may affect a few people. A model mistake built into a product used by millions can spread much farther, much faster, and much more quietly.

This is why AI risk is not only about “bad answers.” It is about bad answers at scale.

A useful analogy is a loudspeaker. A whisper is one thing. But if you put that whisper through a speaker system and project it to a stadium, the same message reaches far more people. AI acts like that amplifier for patterns. If the pattern is good, the benefit spreads. If the pattern is flawed, the flaw spreads too.

That is why organizations care so much about testing, oversight, and limits. The question is not whether AI can sometimes be right. The question is whether it is right reliably enough for the scale at which it is used.

3) Why energy use is built into the design

Now we reach the environmental side.

It is tempting to imagine that AI’s climate impact is separate from its usefulness — as if the system is helpful on one side and energy-hungry on the other for unrelated reasons. But they are connected.

AI needs computation. Computation needs hardware. Hardware needs electricity. Electricity generation often has environmental cost.

So the chain is simple:

More training and more use → more computing → more electricity → more environmental footprint

The key point is that AI does not just “exist” as a thought in a vacuum. It runs on real machines, in real data centers, using real power, with real cooling needs.

This matters because large AI systems often do two things at once:

• They are expensive to build.
• They are cheap to use once built.

That creates a subtle but important pattern. A single interaction may seem small, but the total impact comes from millions or billions of interactions added together. The climate cost is not visible in one prompt. It appears in the full system of training, deployment, and repeated usage.

Here is the deeper connection:

AI becomes more useful by being scaled up, and scaling up is exactly what increases energy demand.

So the environmental question is not separate from the product question. It is part of the same design choice.

4) The “cheap at the surface, costly underneath” pattern

This is one of the most important ideas in the whole course.

AI often feels lightweight to the user. You type a sentence, get an answer, and the exchange seems almost frictionless. That experience can hide a much larger machine underneath.

It is similar to using a tap. A glass of water feels instant and simple. But behind that instant result is a whole invisible system: pipes, treatment, pumps, maintenance, and infrastructure. If the tap works well, you barely notice the complexity. AI works the same way from the user’s perspective.

That hidden complexity is what makes AI easy to underestimate.

So when people ask, “Why does one AI response matter environmentally?” the better answer is:

• one response is small,
• the system behind it is large,
• and the total footprint depends on the whole machine, not the single moment.

This is the same idea showing up again from Section 2: AI outputs are not the whole story; the training and infrastructure behind them matter too.

5) Real tradeoffs in the wild

To see how these forces connect, imagine an organization deciding whether to use AI.

They want speed, lower labor costs, and better handling of repetitive tasks. That is the upside. But once the system is deployed, they must also think about error rates, fairness, oversight, and power use.

Now notice something important: the same feature can create both value and risk.

This table shows why the topic is not simply “AI is good” or “AI is bad.” It is a system of tradeoffs.

A Pakistan-relevant example

Imagine a bank in Pakistan using AI to help sort customer support queries in Urdu and English. The benefit is clear: many routine questions can be handled quickly, customers wait less, and staff can focus on complex cases.

But now look at the other side. If the system misunderstands local phrasing, regional language patterns, or mixed-language messages, it may give poor guidance to customers. And if the bank expands the system widely, that error repeats across thousands of interactions. If the bank also relies on a larger hosted model, there is added electricity and infrastructure cost hidden in the background.

So the same deployment can be:

• efficient,
• fragile,
• and energy-consuming

all at once.

That is what makes AI governance hard. You are not choosing between a perfect tool and a useless one. You are choosing how much benefit is worth how much risk and cost.

6) Why people disagree about AI

Once you see the system clearly, the public debate makes more sense.

People who emphasize usefulness are usually looking at speed, convenience, and scale. They are seeing the upside of pattern automation.

People who emphasize risk are usually looking at bias, hallucinations, job disruption, and power concentration. They are seeing what happens when a high-scale guesser is trusted too much.

People who emphasize climate impact are looking at the hidden machine underneath the convenience — the energy, cooling, and infrastructure that make the experience possible.

All three views are partially right.

That is the real insight:

AI is not one effect. It is a bundle of effects produced by one system.

When you understand the system, the debate becomes less confusing. The arguments are not random. They are each pointing at a different part of the same machine.

7) The core connection to carry forward

Let’s bring everything together.

AI helps because it is good at detecting and reproducing patterns. AI harms because pattern systems can reproduce bias, uncertainty, and mistakes at scale. AI costs energy because making those pattern systems strong requires computation, and computation requires infrastructure.

So the relationship is not:

• usefulness here,
• harm over there,
• climate cost somewhere else.

It is one connected chain:

more capability → more scale → more reliance → more impact

and that impact includes both social and environmental consequences.

If you remember just one sentence from this section, make it this:

AI’s benefits, harms, and energy costs are linked because they all come from the same thing: scaling pattern recognition into real-world systems.

In the final section, we’ll turn that understanding into judgment: how should you think about AI responsibly, and what questions should you ask before trusting or expanding it?

4. What This Means for Choices and Responsibility

You’ve now built the full picture.

You can see why AI feels impressive: it is a pattern system that can turn messy information into useful outputs. You can also see why it can be dangerous: it can sound confident without really understanding, and when it is scaled up, small errors can become large ones. And now you can see why the climate question belongs in the same conversation: the more capable and widely used the system becomes, the more computation, infrastructure, and electricity it tends to require.

That is a real shift in how you think. You are no longer looking at AI as a mysterious black box. You are looking at it as a set of tradeoffs.

That is what mature understanding looks like. Not excitement alone. Not fear alone. The ability to hold both at once.

You now think like someone who can ask better questions:

• What kind of pattern is this system learning?
• Where might it be wrong?
• Who pays the cost when it fails?
• What happens when this is used at scale?
• What hidden energy and infrastructure make this possible?

Those questions are the difference between being impressed by AI and being able to judge it.

And that matters in the real world. It means you can explain why a tool may be helpful in one setting and risky in another. It means you can spot when convenience is hiding cost. It means you can talk about AI with clarity instead of hype.

M. Shahzad, that is the real win here: you now understand AI not as a miracle, but as a system with power, limits, and consequences.

So carry this forward:

When you understand the tradeoff, you are no longer just using technology — you are thinking responsibly about it.

That is a stronger position than knowing the buzzwords. It is the position of someone who can make sense of the machine, not just consume it.

And once you see AI this way, you start seeing many technologies the same way too: as systems that give something, take something, and ask for judgment in return.

That’s the kind of understanding that stays with you.