Lesson video

Hello, my name is Mrs. James.

Welcome to Computing.

I'm so pleased that you decided to join me for the lesson today.

In today's lesson, you will be exploring environmental impacts of AI, describing the energy and resource demands of AI, and be able to identify energy efficient tools for digital tasks.

Welcome to today's lesson from the Unit: Using AI and Digital Tools Responsibly.

This lesson is called The Environmental Impacts of AI.

And by the end of today's lesson, you will be able to compare the energy and resource demands of AI systems with traditional digital tools.

Shall we make a start?

There are four keywords from today's lesson.

Server.

A server is a computer that stores, manages, and shares files, data, and resources in a network.

Data center.

A data center is a facility that houses servers, storage systems, and networking equipment to store, process and distribute data.

Training.

Training is the process of an AI system analyzing large data sets to identify patterns and refine the algorithms it uses to predict outputs.

Electronic waste or E-waste.

E-waste is discarded electrical or electronic equipment like phones, laptops, and batteries.

There are two sections within this lesson on the environmental impact of AI.

The first section is called Describe the Energy and Resource Demands of AI.

The second section is called Identify Energy Efficient Tools for Digital Tasks.

Should we begin?

Every digital search or AI prompt starts in a data center.

Data centers are huge warehouses that are filled with thousands of servers.

To work, these servers need electricity to power the servers, water to keep the servers from overheating.

You can see in this illustration of a data center, it has been built near pylons that are bringing in the electricity and near a water source which it's using to pump water into the data center.

Andeep says, "I didn't realize that computers needed water.

Do they get thirsty?

" Sofia replies, "Because the servers store and process huge amounts of data, their processes heat up and need to be kept cool.

Data centers use fresh water to stop the servers from overheating.

" First question.

Why did data centers use a lot of water?

Is it A, to cool the servers when they're processing large amounts of data?

Is it B, to generate the electricity to be used by the servers?

Or is it C, for employees to drink as the data centers are hot places to work?

What do you think?

If you said answer A, well done, you're correct.

Andeep is saying, "I use generative AI for everything, even to find out what day of the week it is.

" Sofia replies, "Did you know that a single generative AI prompt uses much more energy and water than a traditional web search?

" Comparing energy use: AI versus standard search.

A standard search, like a search engine, just finds existing links, whereas a generative AI search must calculate new patterns every time a prompt is entered.

Therefore, generative AI searches require much more energy.

On the screen now is a table that compares a traditional web search with a generative AI search and their water usage and electricity usage.

A traditional web search uses about 0.

3 of a milliliter of water, whereas an average generative AI search query uses between 10 millimeters and 25 milliliters of water.

This is about 30 to 80 times more water than a standard search.

When it comes to electricity usage, one standard traditional web search uses about 0.

3 of a watt hour of energy, whereas a single basic AI query uses roughly three watt hours.

This is 10 times more electricity than a standard search engine request.

However, it's not the prompting and searching that's the biggest problem.

Generative AI models must be trained first.

Training involves the AI system analyzing trillions of pieces of data to build a model that can then be used by users.

The diagram shows that raw data is fed in and then this trains the AI model.

This takes a long time.

After this, the AI model is tested and only then can it be used by users.

Training an AI model.

Thousands of specialized processes inside servers must run at their maximum capacity to handle these calculations.

These systems often run nonstop for weeks or months to complete the initial training of a model.

In 2022, training the GPT-3 language model in data centers used 700,000 liters of clean, fresh water and 1,287 megawatt hours of electricity.

That's enough electricity to power 120 homes for a year.

We've now reached the point in history where we've trained GPT-5 so we can only imagine how much water and electricity that used.

Next question.

True or false?

Training an AI system takes more energy than using the AI system by prompting it for answers.

What do you think true or false?

If you said true, well done, you're correct.

The training of an AI model uses up much more energy than the use of the model itself.

What other environmental impacts besides water and electricity usage do you think data centers cause?

Have a think.

If you said rare minerals are used to produce the computers and processes, well done.

Or you might have said large amounts of electronic waste or E-waste are produced.

Also, well done.

And if you said large areas of land need to be built on, then that's another factor, well done.

The challenge of E-waste.

Because AI technology moves so fast, the hardware often becomes out of date in just two to three years, compared to the five to seven years for normal computers.

This creates millions of tons of E-waste or electronic waste, which is hard to recycle because of the hazardous materials inside the servers.

Generative AI could add up to five million metric tons of extra electronic waste worldwide by 2030.

Rare minerals and hidden hardware.

AI technology requires specialized chips called GPUs, Graphics Processing Units.

Making these chips involves over 30 different rare materials mined from around the globe.

Producing a single two kilogram computer can require up to 800 kilograms of raw materials, fossil fuels, chemicals, and water during manufacturing.

Data center land use.

The increase in demand for AI features has led to more data centers needing to be built around the globe.

Each data center is a vast complex of server warehouses that need a supply of fresh water and electricity.

Building new data centers often takes water away from local communities and increases their electricity bills.

Often, it is farmland that is used as land for data centers, which negatively impacts the local economy.

Okay.

The first task for this lesson cycle.

You need to fill in the blanks.

I'll read out the paragraphs and I will say blank for the words that you need to replace.

AI systems are powered by thousands of servers located in large buildings called blank.

These facilities require massive amounts of blank to run the processes and large volumes of blank to keep the equipment from overheating.

While a traditional web search finds existing links, blank AI must calculate new patterns every time a prompt is entered.

This means a single AI query uses about blank times more electricity than a standard search.

The most resource heavy stage is blank, where the system analyzes trillions of pieces of data this process can take weeks and use enough energy to power over 100 homes for a year.

AI also contributes to blank because the specialized hardware often becomes outdated in just a few years.

So there are seven blanks that you need to fill in.

Pause the video and fill them in.

If you filled in data centers for the first blank, then electricity for the second, then water for the third, generative for the fourth, 10 for the fifth, training for the sixth, and E-waste or electronic waste for the last blank, you've done really well.

Even if you've got a few of those right.

Well done anyway.

There's a second task for this learning cycle.

A large technology company is looking to build a new data center.

Describe the factors that would influence where they would want to build this new facility and explain why.

Here's an example student answer.

"They would probably look for a place that is naturally cold because the thousands of servers inside get really hot when they're processing AI data.

And using the cold outside air to cool them down saves a lot of electricity and water.

They would also need to build it near a strong power grid or renewable energy sources because data centers use a massive amount of energy to stay on 24/7.

Finally, they would look for large areas of flat land as data centers take up a lot of space.

However, they would have to be careful not to take away too much water or land from local farmers, or it could cause problems for the people living nearby.

" If you covered those sort of points in your answer, well done.

Okay, we're now onto the second section of this lesson.

Identify energy efficient tools for digital tasks.

The hidden energy use of AI.

Digital tools rely on massive data centers that require electricity and water for cooling.

While a traditional search engine identifies existing information, generative AI must create a response using complex networks, which is significantly more energy intensive.

How do you choose when to use AI features and when not to?

Are you looking up existing information?

Well, you can probably just use a standard search engine.

However, are you trying to create something new like a story or a new image, then this is a perfect task for a generative AI tool.

Okay, a quick question.

Which of these tasks requires the most energy?

A, searching for a weather report on a standard search engine.

B, asking an AI tool to write a 500 word story.

Or C, finding a recipe to bake cookies.

Have a think.

Well done if you said B, asking an AI tool to write a 500 word story.

The other two, you could just use a standard search engine.

Andeep is saying, "I had no idea that the environmental impact of using AI was so much worse.

I thought that creating funny pitches using generative AI was okay to do as it was free.

" Sofia replies, "But now you can consider whether it really is worth using up extra water and electricity in a remote data center just to make funny pictures.

How do you choose when to use AI features and when not to?

" Making power smart choices.

Digital problem solving involves selecting the appropriate tool for a task.

If a task can be solved by a five-second traditional search using an AI model that consumes 10 times the power is an irresponsible digital choice.

Next question.

Scenario: You need to find the capital of France.

Which tool is the least environmentally responsible?

A, a generative AI query.

B, a standard search engine.

Or C, asking a friend sat next to you in the classroom.

Have a think.

If you said A, a generative AI query, well done.

That is the least environmentally responsible.

Okay, the final lesson task here.

We've given you a set of tasks and you need to identify which ones are high energy demand and which ones are low energy demand.

The list of tasks are, looking up the definition of algorithm, using a chatbot to explain a complex science theory, using a calculator to multiply two large numbers, prompting AI to generate a new image of a futuristic city, watching a saved video on your device, using AI to swap a face into an existing video clip, reading a blog post about digital inclusion, summarizing a 50 page document into bullet points, or searching for a local shop on a map.

So choose which of those tasks are high or low energy demand.

Let's have a look at the answers.

So you will see that anything that involves AI or a chatbot or summarizing is a high energy task, whereas traditional tasks like searching for a definition, or calculating numbers, or watching a video on your device is a low energy task.

Well done if you got that right.

Okay, we've reached the end of the lesson.

So in summary, data centers used huge amounts of electricity to power servers and lots of water to cool hardware.

They also require a number of rare minerals and generate electronic waste.

Training AI systems, the first step in making an AI model, requires more energy and water than the basic use of AI tools.

Using AI tools is more resource hungry than a standard search query.

To use digital tools responsibly, choose energy efficient methods like search engines for simple information look up.

Well done for completing this lesson on the Environmental Impacts of AI.