Lesson video

Hi, my name's Mr. Hall, and welcome to this Oak National Academy lesson, which is called Machine Learning and Artificial Intelligence, and is from the unit Computer Systems and Data Science.

I'm delighted you've joined me for this lesson today, so let's get started.

The outcome of this lesson is that, "I can describe artificial intelligence and machine learning, and explain how machine learning differs from traditional programming.

We have three keywords or phrases in this lesson.

So the first key phrase is artificial intelligence, or AI, and artificial intelligence is computer systems that adjust their outputs based on data or follow rules to perform complex tasks.

The second key phrase is machine learning, or ML, and machine learning is a system that uses data to improve its outputs without being programmed.

And then finally, our last keyword is programming, and programming is the process of writing rule-based computer programmes.

So there's are three keywords or phrases, artificial intelligence or AI, machine learning, or ML, and programming.

And we have three learning cycles in this lesson.

So first of all, we're going to describe artificial intelligence, and then later on in the lesson, we'll describe machine learning and compare machine learning and programming.

So let's get started with that first learning cycle, to describe artificial intelligence.

So what links the images below? We've got a smartwatch, online shopping site, and a smartphone.

Well, what links them is that they all use artificial intelligence, so let's find out a bit more about what that means.

So Andeep starts off with a question.

"Artificial intelligence, is that the same as AI?" And Sam says, "Yes, artificial intelligence is often shortened to AI.

AI systems are used all around us." And Sam's right, AI systems are used in many different tools and technology that we use daily, like voice assistants, search engines, and video games.

AI technology is about building computer systems that can perform complex tasks and make decisions by adjusting their output based on data, sometimes called learning, or by following rules.

There are many different types of AI, and they are designed to be useful in different ways.

So in this lesson, we'll look at some of the different types of AI that you may come across.

First, time for a question.

What is artificial intelligence? Is it A, computer systems that can learn from data or follow rules to perform complex tasks? Is it B, a machine that learns through experience, like a human would? Or is it C, a robot that can work without software? Well done, the answer is A, artificial intelligence is a computer system that can learn from data or follow rules to perform complex tasks.

So a question this time from Sam.

"What industries are AI systems used in?" Well, AI systems are used in many industries.

Some include healthcare, entertainment, and transport.

In healthcare, AI systems help doctors spot diseases in scanned images.

AI tools can also help doctors select improved, personalised treatments and medicines to help patients recover faster.

In entertainment, AI systems can recommend music, movies, and TV shows that users may enjoy.

It does this by finding patterns in your previous choices and comparing them with what other people enjoyed.

And in transport, AI systems help power self-driving cars.

AI systems use cameras and sensors to see what's around the car, like traffic lights, people, and other cars, which allows it to stop, go, or turn safely.

1996 was an important year for AI development.

Do you know why? Well, it was the year that an AI system called Deep Blue famously beat one of the world's best chess players.

This was significant because programmes had been trying to develop a chess system that would beat a human being for years and years, and by actually beating one of the world's best chess players, they achieved a major breakthrough.

So Andeep says, "Wow, AI systems must be really smart," and he's kind of right.

Although AI systems may seem smart, it's important to remember that all AI systems are developed by humans who have control over their design.

AI systems can improve their performance by learning from data, and Sam says, "Well, that's why they can appear to be really smart." And Andeep says, "That must be why AI systems seems smart and suggest programmes that I might like to watch." Humans learn from experiences and emotions.

AI systems are designed to also learn from experience, but do this by using data and patterns, not by feelings.

There's no emotion involved in artificial intelligence systems. AI systems are built through algorithms and trained on data provided by humans.

AI systems follow programmed rules or improve performance from data, but they do not think or have emotions like humans.

And that's a really important point to remember.

When data is used to improve the performance of an AI system, it's called machine learning.

Machine learning allows the AI system to identify patterns in the data and change its output to improve its own performance or achieve specific goals, so it's all data-based.

Generally, as the AI system gathers and processes more data, it improves in its ability to perform the task it was designed to do.

So more data equals a better output.

Time for a question.

True or false? All AI systems are designed and developed by humans.

Is that true or false? Well done.

It's true.

And another question, which of the following is an example of an AI system used in real life? Is it A, a calculator solving maths problems, B, a remote control car, or C, a voice-activated assistant? Well done.

It's C, a voice-activated assistant uses AI.

Another question, AI, can, A sleep and dream, B, feel happy or sad, or C, recognise data patterns well done.

It's C.

AI is not human, it can't do the first two, which are human things.

It can only recognise data patterns, so the correct answer is C.

Now, time for your task.

Fill in the blanks in this sentence to describe artificial intelligence.

So artificial blank is about building computer systems that can perform blank tasks and make decisions by learning from blank or following blank.

This includes things like analysing medical blank to help doctors or recommending music and videos that users might enjoy.

AI can appear to be blank, but it's not human and doesn't think or feel things.

And the words you can use are complex, images, data, rules, intelligence, or smart.

Okay, let's fill in those gaps.

So this is the complete answer.

Artificial intelligence is about building computer systems that can perform complex tasks and make decisions by learning from data or following rules.

This includes things like analysing medical images to help doctors or recommending music and videos that users might enjoy.

AI can appear to be smart, but it's not human and doesn't think or feel things.

We can now move on to the second learning cycle, which is to describe machine learning.

So machine learning is a part of AI that's about getting computers to be better able to complete a certain task over time.

With machine learning, the computer adjusts the systems' behaviour or output based on data rather than being directly programmed with rules for each task.

Machine learning is often shortened to ML, and ML is probably a phrase that you've heard quite a lot in recent times.

Sam has a question about machine learning.

"Are humans still involved in developing machine learning?" They are, humans are still involved in machine learning at nearly every stage of the process.

ML is a tool that depends on human knowledge, judgement , and control to be useful and trustworthy.

A machine learning model is what you get after a computer learns from data.

There are different types of ML models, but each share a number of common features, so let's look at those features now.

They all learn from training data, they all find patterns or relationships, and they all make predictions on new data.

So now, we'll have a look at that in a little bit more detail.

Machine learning is a data-driven approach to AI.

This is because ML systems need lots of examples and data to work well.

There are three main types of machine learning.

There's supervised learning, unsupervised learning, and reinforcement learning.

Supervised learning uses known answers to train the machine learning model.

With supervised learning, you give the machine both the input data and the correct output.

Over time, it finds patterns that connect the inputs to the right outputs.

Classification is a way of using supervised learning to categorise data.

A classification model assigns data to a class by applying labels.

So here are three potential labels you could apply to some data, so we have apple, banana, and orange.

Those are our three labels.

Classification models must be trained with example data that already has labels assigned by a human.

So a human being has already decided that the first image is an apple, the second image is a banana, and the third image is an orange, and those images have been assigned to the correct labels by a human.

The more training data you use, the more accurate the model will be.

So if you have a look at the images under apples, it's quite important, so an apple can be read or green.

So we've got examples of some red in different shades and green apple under apples.

We've got different examples of bananas, and again, we have different examples of oranges.

So this will be better training data than just one image for each label.

Once the model is trained, new data can be fed into the model and it will produce a prediction of which class it belongs to.

So here's our new input data, so here's an apple.

So we've got a whole apple and an apple sliced in half.

It will go through the machine learning classification model, which will analyse it against the images it's already been trained on and will predict with a confidence of 92% that this is an apple.

In unsupervised learning, no answers are given to the machine.

Instead, the machine is designed to group and organise the data itself, and then find useful hidden patterns.

Unsupervised learning is used by online stores and supermarkets to find products that are often bought together.

It could, for example, detect patterns, like people who buy bread often also by butter.

So next time you're in a supermarket, have a look at how products are grouped.

They are not grouped by accident.

They're grouped using lots of data, which can predict if you buy one product, you may also buy another.

In the last model, reinforcement learning, the machine adjust its behaviour by trying to do a task and getting feedback.

If it does the task well, it will receive a reward, but if it fails or doesn't do the task well, it will receive a penalty.

Over time, the trial and error approach means that the machine improves its output to better complete the task.

So an automatic robot vacuum cleaner is a really good example of this.

Imagine if you have a robot vacuum cleaner that improves at the task of cleaning a room using reinforcement learning.

How would that look? So there we have the vacuum cleaner itself, the automatic robot vacuum cleaner, some dirt, and a wall on the right hand side.

If it cleans up the dirt, it gains a reward, so it gets one point.

If it bumped into a wall, it's given a penalty, so it loses a point.

Over time, it will map out where the dirt usually is, so where it gets points, how to avoid bumping into walls, where it loses points, and furniture, and which parts are more efficient.

So over time, the robot vacuum cleaner will become more efficient by working out where it gets rewards and where it gets penalties.

In machine learning, some data is used to train the model and some is set aside to test the model.

So we've got two types of data, we've got training data and testing data.

Note that testing data is used by developers to check the model works as intended after training.

Time for a question.

Machine learning is A, a way for computers to follow step-by-step instructions, B, a method in which computers identify patterns in data and adjust their outputs, or C, a machine that copies human thoughts? Well done, machine learning is B, a method in which computers identify patterns in data and adjust their outputs.

Another question, what does machine learning need to work well? Is it A, lots of data and examples, B, a strong internet connection, or C, batteries and wires? Well done, the correct answer is A, machine learning needs lots of data and examples in order to work well.

Another question, supervised learning is learning, A, by copying other machines, B, by trial and error only, or C, with examples and correct answers? Well done, the answer is C, supervised learning is learning with examples and correct answers.

Unsupervised learning is when the machine is, A, given no data, B, finds pattern in data without answers, or C, memorises all data? Well done, the answer is B, unsupervised learning is when the machine finds patterns in data without answers.

Reinforcement learning is when the machine, A, adjust its behaviour through trial and error and receives feedback, B, identifies patterns in data without answers, or C, is directly programmed by humans to complete a task? And Well done, the answer is A, reinforcement learning is when the machine adjusts its behaviour through trial and error and receives feedback.

Now, time for your task B.

Match the type of machine learning to the correct description.

So we've got three types of machine learning, supervised learning, unsupervised learning, and reinforcement learning, and three descriptions, so the first description, the machine groups and organises the data without humans directly programming to find useful hidden patterns.

The second description, the machine adjusts its behaviour by trying to do a task and getting feedback.

And the final description, uses known answers to train the machine learning model.

So let's have a look at how those link up.

So supervised learning uses known answers to train the machine learning model, unsupervised learning, the machine groups and organises the data without humans directly programming to find useful hidden patterns, and finally, reinforcement learning is when the machine adjusts its behaviour by trying to do a task and getting feedback.

Second part of your task, in your own words, write two or three sentences to describe machine learning.

And you could have written something like this.

So machine learning, or ML, is part of AI that helps computers get better at completing a task over time by using data.

With ML, the computer uses data to adjust how it performs tasks rather than being programmed with fixed rules for each task.

There are different types of ML models, but they all share some common features, such as working with training data, identifying patterns or relationships, and making predictions on new data.

We can now move on to the third learning cycle in this lesson, which is to compare machine learning and programming.

So Sam's observed through this lesson that, "Machine learning sounds useful, but how is it different to traditional programming," which is a great question.

And Andeep says, "In traditional programming, every rule is written by a human for a computer to follow." In machine learning, the machine form its own rules from data.

So that's the key difference.

Although machine learning models figure out rules from data, human programmers still write the initial code, and Andeep says, "Humans also decide what ML model to use, how to train it, and which settings to use." Humans select and clean the data, check results, and make sure the model is working as intended.

So in machine learning, there are key roles for humans.

To work effectively, machine learning models need large amounts of data to develop their own rules.

The more examples of model sees, the better it can spot patterns and make actual predictions.

In traditional programming, the rules are written by humans, so very little data is needed.

That's highlighted by the fact in machine learning, you have lots of data, so a big representation of the word data, and in traditional programming, it's more about rules, so a much smaller representation of the word data.

In machine learning, the machine can improve its own performance as it uses more or better data.

In traditional programming, improvements are only made if a human updates the code.

With traditional programming, it can be hard to solve complex or unclear problems. However, in contrast, machine learning can help solve complex problems by learning patterns in huge amounts of data that will be hard for humans to understand.

Traditional code follows clear steps to do its job, and Sam says, "In a calculator, code directly tells it how to add or subtract." Traditional code is great for tasks with known rules, like converting units or working out calculations.

It's transparent, predictable, and easy to debug.

With machine learning, it can be hard to understand why the machine makes certain decisions, and Sam says, "It's not possible to go into the ML model and clearly see how it learned from the data." So this means you can't really debug a machine learning model in the same way you debug normal code.

So that's a key difference between programming and machine learning.

The table below shows some key features that can be used to compare traditional programming and machine learning.

So let's look across the table, so how it works.

Traditional programming works by a programmer writing the rules manually.

In machine learning, a model learns or adjust its output based on the rules from the data.

So what's the role of data in those two different things? So in traditional programming, data is used as an input.

Small amounts are often enough.

In machine learning, the role of data is needed in large amounts to train and improve.

So how do they improve over time? Traditional programming only changes if a human updates the code, whereas machine learning can improve with more or better data.

And finally, how does it handle complex tasks? Well, in traditional programming, it can be difficult to code for complex or unclear problems, whereas machine learning is good at learning patterns in hard tasks like images, speech, or data.

Time for some questions.

Traditional programing works by A, learning from examples, B, guessing what the user wants, or C, using fixed rules written by a programmer? Well done, the answer is C, it works by using fixed rules written by a programmer.

Machine learning is different from traditional programming because it A, improves by adjusting its output based on data, B, works without a computer, or C only works for math problems? And the answer is, A, it improves by adjusting its output based on data.

True or false? Is it easy to understand the decisions made by machine learning models and they are easy to debug? Is that true or false? Well done.

The answer is false.

Can you explain why? And it's because it can be hard to understand why machine learning models make certain decisions, and you can't debug a machine learning model in the same way you debug normal code.

So for task C, Jun has a question.

"How are machine learning and programming different?" And this prompt to help you, in your own words, compare some of the differences between machine learning and programming to Jun.

So you could have written something like this.

Machine learning and traditional programming are different ways of getting computers to solve problems. In traditional programming, a human writes the rules and the computer follows them exactly.

It works well when the rules are clear and don't change much.

Machine learning is different because instead of writing the rules, we give the computer lots of data and let it identify the patterns without human help.

This makes ML better for complex tasks, where writing rules will be too difficult.

Another difference is that traditional programming is transparent, predictable, and easy to debug.

However, you can't debug a machine learning model in the same way you debug normal code.

Here's a summary of this lesson, Machine Learning and Artificial Intelligence.

Artificial intelligence is about computer systems that adjust their behaviour based on data or follow rules to perform complex tasks.

Machine learning is a type of artificial intelligence where computers adjust their behaviour based on data and improve their performance over time.

Machine learning is different from traditional programming because it identifies rules from data instead of being programmed what to do by a human.

Thank you for joining me for this lesson, and well done for all the hard work you've put in.

I do hope to see you again in the near future in another Oak National Academy lesson.