How does AI work? A simple explanation with examples

AI can help you write content for your website or emails, and it can even assist you in generating images and designing your website, but how does it actually work?  

In this article, we’ll answer that fundamental question: how does AI work? We’ll use clear examples and explain it in a way that everyone can understand. 

What is AI?

Artificial intelligence, or AI, refers to a computer program that behaves in a way similar to human intelligence. It can learn, interpret natural languages, and communicate in a way that feels human-like. 

Examples:

• An AI can learn to write by recognising patterns in human texts. It can respond to questions and prompts in a way that resembles a human. This ranges from a chatbot answering customer queries to a specialised AI writing about technical questions in a workplace. 
• AI can also be used in hardware, such as in self-driving or driver-assist cars. AI can interpret human-designed infrastructure and control the vehicle as a human would. 

What AI is not

Today’s AI technology is not human and does not possess self-awareness. While it may be skilled at mimicking human behaviour, it has no consciousness.  

Artificial general intelligence (AGI), which could have more human-like traits, may appear in the future, but this technology does not exist today, and its development is purely hypothetical. 

What are machine learning and deep learning? 

Many AI systems use machine learning. Later in this article, we’ll discuss how exactly AI can learn, but in short, machine learning is the technique AI uses to identify patterns in topics, languages, or expressions. Machine learning is often how AI processes data that it can use to generate responses. 

Deep learning is an advanced form of machine learning used by many modern AI systems. It employs neural networks, which allow the AI to recognise patterns in a way that resembles the human brain. 

How does AI work? 

AI is a complex technology, so we’ll explain it only at a basic level with major simplifications in this article. With that in mind, here’s how AI is created, how it works, and how it can be used. 

Data collection 

To create an AI, a large amount of data is required. One reason AI has developed rapidly in recent years is that the ability to collect and process data has advanced quickly. Generally speaking, AI becomes more sophisticated when it has access to more data. Therefore, it is essential to gather as much relevant and high-quality data as possible. 

We’ll set up an example to show how this works. Suppose we want to create AI that can generate images of animals. For AI to learn to illustrate animals, we first need to gather a collection of images of animals that it can learn from. It’s not enough to collect a few hundred or even a few thousand images. Many AI systems use millions, billions, and in some cases, trillions of data points. 

With millions or billions of animal images, AI can go far, but it’s crucial that these images are of high quality. This is where categorisation and training come into play. 

Categorisation and training 

There are primarily two methods of AI training that you need to know about: 

• Supervised learning 
• Unsupervised learning 

Supervised learning 

In supervised learning, AI learns by studying both the input data and the expected outcome. For example, if we want AI to learn to recognise animals, we feed it a large number of animal images. Each image comes with an expected answer, such as “elephant.”  

This way, AI can create a connection between the image of an elephant and the word elephant. When this is repeated enough, it can identify a pattern of what an elephant looks like. 

We can liken this to a parent sitting with their child, flipping through a book of animal pictures. The parent points to each animal and says the animal’s name. The child can then associate the picture of the animal with the word for it. After enough repetition, the child can point to an elephant and say elephant. 

Supervised learning can create high-quality AI. Since the learning occurs with data that has a given result, there’s less risk of unwanted patterns being identified or misunderstandings occurring. The downside to supervised learning is that finding quality data is expensive. It’s easier to find images of animals than images where each one has a verified, correct description of the animal. 

Data for supervised learning can come from many sources. You’ve likely contributed to a supervised learning source when you’ve logged into a website and been asked to click on images showing a bus, a crosswalk, or a bicycle. Data from these questions not only helps websites verify that you are human but also helps validate the content of images used in supervised learning. 

Common uses: Image identification, text analysis, and answering questions. 

Unsupervised learning 

In unsupervised learning, AI is given a large dataset to identify patterns without clear instructions or access to answers. These patterns can be used, for example, to imitate human creativity. An AI that reads thousands of fantasy books can identify a pattern for how such books are written and recreate seemingly creative texts by mimicking the patterns it has learned. 

The advantage of unsupervised learning is that it can be done with data that is easier and cheaper to collect. This makes it more cost-effective and simpler to conduct learning on a larger dataset, which can improve both quality and scope. 

Common uses: Generating texts, images, music, and voices. 

Semi-supervised learning 

Many AI systems are trained using semi-supervised learning, which combines learning methods. In our animal example, the training database may include a total of one million animal images, with 100,000 images having descriptions and 900,000 lacking any description.  

Using this method, the AI can, for example learn correct descriptions for various animals while still accessing a vast library of images to use for generating pictures of elephants. 

How AI writes 

AI can generate results in many forms: it could be an image, text for an email, a design for a website, a spreadsheet, and more. One common feature across all AI applications is that some type of input is required to produce an outcome. For instance, this could be an AI prompt, which is a text submitted as a question to the AI. 

How AI generates a response depends on what is being created, but if we take ChatGPT, which powers our AI writing assistant, as an example, it always starts with a prompt. Suppose you write, “write a short description of the animal elephant”. The writing assistant starts by analysing the user’s prompt and breaking it down into tokens. A token can be a word, part of a word, or several words. These tokens help identify patterns that can be used to generate new text. 

As a response to a prompt, the writing assistant generates a token that it believes is the most likely continuation of the text. It then repeats this analysis and generates the next token based on the user’s prompt as well as the previously generated token. This process continues, generating new tokens in sequence. 

Example

INPUT (PROMPT): Write a short description of the animal elephant. 

RESPONSE: A 

INPUT 2: Write a short description of the animal elephant. A 

RESPONSE: A mammal 

INPUT 3: Write a short description of the animal elephant. A mammal 

RESPONSE: A mammal with 

INPUT 4: Write a short description of the animal elephant. A mammal with 

RESPONSE: A mammal with large 

INPUT 5: Write a short description of the animal elephant. A mammal with large 

RESPONSE: A mammal with large ears. 

In this example, we can see how AI constructs a response based on the user’s prompt and all the tokens it has previously generated. Tokens are generated sequentially, with each new token being the most likely continuation of the previous text. 

Note that a writing assistant does not understand what it is writing and has no understanding of what an elephant is. It merely tries to generate a likely result that works as a continuation of the prompt it received. Large amounts of data and advanced learning allow AI to create results that appear human-like. 

AI for images, websites, and more 

AI can be used for much more than text. As long as you’re using the right AI system, you can generate images, videos, music, documents, websites, etc. The basics covered in our text on “How AI Works” can apply to many AI systems. 

The same principle can be applied to websites. AI can learn to design a website by analysing a large number of websites, identifying patterns in design, layout, navigation, text, and more. 

One example is one.com’s AI onboarding, which uses AI to generate a website in seconds. Users only need to provide the industry they work in, the name of the website, and preferably some information about the company. The AI then creates a complete website, including a custom design with selected images and generated text that fits the business and the purpose of the website. 

Use AI in your business

There are many ways to use AI in your business to increase productivity and improve your products and services. You can use AI to quickly get started with your website and write content for it. With one.com’s Website Builder, you can get up and running in no time. 

Once you’ve created your website, you get free access to the Writing Assistant, which can help you write new texts and edit existing ones. You can, for example, ask it to shorten a text or write a longer one based on a brief description. 

You can also use AI to craft your sales emails. The Writing Assistant can draft emails based on your requirements, write professional emails to your clients, and compose appropriate responses to incoming emails. This can make you more efficient and ensure your emails are well-written and professional. 

Of course, there are many other uses for AI. In Microsoft Teams you can use AI to transcribe meetings with your clients, and other AI features in Microsoft 365 allow you to search and manage documents in new ways. 

Testa AI

You can get started by trying AI for free for 14 days with one.com’s Website Builder. With it, you can use our AI onboarding to get started with your website and use the Writing Assistant to generate content for it. You don’t need to register a credit card, so it’s entirely risk-free to build your new AI-powered website.