1. What is artificial intelligence, and how does it work?

This chapter provides a comprehensive overview of the nature and history of artificial intelligence (AI), covering its development and applications in everyday life. You will learn how AI works and how text-based AI models, such as ChatGPT, differ from other AI applications. The importance of neural networks and the transformer architecture in AI development will also be introduced. The chapter concludes with a discussion of the current technical limitations of AI and potential future directions, helping to anticipate how AI might evolve in the coming years.

What is Artificial Intelligence?

Since the launch of ChatGPT in November 2022, artificial intelligence (AI) has become a widely discussed topic. AI is a branch of computer science that aims to create systems capable of performing complex tasks that traditionally require human intelligence and reasoning.

AI, in some form, has existed for over 50 years, and we have encountered AI applications in everyday life long before ChatGPT. For example, Google Translate, launched in 2006, uses machine learning models similar to those used in text-based AI. However, earlier AI applications were much less complex than the ones available today. The availability of vast amounts of digital data on the internet and the significant increase in computational power has driven the rapid progress in AI.

AI is a computer program that possesses predictive and computational abilities similar to human intelligence, allowing it to generate content, plan, learn, and be creative.

At its core, AI is a computer program that possesses predictive and computational abilities similar to human intelligence, allowing it to generate content, plan, learn, and be creative. While it is debatable whether AI in its current form fully matches human cognitive abilities, it has already mastered many fundamental tasks.

From their inception, computers have essentially been advanced calculators with vast computational abilities. However, traditional computers strictly follow predefined instructions, whereas AI systems can generate multiple possible responses to an input—much like a human would.

Below is a detailed explanation of key terms necessary for discussing and understanding AI. As mentioned, AI is a computer program that analyses existing data and, based on input, provides a prediction or solution. AI applications are divided into two categories: narrow AI and general AI.

• Narrow AI refers to applications designed for specific tasks.
• General AI refers to applications capable of learning and operating independently across different fields and tasks. Currently, widely available AI applications fall under narrow AI.

Generative AI can create new texts, images, music, or other content based on existing data.

In addition, it is important to distinguish generative AI, which can produce new texts, images, music, or other content based on available data. Depending on the type of medium generated, different AI applications are categorized accordingly. The most common types of generated media include text, images, music, and video. This learning material focuses on generative AI applications most commonly used in universities, specifically text-based AI applications.

Neural networks: the core of AI

A neural network is one of the key tools in machine learning, and as the name suggests, it is inspired by the human brain. A comprehensive explanation of neural network programming and operation is beyond the scope of this material. Instead, the following is a simplified overview of how a neural network works and its role in AI systems.

A neural network consists of many interconnected points called neurons. These neurons work together to recognise patterns, perform calculations, and make decisions. Neurons are not just connected in a single layer but are organised into multiple interrelated layers. In a simplified form, a neural network consists of an input layer, hidden layers, and an output layer. When you enter a prompt into a text-based AI, the input layer converts words into numbers, the hidden layers process the information, and the output layer provides a response. In other words, the hidden layers take information from the input layer and transform it into an appropriate output.

Each neuron in the hidden layer calculates a weighted sum of the inputs. The output layer then generates the final response or prediction. In such a multi-layered neural network, there can be a vast number of neurons and interconnections, known as parameters. Essentially, parameters are the weights and biases between neurons that the model has learned during training to achieve the correct output from the input. The number of parameters in an AI neural network model is often used as a measure of its capability.

Beyond parameter count, an AI model’s computational power (processors, GPUs, and memory) and the quantity and quality of training data also determine its effectiveness. More parameters typically make a model computationally more expensive to run. While greater computational power often leads to a better AI model, it also consumes a significant amount of energy. The volume and quality of training data are crucial because the more high-quality natural language data the AI model is trained on, the better it is likely to perform.

A key concept related to neural networks is the transformer. The transformer is a relatively recent type of neural network architecture developed by Google in 2017. It allows AI to focus on the most important parts of an input. This means that the significance of words in a given prompt is determined not just by their proximity to each other but also by the broader context. Another major advantage of the transformer architecture is that it processes words and sentences in parallel rather than sequentially, making these AI models significantly faster.

Future trends in AI development

AI is evolving so quickly that keeping pace is a challenge. In the near future, the goal is to develop models capable of solving complex tasks at a human level. This, in turn, will continue to reshape the job market.

When considering AI models and their development after 2022, one clear trend emerges – one that is unlikely to change in the near future: the pace of advancement in AI is so rapid that keeping up with it is extremely difficult. Although models, mostly classified under narrow AI, made a significant leap forward between 2022 and 2024, the goal remains to advance toward general AI – an AI capable of performing at a human level across various media and specialized tasks. Models are continuously improving, and after just a few generations, they can tackle multi-step tasks, where AI asks follow-up questions to refine solutions and predicts the necessary steps to complete large-scale and complex problems. While such models may initially be slower than the best AI models available today, they will be capable of solving tasks that require highly sophisticated reasoning.

There has been much discussion about AI’s impact on the labour market: in the future, the widespread adoption of AI and automation will reshape employment. Routine and repetitive jobs, such as data entry, basic customer service, and assembly line monitoring, are likely to become increasingly automated. At the same time, new roles will emerge, including AI specialists, AI system maintenance experts, and ethics professionals who assess and regulate the impact of AI applications. This shift means that workers will need to acquire new skills or transition to fields where human creativity and complex problem-solving remain irreplaceable.

Self-assessment