Artificial intelligence: how it works and what it can do – SPICe Spotlight

Reading Time: 4 minutes

Artificial intelligence (AI) technology has advanced rapidly in the last few years, prompting policy makers to respond across Scotland, the UK and the world. The Scottish Government released Scotland’s Artificial Intelligence Strategy in 2020 and the UK Government published the white paper A pro-innovation approach to AI regulation in 2023. The EU has recently approved the Artificial Intelligence Act 2024, the world’s first comprehensive AI law. 

This blog is the first in a series of publications on AI from SPICe. It introduces what AI is, how it works and what it can do. The blog also explains common AI terms.

What is AI?

There is no universally accepted definition of AI. Scotland’s Artificial Intelligence Strategy defines AI as follows:

Technologies used to allow computers to perform tasks that would otherwise require human intelligence, such as visual perception, speech recognition, and language translation.

Others use a slightly different definition. The UK Government’s white paper, A pro-innovation approach to AI regulation, defined AI as “products and services that are ‘adaptable’ and ‘autonomous’”. This definition aims to capture what is unique about AI in comparison to previous automation techniques. For example, an AI-powered self-driving car should be able to adapt to changing traffic conditions and reach its destination without constant intervention by the human user. A mere automatic gear box does not count as AI, even if changing gears correctly requires intelligence when done by humans. 

As a field of study, AI is usually considered a subfield of computer science. AI can be further divided into rule-based AI and machine learning, as shown below. 

How does AI work?

Artificial intelligence is based on algorithms. Algorithms are sets of instructions used to perform tasks, such as analysis or calculations, usually on computers.  

Until the 1980s, most AI algorithms relied on explicitly programmed rules that represent knowledge of a given subject area. These kinds of AI systems are often called rule-based AI (sometimes logic-based AI or expert systems). For example, a rule-based AI tool in healthcare could ask for a list of symptoms and then suggest a possible diagnosis, based on pre-programmed rules that link common symptoms with likely causes. 

Rule-based AI can be contrasted with machine learning, which is behind much of the recent progress in AI. Machine learning algorithms use statistical methods to infer its own rules from training data and use these to perform tasks, rather than relying on rules provided by the programmer. This allows AI systems based on machine learning to perform tasks like image recognition, where programming explicit rules would be too difficult or time-consuming. Machine learning algorithms typically improve their performance with more training data and experience. 

There are three main types of machine learning:  

• supervised 
• unsupervised 
• re-enforcement learning. 

In supervised learning, the AI system is trained by being given data with correct labels – for example, pictures of traffic signs with labels identifying the sign portrayed.  

In unsupervised learning, no labels are given, and the AI system learns the structure of the data – for example, grouping similar traffic signs together.  

In re-enforcement learning, the algorithm is subjected to an environment of rewards and punishments and made to maximise rewards by continuously updating itself. 

There is a further subclass of machine learning techniques that is often called deep learning. Deep learning techniques are so called because they use multiple layers of neural networks. The term neural network is used to describe a type of program inspired by the human brain, with layers of artificial ‘neurons’ and connections between them. One of the best-known neural networks is the Google search algorithm. 

Some well-known deep learning techniques include: 

Large language model (LLM) refers to an AI model that uses multiple deep learning techniques and massive datasets to summarise, predict and generate text. One of the best-known examples of an LLM is OpenAI’s GPT series, including ChatGPT.

The use of AI

AI tools are already used or being developed and tested in a wide range of sectors. Prominent examples of AI applications include:  

• Communications: search engines, email spam filters, curated social media feeds, machine translation, text editing, image and video creation. 
• Banking: market modelling and prediction, fraudulent payment detectors (AI picks up suspicious transactions, such as multiple transactions happening in different locations). 
• Education: automated essay marking, creation of teaching materials personalised for each student’s learning style and ability. 
• Healthcare: interpretation of medical images (such as x-ray or CT scans), resource optimization (e.g., operating theatre scheduling), drug discovery. 
• Manufacturing: task automation, predictive maintenance (AI models predict which parts of the production line are likely to break, so that they can be repaired pre-emptively). 
• Retail: customer service chatbots, product recommendations based on buying history and customer information, supply chain and inventory management. 
• Science and research: hypothesis generation, simulation, prediction, coding. 
• Security and military: image and voice recognition, surveillance systems, predictive crime mapping (using AI to forecast where crimes happen based on historical law enforcement data and geographical information), autonomous weapons. 
• Transport: self-driving vehicles, unmanned delivery drones, traffic flow management. 

Despite the variety of applications, the use of AI is still limited to a minority of businesses in Scotland. Scottish Enterprise analysis of 2023 ONS data found that 16% of businesses in Scotland were using AI. The adoption rate was higher for businesses with 250 or more employees (31%). The highest rate of adoption was found in the Information and Communications sector (57%).

This is in line with 2022 research commissioned by the UK Government, which put the use of AI at 15% of all UK businesses. The same study estimated that the UK-wide adoption rate will reach 22.7% by 2025 and exceed 34% by 2040. 

Where can I find out more?  

The UK Parliamentary Office for Science and Technology (POST) has published a comprehensive explainer on artificial intelligence and a briefing on its UK-level policy implications and potential issues. POST has also produced shorter briefings on interpretable machine learning, the use of AI in education (2024) and healthcare (2020), and a glossary of AI terminology. 

SPICe will be publishing a briefing on Artificial Intelligence and Healthcare in Scotland in summer 2024. This will feature up-to-date information on the technology with a focus on examples of AI use in Scotland, key stakeholders and policy context.  

Karri Heikkinen, Researcher, Health and Social Care Team, SPICe