![HER[AI]TAGE](https://her-ai-tage.pou-cakovec.hr/wp-content/uploads/2025/03/logo-1.png){kind=logo}
Machine learning (ML) is a cornerstone and a key enabling component of Artificial Intelligence. It represents a method of teaching computers to “learn” from data and progressively improve their performance on specific tasks over time, without the need for humans to explicitly programme every single rule or step for each scenario. Instead of rigidly following a pre-determined and fixed set of instructions, a machine learning system identifies underlying patterns, correlations, and insights within large volumes of data. It then intelligently uses these discovered patterns to make informed decisions, generate accurate predictions, or perform complex classifications. The core idea is that systems can learn from data, identify patterns, and make decisions with minimal human intervention, becoming more accurate as they are exposed to more data.
Consider the familiar example of an email service that automatically filters out spam or unwanted messages. This system doesn’t rely on a manually updated list of all possible spam keywords. Instead, it learns what constitutes spam by analysing thousands, or even millions, of example emails. This “training data” includes messages that have been identified by users as spam and messages that are legitimate. By processing this data, the ML model learns to associate certain characteristics – such as specific words or phrases, unusual sender addresses, peculiar formatting, or the presence of suspicious links – with spam. Over time, as it processes more emails and receives feedback (e.g., when a user marks an email as spam or not spam), its ability to recognise and filter unwanted messages becomes increasingly refined and accurate. This continuous improvement based on new data is a hallmark of machine learning.
In the context of adult education, ML is often the invisible engine powering many helpful features. For instance, it is fundamental to the recommendation engines found on many online learning platforms. These sophisticated engines analyse a learner’s interaction history, including the courses they’ve enrolled in, the modules they’ve completed, their performance on quizzes, the content they’ve spent more time on, and even the types of courses viewed by similar learners. Based on this complex analysis, the ML system can suggest other courses, supplementary reading materials, or specific learning activities that are likely to suit that individual learner’s progress, identified knowledge gaps, and expressed interests, thereby creating a more personalised and efficient learning path.
There are several primary types of machine learning, each suited to different kinds of tasks and data:
| Type of machine learning | Core principle / definition | Training data example | Typical application / example from course |
| Supervised learning | Learns from data that is already labelled with the correct output or category. The model identifies patterns to make predictions on new, unlabelled data. | A dataset of images explicitly labelled as “cat” or “dog” to train a model for image classification. | Email spam filters classifying messages based on pre-labelled spam/non-spam examples; predicting housing prices based on historical data with known prices. |
| Unsupervised learning | Learns from data that has no predefined labels. The model explores the data to find hidden patterns, structures, or groupings on its own. | Customer purchase history used to segment customers into different groups based on buying behaviour, without prior knowledge of these groups. | Market segmentation; anomaly detection (e.g., identifying unusual financial transactions); clustering similar news articles. |
| Reinforcement learning | Learns by interacting with an environment and receiving feedback in the form of rewards or penalties. The model (agent) aims to maximise its cumulative reward over time. | An AI agent learning to play a game like chess by making moves (actions) and receiving rewards for winning or penalties for losing. | Training AI to play complex games (Go, chess); robotics (controlling autonomous robots); optimising resource allocation in dynamic systems. |
Machine learning is the foundational technology that underpins many generative AI systems. When you interact with a generative AI tool like ChatGPT and ask it to write a story or explain a concept, the underlying Large Language Model (LLM) has been trained using machine learning techniques (often a form of supervised learning called self-supervised learning, on an immense scale) using billions of text examples from the internet and books. Through this process, it has learned the statistical patterns of language, enabling it to predict what words or sentences are most likely to follow in a given context, and thus generate coherent and relevant text. This powerful capability is highly beneficial in adult education for tasks such as providing personalised reading support, assisting in drafting professional emails or reports, summarising lengthy and complex documents, or even generating creative writing prompts.
Ultimately, machine learning isn’t just about sophisticated algorithms and complex technology; it’s about creating smarter, more adaptive tools that can better understand and respond to users’ needs. This inherent adaptability has the potential to make learning experiences more personal, engaging, and effective, thereby supporting adults in their lifelong learning journeys and skills development across various domains, including the important work of understanding, preserving, and sharing cultural and environmental knowledge.
PRACTICAL EXAMPLES
A personalised learning app for adult numeracy skills uses ML to adapt the difficulty of maths problems presented to the learner. If the learner consistently answers correctly, the app introduces more challenging concepts. If they struggle, it offers simpler exercises or targeted explanations, ensuring the learning experience is always at an appropriate level.
An adult education centre uses an ML-powered system to analyse detailed learner engagement data from its online platform. This includes metrics like how often learners log in, the time they spend on different modules, their assignment submission patterns, and their participation levels in discussion forums. By identifying patterns associated with learners who previously disengaged or dropped out, the system can proactively flag current learners who might be at risk, enabling staff to intervene with personalised support, encouragement, or additional resources.
An online job platform employs ML algorithms to provide highly personalised job recommendations to its users. The system learns from a user’s explicitly stated work history, skills, and career preferences, but also from their implicit behaviour, such as the types of jobs they click on, save, or apply for. This allows the platform to refine its suggestions over time, increasing the likelihood of a successful match.
In a digital literacy workshop designed for senior citizens, the educator effectively demonstrates how everyday smartphone applications utilise ML. Examples include voice assistant apps (like Apple’s Siri or Google’s Gemini assistant) that learn to better recognise an individual’s voice and speech patterns over time, or predictive text features on messaging apps that learn a user’s common phrases and suggest the next word, making typing faster and easier. Another relevant example for some learners could be AI code generators like GitHub Copilot, which uses ML to suggest code completions and even entire functions to developers, significantly speeding up software development.
A team leader in a dynamic corporate environment uses an ML-powered scheduling tool to coordinate team meetings. The tool learns the typical availability, stated preferences (e.g., preferred meeting times or days), and even the meeting attendance history of each team member. Based on this learned information, it intelligently suggests optimal meeting times that are most likely to accommodate everyone, minimising scheduling conflicts.
Adult learners with limited digital proficiency, particularly those who are new to a country and its language, significantly benefit from ML-powered translation tools. These tools can instantly translate website text, official document instructions, or even spoken conversations into their preferred language, greatly aiding them in navigating essential services, understanding important information, and integrating into their new community.
In the context of the HER[AI]TAGE project, which deals with vast amounts of cultural information, ML could be employed (always with rigorous expert human oversight and validation) to analyse large collections of digitised historical texts, folk songs, or images related to cultural heritage. The ML models might identify subtle recurring themes, linguistic patterns, stylistic similarities, or previously unnoticed connections between different cultural artefacts or traditions, offering new insights to human researchers.