Core AI Concepts (Module A.2)
ModulesCore AI Concepts (Module A.2)
This module introduces the ethics foundations of artificial intelligence (AI) within European policy frameworks. It covers the definition of an AI system according to the EU AI Act, along with core AI concepts and techniques.
It further explores the ethics-by-design approach, showing how ethics principles can be embedded throughout the AI lifecycle. The module also presents the requirements for trustworthy AI as defined by the High-Level Expert Group on AI and introduces the ALTAI framework as a practical self-assessment tool.
Overall, this module equips learners with the essential prerequisite knowledge to understand AI fundamentals and serves as an introductory module to the course.
Learning Goals
- Become familiar with core AI concepts and terminology
Core AI Concepts
Artificial Intelligence (AI) systems process data to recognize patterns, make decisions, and execute actions. At the foundation lies Machine Learning, which enables systems to learn from data to identify patterns and make predictions without manual rule coding, operating through supervised, unsupervised, and reinforcement learning paradigms. Neural Networks and Deep Learning utilize brain-inspired layered architectures to process complex structures like images, speech, and text, though their multi-layered design introduces explainability challenges, often operating as a "black box". Generative AI and Large Language Models (LLMs) scale these capabilities to produce novel content across text, code, images, and media; however, they require critical oversight due to risks such as non-factual "hallucinations" and embedded data bias. Finally, the evolution from task-specific AI Agents to autonomous Agentic AI introduces systems capable of proactive planning and goal-setting, making robust governance, accountability, and human oversight essential safeguards.
Data - The Fuel That Powers AI
Data serves as the foundational fuel for all Artificial Intelligence systems. The overall quality, quantity, representative diversity, and fairness of training datasets directly govern how accurately and equitably an AI model behaves. Safeguarding sensitive information, ensuring broad representation, and eliminating dataset biases are critical requirements to prevent automated systems from reinforcing unfair decisions.
Machine Learning — How AI Learns from Data
Machine Learning allows computers to recognize underlying patterns and formulate predictions from data without requiring human developers to code every rule explicitly. It encompasses Supervised Learning (learning from labeled examples), Unsupervised Learning (discovering hidden patterns or clusters without labels), and Reinforcement Learning (learning through trial and error via rewards). Because models learn directly from historical examples, any embedded biases in training data will inevitably be reflected and potentially amplified in output decisions.
Neural Networks & Deep Learning
Deep Learning is a specialized subset of Machine Learning driven by Neural Networks—models inspired by the human brain that process inputs through layers of interconnected artificial neurons. These layered architectures enable advanced applications such as image recognition, speech processing, and text generation. However, their internal decision-making processes are highly complex, giving rise to the "black box" challenge where explaining the precise reasoning behind a specific output remains difficult.
Large Language Models & Generative AI
Generative AI utilizes patterns learned from vast datasets to generate entirely new content, including text, images, code, audio, and video. Large Language Models (LLMs) specifically leverage massive text corpora to synthesize fluent, human-like language. Despite their sophisticated outputs, these models lack true understanding and can produce "hallucinations"—outputs that appear highly convincing yet are factually incorrect or nonsensical—requiring continuous human verification.
AI That Acts - From Agents to Agentic AI
Artificial Intelligence is evolving from passive content generation to active execution, distinguished by varying degrees of autonomy. While traditional AI Agents operate reactively within a narrow scope to complete specific tasks under explicit instructions, Agentic AI exhibits proactive behavior by setting priorities, breaking down complex goals into multi-step plans, and taking independent initiative over time. As AI systems gain greater operational autonomy and require less step-by-step human guidance, the necessity for formal oversight, clear accountability, and ethical safeguards increases significantly.
