Ateneo - UPO

The course provides cross-disciplinary methods for the critical, verifiable and responsible use of foundation models and generative artificial intelligence tools in scientific, technological, engineering and mathematical contexts. It addresses problem formulation, prompt design and refinement, task decomposition, technical and scientific reading and synthesis, the use of AI with data, tables, charts, formulas and code, independent verification of outputs, and process documentation.

The ethics component covers principles, rights, responsibilities, bias, fairness, privacy, copyright, truthfulness, safety, social and environmental impacts, governance and risk management. Operational tools such as datasheets, model cards, risk assessment sheets, auditing, human oversight and due diligence are examined, with particular attention to the AI Act, general-purpose AI systems and the use of generative AI in study, research and professional settings.

Main textbooks and background readings:
- N. Dhamani, M. Engler, Introduction to Generative AI, 2nd edition, Manning, 2025/2026.
- S. Raaijmakers, Large Language Models, MIT Press Essential Knowledge, 2025.
- P.-Y. Chen, S. Liu, Introduction to Foundation Models, Springer, 2025.
- F. Fossa, V. Schiaffonati, G. Tamburrini, Automi e persone. Introduzione all’etica dell’intelligenza artificiale e della robotica, Carocci, 2021.
- L. Floridi, Etica dell’intelligenza artificiale. Sviluppi, opportunità, sfide, Raffaello Cortina, 2023.

Advanced references and institutional materials include works by C. Huyen and Q. Miao and F.-Y. Wang, as well as the OECD AI Principles, the Council of Europe Framework Convention on AI, the UNESCO AI Competency Framework for Students, the NIST Generative AI Profile, official European Commission materials on the AI Act and the General-Purpose AI Code of Practice, the OECD Due Diligence Guidance for Responsible AI, and the Berkeley CLTC risk-management profile for GPAI and foundation models.

Specific articles, surveys, excerpts and required or optional readings are made available through the course platform.

The course contributes to the programme learning goals by providing methodological tools for the conscious, controllable and responsible use of generative artificial intelligence, without treating it as an automatic source of truth.

By the end of the course, students will be able to:
1. distinguish foundation models, generative systems, search engines, databases, calculators, simulators and disciplinary tools, identifying their capabilities, limitations and appropriate uses;
2. formulate problems and prompts clearly, specifying objectives, constraints, available data, assumptions, output formats and verification criteria;
3. use models to decompose problems, analyse technical and scientific texts, and work with data, tables, charts, formulas, code and other representations while retaining control over the process;
4. verify outputs through sources, original data, calculations, tests, external tools and comparison with reliable references;
5. recognise ethical, social, legal and security risks related to bias, discrimination, privacy, copyright, hallucinations, misuse, sustainability and human oversight;
6. apply governance, documentation, auditing and risk-management principles and tools to realistic use cases;
7. justify choices, communicate limitations and uncertainty, and transparently document the contribution of AI to the work performed.

The course carries 3 ECTS credits and includes 24 hours of teaching activities, together with the individual study workload required by applicable regulations.

No specialised knowledge of AI system design or programming is required. However, students should possess basic digital skills, the ability to read technical and scientific texts and interpret simple tables, charts, formulas or code fragments, basic disciplinary knowledge consistent with their degree programme, and the ability to distinguish data, assumptions, inferences and conclusions.

Any formal prerequisites are those established by the Degree Programme regulations.

The course combines lectures, guided analyses, exercises and laboratory activities. Lectures introduce concepts, models, methodological criteria, ethical principles and governance tools. Interactive activities include prompt comparison, analysis of correct and incorrect outputs, problem decomposition, technical reading and synthesis, interpretation of data and representations, independent verification, case analysis on bias and responsible use, preparation of documentation and risk assessment sheets, and a final integrated laboratory exercise.

Students are expected to participate actively, justify their choices and critically discuss alternative solutions. Materials available on the course platform support both attending and non-attending students.

Teaching materials, reading excerpts, worksheets, examples and organisational information are made available through the institutional platform. When generative AI tools are used in course activities, their use must be declared and documented according to the instructions provided, clearly distinguishing the student’s own contribution from content produced or suggested by the system.

Students with disabilities, Specific Learning Disorders or Special Educational Needs may request dedicated services and tools through the relevant University office. After contacting the office, they may consult the course instructor to agree on the practical implementation of the approved measures concerning teaching activities and assessment.

The examination consists of an integrated individual assessment designed to evaluate both knowledge of the principles and tools covered and the ability to apply them to realistic cases. It includes theoretical and methodological questions and the guided analysis of one or more scenarios involving foundation models or generative AI systems in STEM contexts.

Students may be required to formulate or improve a prompt, identify missing data and assumptions, evaluate generated output, propose independent verification methods, identify ethical and technical risks, suggest mitigation and documentation measures, and justify their choices using appropriate terminology.

Assessment is graded out of 30 and considers correctness and completeness, methodological application, quality of verification, recognition of risks and limitations, independent judgement, and clarity of technical communication. The passing grade is 18/30. Excellent performance requires complete and critical mastery, the ability to address complex or non-routine cases, and rigorous, transparent and well-documented reasoning.

1. Foundation models and generative AI: concepts, capabilities, limitations and appropriate use.
2. Problem formulation and prompting: context, objectives, constraints, data, assumptions, output formats and verification criteria.
3. Problem decomposition and assisted reasoning: variables, missing data, edge cases, counterexamples and alternative strategies.
4. Technical and scientific reading: controlled summarisation, extraction, comparison and detection of unsupported claims.
5. Data and disciplinary representations: tables, charts, formulas, diagrams, code, external tools and reproducibility.
6. Verification and error management: factual, logical, numerical, methodological and bibliographic errors; independent checks and human review.
7. Foundations of AI ethics: principles, rights, autonomy, transparency, accountability and responsibility across the AI lifecycle.
8. Bias, fairness and algorithmic discrimination: data, labels, targets, metrics, proxies, disadvantaged groups and dataset documentation.
9. Ethics and safety of generative AI: truthfulness, privacy, copyright, bias, misuse, robustness, information integrity and human oversight.
10. Governance, the AI Act and risk management: risk-based regulation, GPAI, NIST AI RMF, auditing, documentation and due diligence.
11. Integrated methodological laboratory: complete workflow from problem definition to verification, risk analysis and transparent documentation.

Gender dimension
The course examines how data, metrics, proxies and automated systems may produce or amplify gender disparities and other forms of discrimination. Bias and fairness activities include disaggregated impact analysis and the assessment of prevention and mitigation measures. Inclusive and non-discriminatory technical language is promoted throughout the course.

Knowledge and understanding
Students will understand the characteristics, capabilities and limitations of foundation models and generative AI systems; principles of prompt formulation, problem decomposition and output verification; the use of models with texts, data, scientific representations, calculations and code; major risks relating to reliability, bias, privacy, copyright, safety and sustainability; and key frameworks for ethics, governance, documentation, auditing and risk management.

Applying knowledge and understanding
Students will be able to formulate and refine verifiable prompts, decompose problems, critically use models with different types of content, independently verify outputs, identify ethical and technical risks, propose mitigation measures and document datasets, models, decisions and processes.

Making judgements
Students will be able to decide whether and how to use an AI system, assess output reliability, distinguish supported content from unjustified inference, compare alternatives, recognise trade-offs between performance and values, and justify their choices.

Communication skills
Students will be able to clearly and accurately describe the problem, method, role of AI, checks performed, limitations, uncertainty, risks and mitigation measures.

Learning skills
Students will be able to independently update their knowledge of emerging models, tools, regulations and practices by using technical documentation, scientific literature and reliable institutional sources.

Minimum passing level
Students must demonstrate knowledge of the fundamental concepts, formulate understandable prompts, recognise the most evident errors and risks, apply simple verification procedures and provide an essential explanation of their choices.

Advanced level
An advanced level requires integration of methodological, technical and ethical aspects, the ability to address non-routine cases, select appropriate verification methods, critically compare alternatives, propose governance and mitigation measures, and document the process rigorously, transparently and reproducibly.