Ethics and Artificial Intelligence in the next Athics webinar

Ethics and artificial intelligence are now two inseparable dimensions: with the advancement of generative technologies, the proliferation of decision-making algorithms and the growing adoption of AI in critical sectors such as healthcare, finance and education, it is becoming essential to question not only what artificial intelligence can do, but also what it should do. This evergreen guide tackles the core topics of ethical AI in a structured way, from bias to the European AI Act, all the way to concrete practices for responsible artificial intelligence in business.

As highlighted in the Stanford University AI Index 2024 report, the rapid development of artificial intelligence raises important ethical questions related to transparency, responsibility and social impact. The European Committee on Artificial Intelligence (CAHAI) has also reiterated the urgency of ethical governance that puts human rights and human dignity at the centre.

What do we mean by ethics and artificial intelligence?

Talking about ethics and artificial intelligence means questioning the set of principles, values and practices that should guide the design, development and use of AI systems so that they are fair, transparent and respectful of people's rights. This is not an abstract matter: every time an algorithm grants a loan, suggests a diagnosis, screens applications or moderates online content, it is making or influencing a decision that has concrete consequences on someone's life.

International organizations have tried to set some firm reference points. The OECD AI Principles point to values such as inclusive growth, respect for the rule of law, transparency and accountability. UNESCO, with its Recommendation on the Ethics of Artificial Intelligence, has promoted the first global normative instrument on the subject, adopted by its member states. The common thread is clear: technology must remain in the service of human beings, and not the other way around. From this come the pillars we will examine: fairness, transparency, privacy and accountability.

Bias and fairness: when algorithms discriminate

One of the most delicate themes of ethical AI is bias, that is the systematic distortion that can creep into a model's results. Algorithms learn from the data they are trained on: if that data reflects historical prejudices or does not represent the real population in a balanced way, the system risks amplifying inequalities instead of correcting them. A recruitment model trained mostly on male profiles, for example, could systematically penalize female candidates without anyone having explicitly decided so.

Fairness, that is algorithmic equity, is the answer to this problem. Guaranteeing fair decisions requires attention across the entire AI value chain. The most effective practices include:

• Data quality and representativeness: making sure that training datasets include the different groups involved in a balanced way.
• Fairness testing: measuring the model's results across different subgroups to spot disparities in treatment.
• Periodic audits: monitoring systems over time, because a model that is fair at launch can degrade as data evolves.
• Diversity in teams: those who design AI bring a perspective with them, and diverse teams recognize blind spots more easily.

Stanford's AI Index report has documented for years how the evaluation of fairness and reliability of models has become a central chapter of artificial intelligence research. Tackling bias is not only a moral duty, but a quality requirement: an unfair system is also a system that fails in a predictable and harmful way.

Transparency and explainability of models

Many artificial intelligence systems, in particular those based on deep neural networks, work like "black boxes": they produce an output without it being immediately clear what logical path led to that conclusion. When an automated decision affects rights, access to services or economic opportunities, this opacity becomes an ethical problem and, increasingly often, a legal one too.

Transparency and explainability (explainable AI) aim to bridge this gap. Making a model explainable means being able to reconstruct which factors weighed on a given decision and to communicate it in an understandable way to the people involved. This is achieved by carefully documenting the data and models used, adopting techniques that highlight the contribution of individual variables and, above all, keeping human oversight over the most sensitive choices. The traceability of actions, a principle we explored when discussing artificial intelligence and security, is also the prerequisite for being able to demonstrate, after the fact, that a system operated correctly. Explainability is not a constraint that slows innovation, but the condition that allows people to trust intelligent machines.

Privacy and data protection

Artificial intelligence feeds on data, often personal. Language models, recommendation systems and predictive analytics tools process huge amounts of information about real, flesh-and-blood individuals. This is why privacy is one of the non-negotiable pillars of any serious discussion on ethics and artificial intelligence.

In Europe, the reference framework remains the GDPR, which sets principles such as data minimization, purpose limitation and the right of individuals not to be subject to fully automated decisions when these produce significant effects. Designing a privacy-respecting system means collecting only the data that is truly necessary, protecting it with adequate technical measures, defining who can access it and for how long it should be retained. We discussed the interplay between regulation and technology in our webinar on the AI Act, privacy and the digital future, where it emerges that protecting data is not an obstacle to innovation, but a guarantee of long-term sustainability. A company that handles data with care builds trust, and trust is the real capital of the digital economy.

Accountability: who answers for AI decisions?

When an automated system makes a mistake, the inevitable question is: who is responsible? Accountability addresses exactly this knot, establishing that behind every artificial intelligence system there must be clearly identifiable subjects who are responsible for its consequences. It is not enough to say "the algorithm decided it": responsibility cannot be delegated to a machine.

Building accountability means defining precise roles and processes:

• Clear responsibilities: establishing who answers for the development, release and monitoring of a system.
• Human oversight (human-in-the-loop): keeping a person able to intervene, correct or overturn the most sensitive decisions.
• Documentation and traceability: recording choices, data and actions so that what happened, and why, can be reconstructed.
• Redress mechanisms: guaranteeing people the possibility to challenge an automated decision and obtain a review.

This principle is closely linked to the idea, also expressed in literature, that responsibility always remains human. As Dostoevsky reminds us in a phrase we will return to later, "every man is responsible for everything before everyone": a warning that takes on new meaning in the era of decisions delegated to machines.

The European AI Act: rules for trustworthy AI

The European Union has chosen to translate these ethical principles into binding rules with the AI Act, the first organic regulatory framework on artificial intelligence in the world. The structure, described by the European Commission, is founded on a risk-based approach: the more a system can affect people's safety and rights, the stricter the obligations become.

In short, the regulation distinguishes between several risk categories:

• Unacceptable risk: prohibited practices, such as generalized social scoring by public authorities.
• High risk: systems used in sensitive areas (recruitment, credit, critical infrastructure), subject to rigorous requirements for data quality, transparency and human oversight.
• Limited risk: applications subject to transparency obligations, for example the duty to inform users when they are interacting with an AI system.
• Minimal risk: the majority of applications, which require no specific obligations.

For companies, the AI Act is not mere bureaucratic compliance, but a design compass: those who adopt principles of data quality, explainability and governance from the outset will already be aligned with the regulatory requirements. We explored the concrete implications for businesses and citizens in our piece on the AI Act, privacy and the digital future.

How do you build responsible AI in a company?

Translating ethics into everyday practice is the real challenge for organizations. Responsible AI does not arise from a statement of intent, but from concrete choices that run through the entire life cycle of a project: from defining objectives to choosing data, from testing to going into production, all the way to continuous monitoring. The starting point is always to ask not only "does it work?", but "is it fair, understandable and safe?".

The practices that distinguish a mature adoption of artificial intelligence include defining shared ethical principles, assessing risks before release, protecting data by design and maintaining a human safeguard over critical decisions. This is the same approach that guides the development of solutions like those of Athics, where building AI agents for business is accompanied by control, traceability and oversight mechanisms. The same applies to sensitive fields such as information: in our piece on AI journalism we saw how transparency and source verification are essential conditions for using artificial intelligence without betraying the public's trust. Responsible AI, in the end, is the kind that generates value without sacrificing compliance, trust and people's dignity.

Ethics and artificial intelligence in the book "Dostoevsky's Artificial Intelligence"

"Dostoevsky's Artificial Intelligence", the book by Prof. Luca Mari, Full Professor of Measurement Science at LIUC – Università Cattaneo, tries to answer the deepest questions about the relationship between ethics and artificial intelligence. The volume stands out for its original and reflective approach: it is not a technical manual, but a philosophical essay that intertwines Dostoevsky's thought with the contemporary challenges posed by AI.

In the text, Prof. Mari addresses key concepts such as freedom, responsibility, decision and awareness, proposing a vision of artificial intelligence that is not limited to the technological dimension, but calls into question our humanity. The author suggests that living with AI implies a choice: to be guided by automatisms or to exercise individual awareness and responsibility. From this perspective, AI is not just a tool, but a "cultural fact", capable of transforming our way of thinking, acting and living together. A strong invitation not to forget that, as Dostoevsky wrote, "every man is responsible for everything before everyone".

The Athics webinar: an opportunity to reflect together

Starting from the book "Dostoevsky's Artificial Intelligence", Athics organized a webinar in the "AI Readings for the Summer" series, with Prof. Luca Mari as a special guest, who discussed the topics of ethics and artificial intelligence, exploring the relationship between technological innovation and human values. A precious opportunity for those who work in the world of technology, training or communication, and wish to develop a more critical and aware vision of the present.

Why these topics matter

• To understand the role of the human being in the age of intelligent machines
• To discover how literature can illuminate technological choices
• To reflect on the meaning of the word "responsibility" in the context of AI

Conclusions on ethics and artificial intelligence

Technological progress is never neutral. Talking with awareness about the relationship between ethics and artificial intelligence is essential to build a sustainable, human and shared future. Bias and fairness, transparency, privacy, accountability and compliance with the AI Act are not obstacles to innovation, but the foundations on which trust rests. If you want to adopt artificial intelligence in your organization in an ethical, safe and compliant way, talk to our team: we will help you turn these principles into concrete solutions.