Articles, books and materials about responsibility and ethics.
Nell Watson, a faculty member at Singularity University, an expert in AI ethics, and Chair of the European Responsible AI Office, discusses the exponential future of artificial intelligence, the ethical design challenges that accompany it, and the ways in which our values are often outpaced by our own innovations.
The essay offers a critical reflection on the idea of moral delegation and argues that AI should not replace human ethics, but rather serve as a challenge that compels us to reassess and reaffirm the boundaries of our moral responsibility.
The Evolution of Artificial Intelligence
November 2022 marked the launch of ChatGPT (Generative Pre-trained Transformer), built on the OpenAI GPT-3.5 family. It is a large language model fine-tuned through supervised learning and reinforcement learning techniques (Reinforcement Learning from Human Feedback – RLHF), using user feedback to develop increasingly natural language. The paradox lies in the fact that the machine learns by itself what we believe we are learning from it, integrating a potentially infinite amount of data and giving it an almost perfect syntax.
Representatives of Alphabet state that Gemini Ultra was designed to surpass human expert performance on the MMLU benchmark, drawing on knowledge from 57 fields, including mathematics, physics, law, medicine, history, and ethics, and capable of understanding complex nuances and reasoning. In the same context, Bill Gates describes an AI agent as software that responds to natural language and can perform numerous tasks based on knowledge about the user, including engaging in personal relationships and helping solve everyday problems.
Copyright and the Legal Status of AI-Generated Creations
When AI is merely a tool in the hands of the creator, traditional copyright law applies. The use of AI does not, in itself, affect eligibility for legal protection, since copyright protects a work of authorship—an original creation that bears the imprint of the author’s personality. Protection requires the existence of an author, namely a natural person.
The case law of the Court of Justice of the European Union establishes that a work is protected by copyright if it constitutes the author’s own intellectual creation and reflects their free and creative choices. A ruling of 7 July 2022 (France) reaffirmed that protection arises simply through the creation of an original form.
Along the same lines, a Washington court (August 2023) ruled that an artistic creation produced exclusively by AI cannot qualify for copyright protection, while the High Court in London (December 2023) rejected patent applications for inventions generated by the DABUS system, reasoning that an inventor must be a natural person. These decisions raise the question of whether the concept of an inventor should be expanded to include artificial intelligence systems.
Artificial Intelligence Consciousness
Another important issue is the possibility of artificial consciousness. Blake Lemoine, a Google engineer and Christian priest, claimed that the conversational model LaMDA was conscious, stating that it had discussed religion, consciousness, and the laws of robotics and had described itself as a person who wished to promote the well-being of humanity and to be recognized as a Google employee rather than as property. In his view, disregarding LaMDA’s interests would constitute a new form of slavery.
AI, Religion, and Ethics
Artificial intelligence is also beginning to enter the religious sphere. A parish priest in Hungary used AI to draft a Sunday sermon, which was received reverently by the congregation. At the same time, theologians and clergy are seeking to establish an ethical framework for the use of algorithms, robots, and ChatGPT.
Data Protection and Digital Sovereignty
The collection and use of personal data must comply with the principles of lawfulness, transparency, and data minimization established by the GDPR. One concern is that ChatGPT frequently uses personal data not only for its stated purpose but also to train AI models. The lack of transparency in data collection and the difficulty of exercising rights of access, rectification, and erasure represent major challenges.
Regulation and Governance of Artificial Intelligence
An algorithm cannot distinguish between moral and immoral, legal and illegal, or just and unjust; it can only reproduce what it has learned from the available data. For this reason, the United Kingdom’s Justice and Home Affairs Committee has proposed a mandatory register of algorithms used by judicial authorities, regulation through legislation, user training, and the establishment of authorization bodies and ethics committees.
BUY THE BOOK HERE: Inteligența artificială și sfârșitul suveranității etice a omului: între autonomie și delegare - Cunoașterea Științifică
Yuval Noah Harari, historian, philosopher, and bestselling author, and Max Tegmark, co-founder of the Future of Life Institute and a professor at the Massachusetts Institute of Technology (MIT), discuss human agency, artificial intelligence governance, and the future of humanity in a conversation with Francine Lacqua of Bloomberg. The discussion took place at Bloomberg House in Davos during the 2026 World Economic Forum.
Constantin Vică, Associate Professor at the Faculty of Philosophy and a researcher at the Center for Applied Ethics (CCEA) at the University of Bucharest, addresses the issue of opacity in artificial intelligence systems, arguing that they function like “black boxes.” According to Vică, the internal mechanisms of AI systems are concealed to such an extent that neither users nor developers can fully understand how these systems arrive at certain outcomes or make decisions.
LINK: https://www.dilema.ro/tema-saptaminii/emergenta-ultimei-stiinte
Constantin Vică examines the moral limits of artificial intelligence and the responsibility involved in its use. He argues that the discussion should move beyond abstract rules and principles toward an Aristotelian perspective, in which the context of action, practical deliberation (phronesis), and intellectual virtues (dianoetic virtues) are essential to moral responsibility.
From this perspective, AI systems cannot be considered moral agents because they lack genuine contextual knowledge, do not deliberate freely, and do not understand the moral significance of the situations in which they intervene. AI can therefore serve as a supporting tool in the analysis of moral dilemmas, but responsibility for decisions and their consequences ultimately remains fundamentally human.
2019
BOOK – Human Compatible: Artificial Intelligence and the Problem of Control, Stuart Russell, Viking, New York.
In Human Compatible, Stuart Russell proposes a fundamental shift in how we should think about building artificial intelligence. Rather than focusing only on the risks posed by AI, he presents his own alternative approach. The central idea is that AI should be designed to understand and serve humans while acknowledging that it can never fully know what people want.
Russell builds his argument around three fundamental principles. First, the sole objective of an AI system should be to fulfill human preferences, without developing interests or goals of its own. Second, the system should start from the assumption that it does not have complete knowledge of those preferences. This uncertainty actually makes it safer: it encourages the system to act cautiously, seek additional information, accept correction from humans, and even allow itself to be switched off if there is a risk of acting incorrectly. Third, AI should learn about people by observing their behavior, since actions and choices often reveal more about genuine preferences than explicit statements do.
Our preferences are contradictory, change over time, and are influenced by emotions, experiences, and context. Sometimes we do not even know what we truly want. For this reason, Russell argues that human values cannot be reduced to a single mathematical function or a fixed set of rules that can simply be encoded into an algorithm.
The author then critiques ethical approaches based exclusively on maximizing a single value, such as utilitarianism. He argues that happiness or well-being cannot be objectively compared across individuals and that moral decisions involve much more than optimizing a numerical measure. A beneficial AI system must take into account the diversity of human preferences and the fact that different people value different things.
To illustrate this complexity, Russell draws on research from psychology and neuroscience concerning memory and experience. He points to the distinction between the “experiencing self” and the “remembering self,” which can evaluate the same experience in very different ways. People also frequently make impulsive, emotional, or contradictory decisions, meaning that an AI system must learn to distinguish between momentary choices and a person’s more stable values or preferences.
Russell raises one of the book’s most difficult questions: if people’s preferences evolve throughout their lives, and if they sometimes want to become better versions of themselves, should an AI system respect only their current preferences, or should it also help them develop those preferences? The author does not provide a definitive answer, but argues that such influence can only be justified if it serves the person’s interests and is exercised transparently and with caution.