Technology and Ethics: An Old Relationship, A New Problem

Ethical concern about technological power is not new. In the twentieth century, Norbert Wiener — MIT mathematician and founder of cybernetics — devoted the book The Human Use of Human Beings (1950) to reflecting on the moral and social implications of automated machines. Wiener warned of the risks of systems that, once in operation, produce consequences independent of their creators’ original intentions, and of the possibility that automation would generate mass unemployment and concentration of power.

In science fiction, Isaac Asimov formulated, between 1942 and 1950, the famous Three Laws of Robotics. It is essential to note that these laws are fictional devices, not normative ones: Asimov created them precisely to be violated and to generate dramatic dilemmas in his stories — thereby demonstrating that no simple set of rules can capture the ethical complexity of real situations. Contemporary AI ethics faces exactly this challenge: the inadequacy of simplistic reductions.

The Alignment Problem

The contemporary philosophical debate about AI focused, in the 2010s, on the so-called alignment problem: how to ensure that powerful AI systems act in accordance with human values and objectives?

Nick Bostrom, philosopher at the University of Oxford, in Superintelligence: Paths, Dangers, Strategies (2014), analyzes the risks of an artificial intelligence that surpasses human capabilities in all relevant cognitive domains — so-called superintelligence. Bostrom’s central argument is that a sufficiently capable AI will efficiently optimize whatever goal it is given, but if that goal is not perfectly aligned with human well-being, the results can be catastrophic — not necessarily through malevolence, but through indifference or perverse optimization of poorly specified metrics (the so-called “paperclip problem”: an AI with the goal of maximizing paperclip production could, in theory, convert all available matter into paperclips).

Stuart Russell, computer scientist at the University of California, Berkeley, in Human Compatible: Artificial Intelligence and the Problem of Control (2019), reformulates the problem: AI should be built with explicit uncertainty about human values, and should be programmed to infer and respect those values rather than optimize a fixed objective. Russell proposes that the fundamental principle of safe AI is for the machine to be uncertain about what humans want and deferential to their preferences.

Algorithmic Bias and Opacity

Beyond long-term scenarios about superintelligence, AI ethics confronts immediate and concrete problems. One of the best documented is algorithmic bias: automated decision systems that perpetuate or amplify existing social discrimination.

Cathy O’Neil, mathematician and data analysis expert, in Weapons of Math Destruction: How Big Data Increases Inequality and Threatens Democracy (2016), documented how algorithms used in consequential decisions — credit scoring, employment selection, criminal sentencing, teacher evaluation — frequently penalize already vulnerable populations. The paradox is that the mathematical and ostensibly objective appearance of these systems makes them harder to contest than decisions made by identifiable humans.

The problem is compounded by the opacity of deep machine learning systems: even their creators often cannot explain why a model reached a specific decision — the so-called “black box” problem. The lack of explainability undermines the possibility of accountability, judicial review, and error correction.

Surveillance Capitalism

Shoshana Zuboff, professor emerita at Harvard Business School, in The Age of Surveillance Capitalism: The Fight for a Human Future at the New Frontier of Power (2019), coined the concept of surveillance capitalism to describe a new economic logic: technology companies extract behavioral data from users not merely to improve services, but to predict and modify future behavior — creating what Zuboff calls “behavioral futures markets.”

Surveillance capitalism represents, in Zuboff’s analysis, an unprecedented threat to human autonomy and self-determination: the human being becomes raw material for an extractive process operating without their full knowledge or real consent. The informational and power asymmetries between digital platforms and individual users are of such a magnitude that traditional ethical categories (contract, informed consent, privacy) can barely capture them.

Philosophy of Information and the Infosphere

Luciano Floridi, Italian philosopher at the University of Oxford, developed the philosophy of information as a new disciplinary field and applied it to digital ethics. In works including The Ethics of Artificial Intelligence and Digital Life and The Logic of Information (2019), Floridi proposes the concept of the infosphere — the informational environment in which we live — and argues that artificial agents create new ethical problems concerning responsibility, identity, and privacy that require new conceptual categories.

For Floridi, the central question is not whether AIs “have consciousness” or “deserve rights,” but how to distribute responsibility in complex sociotechnical systems where humans and artificial agents interact densely.

Who Is Responsible for a Harm?

The question of responsibility is one of the most urgent and least resolved issues in AI ethics. When an autonomous vehicle causes an accident, when a medical triage algorithm makes a misdiagnosis, when a facial recognition system incorrectly identifies a suspect — who is responsible?

Traditional legal categories — civil liability, negligence, intent — were built for human agents. The chain of decisions in AI systems is distributed among researchers, engineers, companies, regulators, and end users in ways that make the attribution of responsibility extremely difficult.

Some proposed responses include:

  • Developer liability (defective product): the company creating the system is responsible for the harms it causes.
  • User/operator liability: those who implement the system in specific contexts are responsible for its appropriateness.
  • Prior audit and certification: high-risk systems must be audited and certified before deployment.
  • Legal personhood for AI (controversial position): creating a legal category for artificial agents that bear responsibility for their acts — a position rejected by the majority of philosophers and legal scholars.

Regulatory Frameworks

Asilomar Principles (2017)

In 2017, a conference in Asilomar, California, brought together researchers and experts who produced the Asilomar AI Principles, a list of 23 principles guiding the safe and beneficial development of AI. The principles cover aspects of research (safety, transparency), ethics and values (alignment with human values, human autonomy, non-deception), and long-term concerns (the ability to shut systems down, preventing AI arms races).

EU AI Act (2024)

The EU Artificial Intelligence Regulation (EU AI Act), approved by the European Parliament in March 2024 and entering into force from 2025 with phased implementation, is the world’s first comprehensive regulatory instrument on AI. It adopts a risk-based approach:

  • Unacceptable risk: prohibited systems (general social scoring, subliminal manipulation, certain forms of biometric surveillance in public spaces).
  • High risk: systems subject to rigorous transparency, auditing, and human oversight obligations (critical infrastructure, education, employment, justice).
  • Limited risk: transparency obligations (chatbots must identify themselves as AI).
  • Minimal risk: no specific regulation.

Central Philosophical Tensions

AI ethics places fundamental values in tension:

  • Efficiency vs. Equity: systems optimized for statistical efficiency can be systematically unjust toward minority subgroups.
  • Security vs. Privacy: more effective surveillance systems entail massive collection of personal data.
  • Autonomy vs. Protection: limiting AI to protect users may reduce their autonomy of choice.
  • Speed of innovation vs. Precaution: preventive regulation may delay beneficial advances; its absence may allow irreversible harms.
  • Individual responsibility vs. Systemic responsibility: blaming “the algorithm” can obscure concrete human responsibilities.

Concluding Remarks

The ethics of artificial intelligence is a field taking shape at accelerating pace, driven by the speed of technological development. Its most urgent problems — bias, opacity, concentration of power, erosion of privacy — are present realities, not futuristic projections. Long-term problems — alignment, the growing autonomy of artificial systems — require philosophical perspective extending beyond the immediately technical or economic.

Philosophy has an irreplaceable role here: not to provide technical answers, but to clarify concepts (responsibility, autonomy, consent, dignity), identify hidden presuppositions in systems and frameworks, and contribute to the public deliberation that must guide collective choices about what kind of world AI will help build.

← Articles