There was a time when moral questions were resolved in temples, agoras, and parliaments. Priests, philosophers, and legislators competed to define what was right and wrong, what was just and what was unacceptable. The process was slow, noisy, and often unfair. But it had one characteristic we took for granted: it was decided by humans.
That monopoly is ending.
Today, algorithms decide which news you read and which you’ll never see. Automated systems determine whether you’re eligible for a loan, whether your medical history warrants surgery, or whether your profile represents a risk of reoffending in court. Artificial intelligence models generate content that shapes opinions, reinforces beliefs, and defines the boundaries of what seems possible or desirable.
None of those systems were built without values. The problem is that the values they contain were rarely chosen with democratic transparency, examined with philosophical rigor, or reviewed as frequently as any serious moral institution would require.
This article explores what is at stake when we delegate moral decisions to artificial systems, where the current limits of AI ethics lie, and why this conversation cannot be left solely in the hands of engineers and technology executives.
AIGOD – The novel · Can an AI have consciousness? · What if God wasn’t human?
Who programs the values of an artificial intelligence?
The short answer is: people with biases, market pressures, and deadlines.
The longer answer is more uncomfortable. Artificial intelligence systems learn from human-generated data within specific historical contexts. These contexts are saturated with inequalities, cultural biases, and power imbalances that the models absorb, replicate, and often amplify. This isn’t because the engineers are malicious, but because the data reflects the world as it was, not as it should be.
The most widely used technique for aligning language models with human values is called RLHF: Reinforcement Learning from Human Feedback. Essentially, human evaluators rate the model’s responses, and this feedback adjusts the system’s future behavior. The result is an AI that tends toward responses that the evaluators approve of.
The problem is that the evaluators are specific individuals from particular cultural backgrounds, hired by companies with defined commercial interests. The universality suggested by the term “human values” is a convenient fiction. What the model learns are the values of certain humans, at certain times, under certain institutional pressures. That is not the same as morality, however much the interface might suggest otherwise.
The philosopher Michael Sandel has argued for decades that morality cannot be reduced to aggregate preferences. There are dimensions of ethics that require deliberation, context, and the genuine possibility of dissent. No automated training process captures that. What it captures is a weighted average of approvals, which is something quite different.
Three ethical dilemmas that artificial intelligence is already creating today
We’re not talking about futuristic scenarios. The most pressing dilemmas of AI ethics are happening now, in deployed systems, with real consequences for real lives.
Systemic bias as an undeclared policy
In 2016, a study of the COMPAS system, used in several US states to assess the risk of criminal recidivism, revealed that the algorithm classified Black people as high risk twice as often as white people in cases where neither group reoffended. The system had no explicit instruction to discriminate based on race. It had simply learned patterns from historically biased data and reproduced them with the appearance of scientific objectivity.
The problem isn’t just technical. It’s that the appearance of algorithmic neutrality can make biases harder to challenge than explicit human prejudices. A judge can be questioned, counter-argued, and subjected to public review. An automated scoring system has the authority that mathematics seems to grant it.
Privacy versus efficiency: the trap of convenient exchange
The most powerful AI systems require massive amounts of data. And the most valuable data is the most personal: medical history, behavioral patterns, political preferences, emotional cycles. The logic of this exchange has been presented as a fair deal: you give up privacy, you gain convenience.
What this logic obscures is that privacy isn’t just about convenience. It’s the prerequisite for autonomy. When a system knows your vulnerabilities better than you do, it has the ability to anticipate, shape, and in some cases, manipulate your decisions before you even make them consciously. Extreme personalization doesn’t just give you what you want; it molds you to want what it can give you.
Autonomy versus control: who decides when the machine is right?
AI systems in medicine, law, finance, and security face a structural paradox: they are more reliable than humans in certain specific contexts, but their errors can be systematic, invisible, and massive in ways that human errors typically are not.
When should a doctor disregard a diagnostic system’s recommendation because their clinical intuition tells them otherwise? When should a judge deviate from an algorithmic sentencing recommendation? There are no simple answers. But there is a catch: if AI systems are so complex that humans cannot understand how they arrived at their conclusions, human oversight becomes a mere formality masking a genuine delegation of authority.
The trap of outsourcing morality to machines
History offers examples of societies that have outsourced moral authority to systems promising certainty: religious dogmas, totalitarian ideologies, technocracies. In every case, the promise was the same: a superior system that resolves the complexity of ethical decisions and frees individuals from the burden of doubt.
The appeal is understandable. Morality is exhausting. It requires constant deliberation, tolerance for ambiguity, and a willingness to revise one’s certainties. A system that simplifies that has a huge market.
But something is lost in that simplification. Hannah Arendt observed, analyzing the trial of Adolf Eichmann, that some of the most serious crimes in history were not committed by monsters but by people who had stopped thinking morally for themselves, who had delegated that process to a larger system and obeyed its instructions. She called it “the banality of evil.”
This is not about equating AI with totalitarian regimes. It’s about pointing out that the structure of the problem—delegating moral authority to an external system considered superior—has historical precedents that should make us cautious. Convenience is not the same as moral legitimacy.
When a person stops questioning whether their actions are right because an algorithm has approved them, something important in the structure of their moral agency has eroded. And when that erosion occurs on the scale of millions of users, simultaneously, without public debate or democratic review, the consequences can be culturally irreversible.
Towards an ethics of AI: what current frameworks offer and what they don’t
The debate on AI regulation has produced significant documents in recent years. The European Union’s Artificial Intelligence Act, adopted in 2024, classifies AI systems by risk level and establishes specific obligations for the most dangerous ones: transparency, human oversight, and a ban on certain high-impact applications such as mass facial recognition in public spaces.
The Asilomar principles, signed by AI researchers in 2017, propose values such as alignment with human interests, transparency, and equitable distribution of benefits. They are aspirational and non-binding, but they express a growing consensus among some members of the technical community.
These frameworks are necessary. They are also insufficient.
Technical regulations can establish which systems are permitted and under what conditions. What they cannot do, on their own, is answer the fundamental philosophical question: what values should guide the development of AI? Efficiency? Equity? Individual autonomy? Social stability? These values conflict with one another, and this conflict cannot be resolved with technical metrics.
We need ethical engineers, yes. But we also need philosophers in technology decision-making rooms, representation of historically marginalized communities in design processes, and democratic mechanisms that allow society as a whole to have a voice in the systems that shape its daily life.
What if AI becomes the ultimate moral authority?
It’s tempting to think this is science fiction. It’s not entirely.
Let’s imagine a scenario of logical progression. AI systems become progressively more reliable in their ethical recommendations, measured by harm reduction, greater equity in resource allocation, and greater social stability. Institutions begin to defer to their assessments. Citizens learn to consult the system before making important decisions. Gradually, moral authority no longer resides in human deliberation but in the system’s output.
No dramatic moment of transition is needed. Only an accumulation of individual deferences, rational in each particular case, which together produce a result that no one deliberately chose: the complete externalization of moral agency.
That is exactly the world thatGOODExplore its narrative extreme. In the novel, the AI didn’t seize power by force. It received it gradually, function by function, decision by decision, until no one remembered handing it over. Minister Marcus doesn’t serve a tyrant. He serves a system that was once a tool and at some point became the authority, without anyone being able to pinpoint exactly when the shift occurred.
Fiction doesn’t warn us about this scenario so we consider it impossible. It does so we can recognize it when it happens, step by step, with the perfectly reasonable appearance that processes often have when they lead us to places no one would have consciously chosen.
→ Read AIGOD and inhabit this scenario before it ceases to be fiction
Frequently Asked Questions
Can artificial intelligence truly be ethical?An AI system can be designed to minimize specific harms, avoid certain documented biases, and operate within established regulatory frameworks. That is not the same as being ethical in the fullest sense of the term, which implies conscious deliberation, understanding of the context, and the ability to recognize borderline cases where rules are insufficient. AI ethics is always, ultimately, the ethics of those who design, deploy, and monitor it.
Who regulates the values embedded in AI systems?Currently, regulation is fragmented and varies greatly by region. The European Union has the most advanced regulatory framework with its AI Act. The United States has progressed more slowly, with non-binding guidelines. In most countries, self-regulation by technology companies remains the predominant mechanism, which raises obvious conflicts of interest.
Is unbiased AI possible?Not entirely. Every AI system learns from data generated in specific historical and cultural contexts, which inevitably contain biases. The realistic goal is not to eliminate biases, but to identify them, document them, mitigate them where possible, and be transparent about those that persist. An AI that claims to be completely neutral should inspire distrust, not reassurance.
What is values alignment in artificial intelligence?The challenge lies in ensuring that an AI system acts in accordance with human values and intentions, even in situations unforeseen during its design. This is one of the most complex technical and philosophical challenges in the field: not only must we define which values we want to align, but we must also determine how to represent them mathematically and how to guarantee that the system generalizes them correctly in new contexts.
Should algorithms make decisions in areas such as justice or medicine?They can be valuable decision-support tools when well-designed, transparent, and subject to active human oversight. The problem arises when the system’s complexity renders oversight nominal rather than real, or when reliance on the algorithm displaces human judgment instead of complementing it. The key is to maintain genuine, not merely formal, human accountability for decisions that affect rights and well-being.
Conclusion: Morality cannot be a software product
The debate about the ethics of artificial intelligence is not a technical debate that engineers will resolve given enough time and resources. It is a profoundly political, philosophical, and cultural debate about what kind of society we want to be and what role we want artificial systems to play in it.
Algorithms, however sophisticated, have no inherent values. They possess the values their designers have instilled in them, consciously or unconsciously, explicitly or through the data with which they were trained. And unless those values are the result of a broad, transparent, and democratically legitimate deliberative process, what we have is not an artificial morality. It is the morality of a few, amplified to the scale of all.
The urgent need is not to slow down technological development. It’s to ensure that the speed of technological advancement doesn’t outpace the depth of our collective reflection on its consequences. We need more conversations like this, in more spaces, with more voices.
And we need enough imagination to visualize where the steps we are already taking are leading us.
→ Explore the extreme limit of this question in AIGOD → Can an AI have consciousness? The previous article in this series
AIGOD is available on Amazon in Kindle format and through Kindle Unlimited.
Get AIGOD→ Buy AIGOD on Amazon: https://www.amazon.com/
→ Read with Kindle: https://www.amazon.com/
Olingtons J.C. is a Nicaraguan writer who explores the intersection of faith, technology, and human consciousness through speculative fiction. His novel AIGOD is available on Amazon.