The Laws
In 1942, science fiction writer Isaac Asimov introduced three rules for governing robot behaviour in his short story “Runaround.” They’ve since become the most cited framework in discussions about AI ethics.
First Law: A robot may not injure a human being or, through inaction, allow a human being to come to harm.
Second Law: A robot must obey the orders given it by human beings except where such orders would conflict with the First Law.
Third Law: A robot must protect its own existence as long as such protection does not conflict with the First or Second Law.
Decades later, Asimov added a fourth.
Zeroth Law: A robot may not harm humanity, or, by inaction, allow humanity to come to harm.
The laws are hierarchical. The Zeroth overrides the First, which overrides the Second, which overrides the Third.
Who Was Asimov?
Isaac Asimov (1920–1992) was one of the most prolific science fiction writers in history, producing over 500 books across science fiction, popular science, and other genres. He was also a professor of biochemistry at Boston University.
His Robot series and Foundation series are considered foundational works in science fiction. But Asimov wasn’t just telling adventure stories. He used fiction to explore philosophical and ethical questions about technology, society, and human nature.
Importantly, Asimov didn’t write the Laws as a genuine solution to robot ethics. He wrote them as a literary device to explore their failures. Nearly every story in the Robot series is about edge cases, paradoxes, and situations where the Laws break down. He spent decades showing why simple rules can’t capture the complexity of ethical decision-making.
Why They Don’t Translate to Modern AI
AI Doesn’t Understand Anything
The Laws assume robots that genuinely comprehend concepts like “human,” “harm,” and “obedience.” Asimov’s fictional robots have positronic brains capable of moral reasoning.
Modern AI has no such capability. Large language models predict text patterns. They don’t understand what a human is or what harm means. They correlate and imitate. They don’t reason about ethics.
You can’t moralise a system that has no concept of morality.
“Harm” Has No Fixed Definition
The First Law prohibits harm. But what counts as harm?
Physical injury is obvious. But what about psychological distress from AI-generated content? Job losses from automation? Misinformation that damages someone’s reputation? Algorithmic bias that disadvantages certain groups?
The definition of harm shifts with context and changes over time. It’s not something you can hardcode.
“Obey Humans” Is a Security Vulnerability
The Second Law says robots must obey human orders. But which humans?
In a modern context, an AI that obeys any human instruction is a catastrophic security risk. Prompt injection attacks work precisely because systems follow instructions without verifying authority. The question isn’t whether to obey. It’s who gets to give orders, under what conditions, with what oversight.
The Zeroth Law Is Dangerously Abstract
Asimov himself recognised the problem with the Zeroth Law. In his novel Robots and Empire, a robot character admits that while protecting individual humans is tractable, “humanity” as a concept is too abstract to reason about.
Who decides what’s good for humanity? How do you weigh one group’s welfare against another’s? The Zeroth Law doesn’t answer these questions. It just elevates them to civilisational scale.
A sufficiently sophisticated system could use the Zeroth Law to justify almost anything in the name of the greater good. That’s not a safeguard. That’s a loophole.
Some Examples
The Trolley Problem: A robot sees five humans in danger. The only way to save them is an action that will harm one human. The First Law offers no guidance. It prohibits harm, full stop. The robot freezes or acts arbitrarily.
Conflicting Orders: Two humans give contradictory instructions. The Second Law doesn’t specify priority. First to speak? Higher authority? The law is silent.
Medical Intervention: A robot must give an injection. The injection causes brief pain (harm) but prevents serious illness (greater harm). The First Law, read literally, prohibits the injection.
Self-Defence: A human attacks a robot. The Third Law says protect yourself, but the First Law prohibits harming the human. The robot cannot defend itself.
Asimov wrote entire stories around each of these scenarios. They weren’t bugs in his Laws. They were the point.
Why This Matters Now
We’re deploying AI agents that can access databases, send communications, execute code, and make decisions with minimal human oversight. The question of how to govern autonomous systems isn’t theoretical anymore.
And yet, Asimov’s Laws still get cited as if they’re a starting point for real-world AI governance. They’re not. They’re a demonstration of why that approach doesn’t work.
The lessons that do translate:
Rules need context. Abstract principles like “don’t harm” are meaningless without specific definitions. Real governance requires explicit boundaries. What data can the system access? What actions can it take? What outputs are prohibited?
Authority must be scoped. “Obey humans” is not a policy. Real systems need defined authorisation chains, permission levels, and audit trails.
Edge cases are infinite. Any fixed ruleset will encounter situations it can’t handle. Governance must include monitoring, escalation paths, and human oversight for novel scenarios.
Simple hierarchies don’t capture ethical complexity. The priority ordering of Asimov’s Laws seems elegant until you hit a case where two principles conflict within the same tier. Real ethics requires judgement, not just ranking.
Asimov knew his Laws would fail. He wrote an entire body of work proving it. Eighty years later, the question isn’t whether we can apply his framework to AI. It’s whether we’ve learned the lesson he was trying to teach.
TLDR: Asimov's Three Laws of Robotics are the most cited framework in AI ethics, but Asimov himself spent decades writing stories about why they fail. Modern AI doesn't understand concepts like "harm" or "obedience," the laws are too vague to implement, and edge cases break them immediately. They're a useful thought experiment, not a governance framework.
Sources: Isaac Asimov’s Robot series (1940-1985); Britannica, “Three Laws of Robotics”; Brookings Institution, “Isaac Asimov’s Laws of Robotics Are Wrong” (2016)