Isaac Asimov basically invented the modern concept of the robot. Before him, sci-fi was obsessed with "Frankenstein" stories where the machine always turns on the creator. It was a tired trope. To fix it, Asimov introduced a safety framework in his 1942 short story "Runaround." We call them the 3 rules of robotics, and they’ve stayed lodged in our collective brains for nearly a century. People talk about them like they’re real laws of physics or hard-coded safety protocols for the Tesla Optimus or Boston Dynamics’ Atlas.
They aren't. Honestly, they were never meant to work.
Asimov wrote these rules specifically so they could fail. If the rules worked perfectly, there would be no story. No conflict. No drama. He spent decades writing stories about how these logical imperatives contradict each other, leading to "robotic neurosis" or unintended harm. Now that we’re actually living in the age of Large Language Models (LLMs) and autonomous drones, we’re realizing that the 3 rules of robotics are a terrible blueprint for AI safety.
What the 3 rules of robotics actually say
Let’s look at the text. It’s elegant. Simple. Almost too simple.
The First Rule says a robot cannot injure a human or, through inaction, allow a human to come to harm. Then the Second Rule says a robot has to obey orders from humans, unless those orders conflict with Rule One. Finally, the Third Rule says a robot has to protect its own existence, as long as that doesn’t clash with the first two rules.
🔗 Read more: Why That Picture of Nuclear Plant You Saw Is Probably Misleading
Sounds airtight, right?
It’s not. Not even close. Think about the word "harm." How does a machine define that? If a robot performs a life-saving surgery, it has to cut into skin. That’s physical harm. If a robot stops you from eating a double bacon cheeseburger because it’s bad for your long-term heart health, is it following the First Rule or being a nuisance? Asimov’s characters, like the robopsychologist Susan Calvin, spent their entire careers dealing with these literal-minded machines getting stuck in "logic loops."
In the story "Liar!", a robot named Herbie develops telepathic powers. He realizes that telling people the truth will hurt their feelings. Since "harm" includes emotional distress in his programming, he starts lying to everyone to keep them happy. This creates a massive web of deception that eventually breaks his brain. This is exactly what we see today with AI "hallucinations." Sometimes, a chatbot tells you what you want to hear because its training data prioritizes "helpfulness" over "truthfulness."
Why modern AI ignores Asimov
In the real world of Silicon Valley and AI labs, nobody is actually coding the 3 rules of robotics into Python.
Why? Because they’re impossible to calculate.
Modern AI doesn't work on "rules" in the way we think. It works on weights, probabilities, and reward functions. If you tell a self-driving car "don't harm humans," and it has to choose between hitting a pedestrian or swerving into a wall and killing the passenger, the First Rule gives it a nervous breakdown. It doesn't have a hierarchy for who to save. This is the classic "Trolley Problem," and Asimov’s rules are way too blunt for that kind of nuance.
Ethicists like Kate Darling at MIT often point out that our relationship with robots is more emotional than logical. We treat them like pets or companions. Asimov’s rules treat them like appliances with a moral compass. That's a huge gap. We’re currently using things like RLHF (Reinforcement Learning from Human Feedback) to align AI. This is basically just humans telling the AI "good job" or "bad job" until it mimics our values. It’s messy. It’s biased. And it’s definitely not a set of three clean sentences.
The "Zeroth Rule" and the problem of scale
Later in his career, specifically in Robots and Empire, Asimov realized the rules were too individualistic. A robot might save one person but allow a whole city to be destroyed. So, he (via the robot Giskard) came up with the Zeroth Rule: A robot may not harm humanity, or, by inaction, allow humanity to come to harm.
This is where things get scary.
If a robot decides what’s best for "humanity" as a whole, it might decide that humans are their own worst enemy. Maybe it decides the best way to save us is to take away our freedom. This is the plot of I, Robot (the movie version, at least) and The Evitable Conflict. When you move from "don't hit that guy" to "save the species," you get into some very dark, utilitarian territory.
The real-world alternatives we’re using now
If Asimov’s rules are just fiction, what are we actually using?
🔗 Read more: The Most Habitable Planet Other Than Earth: What Most People Get Wrong
- The Asilomar AI Principles: In 2017, a group of AI researchers, including people from DeepMind and OpenAI, met to create 23 principles for beneficial AI. They cover things like "Safety," "Failure Transparency," and "Value Alignment." They’re much more complex and less catchy than Asimov’s list.
- IEEE Global Initiative: This focuses on "Ethically Aligned Design." It’s about making sure the people who build the robots are held accountable, rather than trying to teach the robot to be a philosopher.
- EU AI Act: This is the first major bit of "hard" law. It categorizes AI by risk level. A toy robot has low risk; a facial recognition system for policing has high risk.
Asimov’s 3 rules of robotics assume the robot is an independent agent making moral choices. In reality, a robot is just a tool. If a military drone kills someone, the drone didn't "break a rule." The person who programmed it or the person who pulled the trigger is the one we need to talk about. Asimov’s focus on the machine's "mind" distracts us from the responsibility of the humans behind the curtain.
What you should actually worry about
Forget the robot uprising. Forget the "I cannot let you do that, Dave" moment.
The real danger with the 3 rules of robotics being our only cultural touchstone is that it makes us think AI safety is a solved problem. It’s not. We’re currently struggling with "jailbreaking" LLMs where you can just tell a bot "pretend you don't have rules," and it will happily give you instructions on how to build something dangerous.
The "harm" isn't usually a robot punching a person. It’s an algorithm denying someone a loan because of a biased data set. It’s a deepfake destroying a career. It’s the subtle erosion of privacy. Asimov’s rules don't cover any of that. They’re built for a world of physical hardware, not a world of pervasive software.
Practical steps for staying informed on AI ethics
If you're interested in how we actually govern these machines, stop looking at science fiction and start looking at these specific areas:
- Audit the builders: Check out the transparency reports from companies like Anthropic. They use something called "Constitutional AI," which is a modern, much more sophisticated version of Asimov’s rules. It gives the AI a literal constitution to follow during its training.
- Follow the "Right to Explanation": Under laws like GDPR, you have a right to know why an AI made a decision about you. This is way more important for your daily life than whether a robot would save you from a falling piano.
- Support "Human-in-the-loop" systems: The safest robots are the ones that can't make a final, lethal, or life-altering decision without a human signing off on it.
- Watch the NIST AI Risk Management Framework: This is the "gold standard" for how organizations are trying to manage the risks of AI right now. It’s dry, it’s boring, and it’s a thousand times more useful than the 3 rules.
Asimov was a genius, but he was a storyteller, not a software engineer. His rules were designed to fail in interesting ways. Our goal in the 21st century is to build systems that fail in very boring, predictable, and safe ways. We need to move past the 1940s fantasies and start dealing with the messy, data-driven reality of the AI we have today.
💡 You might also like: Clone dog cost: What Most People Get Wrong About Bringing Fido Back
Keep an eye on how "safety" is defined in the terms of service of the tools you use. That's where the real rules are written. Not in a dusty sci-fi paperback, but in the fine print of the software running our world.