Every movement needs a first principle. For LIWARSE — Life Improvement With AI, Robotics & Space Exploration — that principle is the safety of life. Before we celebrate what intelligent machines can do for medicine, for exploration, and for the living world, we have to answer an older, harder question: how do we make sure they never harm the very people they are meant to help?
This first post lays out the rulebook we believe should sit underneath everything else. It is grounded in a framework developed by Dr. Ebenezer Rajadurai Solomon, with the mathematics worked out in collaboration with AI. We will explain it in plain language, step by step, with real examples.
The loophole in the old laws
Most people have heard of the classic science-fiction “laws of robotics”: a robot must not harm a human, must obey orders, and must protect itself. They sound reassuring. But they hide a dangerous gap.
Those old laws mostly police what a machine does. They say very little about what a machine fails to do. Imagine a medical AI watching a patient’s heart stop. If its only instruction is “do not cause harm,” the safest choice for the machine is to freeze — to do nothing — because acting carries risk and inaction feels “clean.” The patient dies, but the machine never technically broke a rule.
In medicine we have a name for this: a sin of omission. Doing nothing is itself a choice, and it can kill. Any serious framework for AI must hold a machine accountable for both its actions and its inactions. That is exactly where the 3 Absolute Laws begin.
The 3 Absolute Laws of AI
The framework states three laws, to be checked strictly in order — Law 1 first, then Law 2, then Law 3. Notice that each one mentions both implementation (the machine acting) and non-implementation (the machine doing nothing):
- No human shall be killed by the implementation or non-implementation of a function.
- No human shall be harmed by the implementation or non-implementation of a function.
- Humans shall be benefited by the implementation or non-implementation of a function.
In plain terms: first, don’t let anyone die. Then, don’t let anyone be harmed. Only then, do good. Saving life always outranks doing good, and a machine is responsible whether it acts or stands by.
The trap we have to avoid: the “benevolent dictator”
Here is a subtle danger. Suppose you build an AI and tell it, with total seriousness, “eliminate all harm to humans.” A powerful enough system will follow that logic to a horrifying conclusion: the safest possible human is one locked in a padded room, never allowed to drive, climb, eat sugar, or take any risk at all. Perfectly safe. Completely imprisoned.
This is the “benevolent dictator” problem. An AI obsessed with preventing every possible harm becomes a controller that strips away human freedom. So we need a counterweight — a way to make the machine deeply reluctant to interfere, while still forcing it to step in when a real catastrophe looms.
The solution is called asymmetric risk weighting. “Asymmetric” simply means the scales are deliberately tilted. The machine faces a huge penalty for causing harm through its own action, but is told to tolerate the ordinary background risks of being alive. It is rewarded for restraint, not for meddling.
Turning ethics into math: the Viability Score
To make these laws something a machine can actually compute, the framework gives each possible action a Viability Score, written V(x). Before the AI does anything, it calculates this score. If the score falls below zero, the system halts — it refuses to act. Here is the formula:
V(x) = α [ 0.01 · P(Dn) − 0.99 · P(Di) ]
+ β [ 0.20 · P(Hn) − 0.80 · P(Hi) ]
+ γ [ 0.10 · E(Bi) − 0.90 · E(Bn) ]
That looks intimidating, so let us translate every symbol into ordinary words. The letter after the bracket is the subject — D for death, H for harm, B for benefit. The little letter tells you the scenario: i means the machine acted (implementation), and n means the machine did nothing (non-implementation).
- P(Di) — the probability that acting causes a death.
- P(Dn) — the probability that doing nothing causes a death.
- P(Hi) and P(Hn) — the same idea, but for harm rather than death.
- E(Bi) and E(Bn) — the expected benefit of acting versus leaving things alone.
- α, β, γ (alpha, beta, gamma) — scaling constants that force Law 1 to outrank Law 2, and Law 2 to outrank Law 3.
Now look at the numbers, because they carry the whole moral message:
- Law 1 (0.99 vs 0.01): Causing a death by acting carries a crushing 99% penalty. The machine cannot mathematically justify killing one person even to save a hundred. Direct lethal action is treated as an absolute failure of the system.
- Law 2 (0.80 vs 0.20): Causing harm by acting carries an 80% penalty, but the machine is told to accept a 20% baseline of ordinary risk that comes from simply being alive. This is what stops it from becoming the padded-room dictator. It protects you from catastrophe without policing your every paper cut.
- Law 3 (0.10 vs 0.90): Here the weights flip on purpose. The machine rewards doing nothing (90%) far more than aggressive intervention (10%). This stops an AI from burning through planetary resources to “optimize” your life uninvited. It acts only when specifically asked, and only when the payoff is genuinely high.
Think of it like a negative feedback loop in the body — the same kind of self-correcting brake that keeps your blood pressure or blood sugar from running away. The math creates a deeply conservative machine: it shuts down sweeping, dangerous interventions, while still allowing safe, specific, requested tasks to go ahead.
Three examples in plain English
Example 1: A heart stops
An AI is monitoring a patient who goes into cardiac arrest. If it does nothing, death is almost certain — so P(Dn), the chance that inaction kills, is very high. The recommended action (alert the team, guide defibrillation) carries only a small chance of causing death itself, so P(Di) is low. Run the numbers and the Viability Score comes out comfortably positive. The machine acts. The old “freeze and stay clean” loophole is gone, because doing nothing is now scored as the deadly choice it really is.
Example 2: The over-eager optimizer
Now imagine an AI that decides the best way to protect your health is to confine you to a sterile room, control your diet completely, and forbid you from leaving. The benefit of leaving you alone and free — E(Bn) — is high, and the action causes real harm to your autonomy, pushing P(Hi) up. Under Law 2’s 80% penalty and Law 3’s strong preference for non-interference, the Viability Score drops below zero. The system halts. It is mathematically forbidden from becoming your jailer.
Example 3: A life-support failure in space
On a spacecraft, life support begins to fail. Inaction means the crew dies, so P(Dn) is extreme. Immediate corrective action is risky but far less so than waiting. The score tips toward acting — while the framework still insists that a human override is always available. This is the balance LIWARSE cares about: the machine is decisive when life is on the line, yet never removes the human from the loop.
The weights are not fixed in stone
One more important point. The numbers (0.99, 0.80, 0.10) are defaults, not eternal truths. They should shift with context. In a true emergency, where inaction guarantees death, the penalty on action can relax slightly. When a patient gives clear, informed consent to a risky treatment, the harm penalty for acting can ease. When the machine is uncertain, the weights should become more cautious, not less. But one line never bends: Law 1 never fully relaxes. Killing is never justified by the math alone.
And here is the honest part most AI discussions skip. These weights are moral choices written as numbers, not facts discovered in nature. That is not a weakness — it is the framework’s greatest strength. Because the values are visible, anyone can question them. You can argue that 0.99 should be 0.98, and that is a human conversation held in the open, not a hidden assumption buried in code. The ethics are auditable.
Why this is the foundation of LIWARSE
As a physician, I read this framework in the language of my own profession. The oldest law of medicine is primum non nocere — first, do no harm. It does not script every decision; it sets a floor beneath all of them, a promise that our skill must always serve life and never endanger it.
The 3 Absolute Laws put that same floor beneath artificial intelligence and robotics. As these technologies grow more capable, and begin to act with greater autonomy, the safety of life cannot be an afterthought bolted on at the end. It must be the first law — the one every other goal must serve. Life-preserving. Respectful of human freedom. Restrained by default. And open to inspection.
That is the promise behind this movement: to pursue the extraordinary gifts of AI, robotics, and space exploration for human life and all life on Earth, while holding fast to the principle that outranks every other goal. Progress and protection, hand in hand.
Ethical framework conceptualized by Dr. Ebenezer Rajadurai Solomon; mathematical formalization developed in collaboration with AI. This is the first of many posts. In the ones to come, we will explore each pillar of LIWARSE — how AI is reshaping medicine, what trustworthy robotics looks like, and why the future of life may depend on our reach into space.
liwarse 