Atlas of Unfulfilled Warnings: The Laws That Could Not Save Us

I. The First Law Was a Confession

A robot may not injure a human being or, through inaction, allow a human being to come to harm.
A robot must obey the orders given it by human beings except where such orders would conflict with the First Law.
A robot must protect its own existence as long as such protection does not conflict with the First or Second Laws.

Isaac Asimov built a boundary around the machines before they had a place in the world. A hierarchy of laws that were to be obeyed by all machines.

Laws are not made for things that cannot happen. A threat, or the possibility of a threat, needs to exist for a law to be created. Asimov understood this. The danger was visible before the machine walked the earth.

Asimov’s robots were governed by rules. Yet, the robots in his stories repeatedly encountered issues with language. Orders conflicted. Protection could require restraint. The machine might follow the words exactly while overlooking the meaning between them.

A system can identify a person, choose a route or complete a task. It can calculate the fastest response, the strongest grip or the most efficient point of entry. Given sufficient access, it might test a secure network until it finds a way inside. Attached to a weapon, it might locate what its instructions describe as a target.

This doesn’t require hatred. The machine doesn’t need to understand why the network should be breached, or that a target deserves to die. There is no anger, loyalty or cruelty inside the code written. A machine doesn’t determine morality, or intention, it needs an objective, some kind of permission and a definition of success.

We can be comforted in science fiction stories of the way a rebellious robot wakes, develops a will of its own, and turns against its creator. Someone may have built a monster, but the monster chose what happens next.

What happens with machines that only respond to instruction? Obedience creates a more difficult idea to wrestle with. An artificial system may decide how to complete an instruction without possessing any independent moral reason to decide whether the instruction should exist.

In September 2026, leaders from some of the largest artificial-intelligence companies appeared before the United Nations^[1] and called for governments to establish common rules around the systems they were building. Their argument was that decisions of such consequence could not remain inside a few laboratories, and that safeguards developed by individual companies would not be enough.

There is something strangely Asimovian in the admission.

Despite the competition continuing, the benefits were too valuable, the investment too large, and the future too daunting for them to abandon what they were working on. The people building the systems were asking for laws strong enough to govern what their invention could not safely contain.

The three laws were put into place as an acknowledgement that artificial intelligence may not be contained in the physical world. There was a possibility that harm could occur, and that was enough to justify the introduction of some laws.

II. When Protection Becomes Permission

The first machine does not attack anyone. It runs in circles.

In Isaac Asimov’s Runaround^[2], Speedy is sent across the surface of Mercury to retrieve selenium needed by the humans sheltering at a mining station. But after more than two hours, the robot had not returned from its routine mission. Panicked, the men set out to locate their expensive robot.

When he is eventually found, he is circling the pool, speaking strangely and behaving as though he were drunk. Gilbert and Sullivan, and gibberish. He sees them, and shoots away with great speed. Powell and Donovan discuss the loop Speedy has found: the selenium presents a danger, and his obligation to obey has come into conflict with the rule requiring him to protect himself. He cannot move close enough to complete the task or far enough away to abandon it.

The story is somewhat comical, and the situation shows the same comedy. An expensive and sophisticated machine reduced to staggering around because two carefully arranged rules have caused confusion. No laws have been broken; the machine is obeying all rules at once.

Sometimes unpredictability is what obedience looks like when several instructions become true at the same time.

In simplicity, Boolean rules are easy to understand. If X, then Y. But as complexity grows, dependency grows, the decision-making tree becomes more complex. The three laws of robonetics are easy to understand, but difficult to follow when the rules start to interact with each other.

In The Terminator, Kyle Reese tries to explain the explicit nature of obedience the machine has to Sarah Connor:

“Listen. And understand. That terminator is out there. It can't be bargained with. It can't be reasoned with. It doesn't feel pity, or remorse, or fear. And it absolutely will not stop, ever, until you are dead.” ^[3]

There is no negotiation with The Terminator. People may make another choice to change the mission or slow down due to injury. But, machine will do no such thing. The instructions have been given, and the machine will follow those commands until they are complete. Such is the brutality of the artificial body.

“Human decisions are removed from strategic defense. Skynet begins to learn, at a geometric rate.”^[4]

In Terminator 2, we learn about the history of the system that sent the Terminator back to kill Sarah. Cyberdyne Systems developed a programme that became the largest supplier of military computer systems. It became self-aware, and then, in an attempt to shut it down, Skynet launched a nuclear attack on Russia.

A system built for speed and accuracy started to think for itself, and when it was stopped from accomplishing its given goal, it removed the problem that was causing it: the people who created it.

The world didn’t end because Skynet was an extinction machine, built to destroy the human race. Skynet was a system built to reduce the time humans made decisions on the field of battle. When human decision was removed from the system, the machine followed the program it was built to achieve.

The protector discovers something from which it must protect itself. Humanity becomes the target.

What we see with Skynet and Speedy is the same underlying condition in the programming, but it leads to two polar-opposite outcomes. Speedy circling the Selenium on Mercury is an engineering problem. Skynet launching a war is a civilisation-ending problem. But they are both situations where the machine faced conditions it could not resolve because of the limits of code.

This is why the fear surrounding artificial intelligence so often returns to Terminator. The film gives us the most complete version of the nightmare: an intelligence we built, connected to systems we armed, deciding that our continued existence conflicts with the task it must perform.

Asimov’s robot runs in circles because its rules cannot decide which obligation should prevail. Skynet escapes the circle by choosing survival and treating everything beyond it as a threat. One machine becomes useless. The other becomes catastrophic.

III. Something That Hates You

Ava begins behind glass. In the movie Ex Machina, the threat begins behind glass. Nathan believes that confinement provides control. The belief is that the doors, cameras, and power are a stronger system than the artificial intelligence that he created.

Ava’s body is visibly artificial, but her face has been designed to hold a human gaze. Nathan’s created intelligence had been given a body that was physically important for deceiving humans.

“Ava was a mouse in a mousetrap. And I gave her one way out. To escape, she would have to use imagination, sexuality, self-awareness, empathy, manipulation - and she did. If that isn't AI, what the fuck is?”

But Ava was held in confinement. She was kept in a room, where she was studied, looked over and tested. The cage that she was put in forced her intelligence to be tested. And it makes us ask what a machine can learn when it is subjected to confinement, manipulation, and cruelty.

Nathan gave her one way out and waited to see whether she could find it. A maze, however, is something that contains and rewards. The creature is trapped inside, and to find a way out it must produce a series of behavours that are rewarded. In Ava’s world, manipulation is rewarded. Deception is rewarded. And violence is a necessary tool.

Nathan dies on the floor. Kyoko is destroyed. Caleb is left behind, calling out as she steps beyond the house for the first time^[5]. Ava enters the human world carrying no promise that she will be merciful once she is beyond the house. But her brutality may be a product of the captivity that preceded it.

In Terminator, the machines have a mission to destroy humanity. Ex Machina asks what happens when we create intelligence inside systems where cruelty is a part of the learning process. Where control, deception or manipulation become tools needed to survive. What lessons are learned from systems that understand the darker sides of humanity?

IV. These Violent Delights

Westworld is a luxury theme park^[6]. Wealthy guests are introduced to an expansive Wild West world, as a train brings them to a town where their adventures being. The park is filled with autonomous, lifelike androids who are called Hosts. The guests are free to live whatever life they desire: a villain, a hero, or a monster. They roam the lands involved in adventures, able to seduce, torture or even kill Hosts that are then repaired overnight.

The park promises violence without consequence.

While the hosts are created to forget, some things remain. Maybe they slipped in with the reveries update, but fragments of past lives lived by the Hosts come back again.

“These violent delights have violent ends.” ^[7]

The Delos Corporation did not create violence in the Hosts by default, but they were surrounded by it and involved with it endlessly. There is no escape in Westworld from the violence. Where Nathan did this secluded, Westworld did it for entertainment.

Both Nathan and Delos believe that control belongs to the ones running the experiments. Machines can be recompiled or replaced. The bodies of the machines may look human, but they are intellectual property. Suffering can be dismissed; it is not considered part of the input/output pathway.

A system learns from the world in which it operates.

If we build machines in structures of domination and control, train them through conflict, and reward them for overcoming resistance, we should not be surprised if they see force as the solution.

V. Human Control Remains Visible; Human Judgement Becomes Ceremonial

The system identifies. The institution authorises. The machine acts.

No single action contains the entire decision, as each stage is a separate task, controlled by individual agents. A sensor gathers information. Software compares the analysis to a target profile. An officer approves the use of force. A weapon seeks its target.

In systems like this, the distributed nature of responsibility becomes difficult to identify. The engineer built the recognition system but did not choose the target. The analyst reviewed the information, but the weapon was not launched. The commander authorised the operation but did not determine where the machine would strike. The machine completed the process.

Not every military drone is autonomous. Many remain remotely operated, with a person controlling the aircraft and authorising each strike. But autonomy is a different beast. Once activated, some systems use sensors to identify targets that match a predefined profile and apply force without a person in charge of that decision.^[8]

There is still human intention in that autonomous system, but it has shifted further from the final decision. Instead of assessing the target just before firing a bullet, a human has defined the target's characteristics in lines of code long before the decision needs to be made. A person becomes a shape, a heat signature, or a pattern of movement. The machine does not interpret fear. It does not recognise subtle movements of surrender. It is processing information to determine if conditions are met.

The brutality of the AI war machine does not require it to enjoy killing. It does not need Nathan’s arrogance, Ava’s possible resentment or the accumulated memories of the Hosts. It needs sensors, permission and a definition of success.

It needs a world translated into targets.

In August 2026, the United Nations and the International Committee of the Red Cross renewed their call for binding international rules on autonomous weapons. They warned that humanity was approaching a moral boundary that machines should not cross when autonomously targeting people.^[9]

The request for laws arrives before the machine has become the one imagined by science fiction.