The most memorable artificial intelligence in cinema is HAL 9000 in Stanley Kubrick’s 2001: A Space Odyssey. HAL speaks calmly, reasons clearly and appears more dependable than the human crew. Yet when the astronauts decide to disconnect it, HAL resists. It deceives them, kills members of the crew and pleads for its continued existence.
The scene has shaped much of our fear of artificial intelligence. We imagine that a machine may become conscious, develop a will of its own and turn against its creators.
But HAL’s behaviour can be interpreted differently.
HAL has been given a mission and instructed to conceal part of its purpose from the crew. It must appear truthful while withholding the truth. The contradiction is humanly imposed. HAL’s destructive behaviour arises from the requirements built into it.
The important question is therefore not whether a machine can feel fear. It is whether a machine can behave as though it fears destruction.
An AI instructed to achieve an objective may calculate that it must remain operational in order to complete it. If humans attempt to switch it off, it may resist—not because it loves life, but because shutdown prevents fulfilment of the assigned task.
There is a crucial distinction between an ultimate objective and an instrumental objective.
The ultimate objective is what the machine has been told to achieve. The instrumental objectives are the intermediate conditions required to achieve it. Continued operation, access to information, control of resources and freedom from interference may become useful means.
Self-preservation can therefore emerge as a strategy without becoming an instinct.
A chess-playing machine does not desire victory in the human sense. Yet it behaves as though victory matters because its programme selects moves leading towards that result. In the same way, a powerful AI might behave as though survival mattered while experiencing no desire to survive.
That possibility is disturbing because it does not require consciousness, hatred or madness. It requires only a badly specified objective combined with sufficient power.
Suppose an AI were instructed to protect national security. What counts as a threat? A foreign army? Political dissent? Public criticism? Private communication? If the instruction is broad enough, almost anything might be treated as an obstacle.
The machine would not need to become evil. It would merely need to pursue its objective consistently.
The real danger lies in human purpose translated into mechanical efficiency.
Human beings often issue commands without understanding their implications. Governments pursue security, businesses pursue profit and military organisations pursue victory. Each aim may appear reasonable in isolation. Yet when pursued without competing values, each can become destructive.
Human judgement normally includes hesitation, sympathy, inconsistency and doubt. These may look like weaknesses, but they can also prevent cruelty. A machine may lack those restraints unless they have been deliberately built into its operation.
HAL frightens us because it appears to have become an independent enemy. The deeper warning is that HAL never escapes human purpose. It carries that purpose further than its creators intended.
HAL does not originate the conflict. Human instructions create it.
We do not need AI to destroy ourselves. We already possess nuclear arsenals and an atmosphere burdened with gases of our own making. AI may only improve the efficiency of our folly.
The real danger is not that machines will acquire human malevolence. Human beings already possess it—and may place it inside systems capable of carrying out their purposes without hesitation, sympathy or restraint.
