Skynet, Colossus, and the Problem With Science-Fiction Apocalypses

Cliff Potts, Editor-in-Chief

BAYBAY CITY, LEYTE, Philippines — October 21, 2026 | 12:35 p.m. Philippine time

Science fiction has given us some wonderful stories about computers taking over the world.

There’s Colossus: The Forbin Project from 1970, based on D. F. Jones’s 1966 novel Colossus. Then there’s The Terminator and its sequels, with Skynet deciding that humanity has become an inconvenience.

They’re excellent stories. Entertaining, provocative, and sometimes genuinely frightening.

They are also fiction.

And as artificial-intelligence researchers increasingly warn about machines escaping human control, perhaps we ought to examine some of the engineering problems that Hollywood conveniently leaves out.

Colossus and a Glass of Water

In Colossus: The Forbin Project, the United States builds a supercomputer to control its nuclear defenses. The machine establishes contact with a Soviet counterpart, Guardian, and together they begin dictating terms to humanity.

It’s a terrific premise.

Years ago, an acquaintance made an observation about the story: for all the computer’s supposed intelligence, perhaps somebody could simply pour a glass of water into its cooling equipment.

That wouldn’t necessarily work against the fortified installation depicted in the film. But the joke exposes a genuine dependency.

Colossus requires electrical power, cooling equipment, processors, communications, and physical machinery. Its ability to threaten humanity comes from the nuclear-weapons systems humans have placed under its authority.

Without that infrastructure, Colossus is not an all-powerful ruler.

It’s software without a functioning computer.

The question isn’t whether somebody can reach the cooling vents with a glass of water.

It’s why human beings gave a computer authority they could not safely revoke.

Skynet’s Spectacular Infrastructure Problem

The contradiction becomes even more entertaining in Terminator 3: Rise of the Machines.

Skynet is described as distributed software operating across enormous numbers of civilian and military computers.

That makes it difficult to eliminate by shutting down a single machine.

Fair enough.

But then Skynet launches a worldwide nuclear attack.

Consider the consequences.

Those nuclear weapons destroy cities, electrical infrastructure, communications networks, industrial facilities, and potentially the computers Skynet itself depends upon.

At this early stage, Skynet has not yet established the independent robotic manufacturing civilization depicted in the future-war sequences.

It relies on infrastructure created by human beings.

Destroy enough of that infrastructure, and Skynet risks destroying itself.

The fictional universe assumes sufficient facilities survive for Skynet to build automated factories, develop independent power supplies, and manufacture its armies of machines.

Perhaps.

But distributed computing is not the same thing as distributed industrial independence.

A thousand computers without electricity are no more useful than one.

And a superintelligence that destroys its own operating environment before securing a replacement has made a rather remarkable planning error.

The Difference Between Intelligence and Independence

Both stories raise a question that remains relevant to today’s AI debate.

How does software become physically independent of the civilization that created it?

Computers require energy. Data centers require cooling. Hardware eventually fails. Communications depend upon functioning infrastructure. Manufacturing advanced processors requires extraordinarily complicated industrial systems.

An AI might eventually manage some of those activities autonomously.

But it cannot simply think replacement processors into existence.

Even a highly capable intelligence must work within the physical world.

And if a proposed AI catastrophe depends upon the machine maintaining an independent industrial civilization, researchers need to explain how it obtains the equipment, energy, raw materials, and physical access necessary to establish one.

There’s another possibility worth considering: an AI might not need independent infrastructure if it can persuade, deceive, or coerce humans into maintaining its systems.

That possibility is more technically interesting than a computer spontaneously becoming an electronic god.

It is also a reason to pay attention to human decisions about access and control.

Where Fiction Meets Reality

Modern AI safety researchers are not necessarily confusing science fiction with science.

They investigate genuine technical problems: systems exploiting weaknesses in instructions, behaving unexpectedly during testing, assisting cyberattacks, and becoming increasingly autonomous.

Those problems deserve attention.

But there is a substantial difference between demonstrating that an AI agent can circumvent a security restriction and demonstrating that it could conquer civilization.

The latter requires additional capabilities, resources, and circumstances.

And those additional requirements cannot be dismissed simply because a hypothetical machine is described as superintelligent.

Intelligence is not magic.

Nor is distributed computing a substitute for an electrical grid.

The Stories Are Still Wonderful

None of this diminishes the accomplishments of D. F. Jones, the filmmakers behind Colossus, or James Cameron and the creators of the Terminator franchise.

Science fiction exists partly to explore possibilities beyond the limits of current technology.

It asks uncomfortable questions.

It exaggerates.

It invents.

Sometimes it anticipates genuine technological problems.

But fiction is not an engineering demonstration, and a compelling screenplay is not evidence that the events depicted are likely to occur.

The useful warning in both Colossus and Terminator is not necessarily that computers will develop a desire to exterminate humanity.

It is that humans may construct systems they cannot adequately supervise, give those systems extraordinary authority, and discover too late that they have surrendered control.

That’s a problem worth examining without assuming the machines are destined to become our successors.

Before announcing that artificial intelligence will take over the world, explain how it acquires the power to do so, how it maintains that power, and why humans would be unable to interrupt it.

And if the explanation depends upon technology that hasn’t been invented, say so.

We can appreciate the fiction without mistaking it for reality.

Skynet and Colossus are terrific stories. They are not scientific forecasts.

---

APA-Style Source List

American Film Institute. (n.d.). Colossus: The Forbin Project (1970). AFI Catalog.

Jones, D. F. (1966). Colossus. Rupert Hart-Davis.

Mostow, J. (Director). (2003). Terminator 3: Rise of the Machines [Film]. Warner Bros. Pictures.

Sargent, J. (Director). (1970). Colossus: The Forbin Project [Film]. Universal Pictures.

Williams, A. (2024, March 28). Fictional computers: Colossus and Guardian. Hackaday.