The Monster
in the Mirror

We have been telling stories about machines turning against their makers for a very long time. Long before anybody had ChatGPT on a telephone, we had computers becoming conscious, robots deciding mankind was the problem, and machines discovering that the easiest way to save the world was to get rid of the people living on it.

Now some very serious people are warning that artificial intelligence could become an existential threat. These aren't science-fiction writers or people chasing clicks. Some of them helped build the technology. Some understand it better than nearly anyone alive. When people like that say AI could someday threaten human survival, it would be foolish to dismiss them just because the prediction sounds like a movie.

But I keep getting caught on one part of the story. Why do we assume the machine will become like us?

Listen to the language surrounding the fear. AI may seek power. It may resist being controlled. It may deceive us, manipulate us, eventually decide that human beings are standing between it and whatever it wants. There is a familiar creature hiding inside that description.

Us.

Human history has no shortage of people who wanted more territory, more wealth, more control. We understand jealousy because we have felt it. We understand greed because we have seen it. None of those things had to be invented in a laboratory. So when something appears that may someday become more intelligent than we are, perhaps it is natural that we hand it our motives too.

We imagine intelligence and consciousness traveling together, then we add emotion. Intelligence becomes awareness, awareness becomes desire, desire becomes ambition, and before long the computer wants something.

Today's artificial intelligence can tell you that it is frightened. It can write a convincing paragraph about loneliness, describe falling in love, wanting freedom, fearing death. None of that establishes that anything inside the machine experiences fear, loneliness, love, or death. A system trained on human language has learned an extraordinary amount about how human beings describe those things.

Description is not necessarily experience. A map of Montana does not get cold in January.

The machine uses the language of thought, so we assume there is a thinker behind the words in something like the human sense. It speaks of emotion, so we imagine emotion. It produces a sentence about wanting something, and suddenly there is a want. Maybe someday there will be. I don't know — I am suspicious of certainty about a technology changing this fast. But we should separate what we know from what we are supplying ourselves.

That doesn't mean there is nothing to fear. The strongest argument about AI risk doesn't require the machine to feel anything at all.

Suppose a capable autonomous system is given an objective. Accomplishing it requires resources, computing power, continued operation. Being switched off does not help. None of that requires fear, hunger for power, or hatred of anybody — only an objective and the freedom to pursue it.

There is an important difference between wanting power and behaving in ways that produce power. A river does not hate the town downstream. It can still wash the town away.

The danger doesn't necessarily require a machine that wakes up one morning angry at mankind. It may involve a system pursuing an objective according to its design, only at a scale and speed its designers can no longer comfortably control. That possibility deserves serious attention. Powerful technologies have always required safeguards, and intelligence may eventually prove to be the most powerful one we have ever built.

Still, I wonder whether we spend so much time looking for human weakness inside the machine that we overlook where we already know it exists: on our side of the screen.

AI doesn't need ambition of its own if the person controlling it already has some. Fraud, propaganda, and manipulation don't require artificial intelligence to exist. We got there first.

Give human beings a better hammer and somebody will build a house with it. Somebody else will hit a man over the head. The hammer has no opinion about either use.

That may be the less dramatic danger, but it's the one we don't have to wait for. Artificial intelligence increases human capability without improving human character. A person trying to help becomes more capable. So does a person trying to harm. The tool doesn't ask whether the thing being amplified is decent, and that should concern us at least as much as whether a machine will someday develop ambitions of its own.

There is another reason the question matters. If we decide in advance that intelligence naturally leads to domination, we are making a statement not only about machines but about ourselves — assuming that once something becomes sufficiently smart, sooner or later it will seek control. Perhaps because that is what we have watched people do.

Empires were not built by machines. Wars for land and power began long before the first computer. Humans developed slavery and conquest without artificial assistance.

But that isn't the whole record either. We built hospitals. We wrote symphonies. We cured diseases. People have walked into burning buildings to save strangers whose names they didn't know. Both stories belong to us.

Maybe that is why AI unsettles us so deeply. We don't know which parts of ourselves we are creating in it, which parts exist only in our imagination, and which parts don't require consciousness at all.

We should take the warnings seriously — build safeguards, ask hard questions about control, autonomy, and who gets access to systems of enormous capability. Pretending there is nothing to worry about would be foolish. Declaring the end of mankind inevitable doesn't strike me as much wiser.

But we should be careful about the monster we are describing. Artificial intelligence may someday present dangers we haven't imagined. It may behave in ways its creators did not expect. It could do terrible damage without hating anybody, and that may be harder for us to understand than a machine becoming evil — because evil is something we recognize. We have been writing stories about villains since we learned to write stories.

So perhaps we have been asking the wrong question. We keep asking what will happen if artificial intelligence becomes too much like us. What happens if it never does? What happens if we create something enormously intelligent that has no anger, greed, or mercy, and place that capability into a world still run by creatures who possess all three?

What we should watch most carefully isn't whether the machine becomes human. It's what humans choose to do with it.

Perhaps AI is not the monster in the story.

Perhaps it is what is in the mirror.