The Oldest Analogy in the Newest Conversation

Fire is perhaps the most ancient example of transformative technology. Prehistoric humans who controlled it could cook food, ward off predators, survive cold, and signal across distances. Those who lacked it were at a permanent disadvantage. And yet fire also burned settlements, spread disease through smoke, and was used as a weapon in every era of human conflict. Nobody argues we should have left fire alone. The conversation we have had about fire for thousands of years is not whether to use it but how to use it well — how to build fireplaces and chimneys, how to teach children not to touch it, how to fight it when it escapes. That same conversation, in a more complex form, is the one we need to be having about artificial intelligence right now.

The Tool Does Not Choose

There is a category error that runs through a lot of AI commentary. The technology gets credited or blamed for outcomes that actually result from human decisions. Algorithms recommend content — but humans built the recommendation systems and chose the metrics they optimize for. AI can generate convincing false information — but humans decide whether to deploy it for disinformation campaigns. The tool does not choose. The people who build it, deploy it, and use it choose. This matters because it locates responsibility accurately. When a social media platform uses AI to maximize engagement and the result is increased political radicalization, the cause is not the AI. The cause is the choice to use engagement as the primary metric, made by executives who understood the downstream effects well enough to be concerned about them in internal documents. Research from MIT Media Lab on algorithmic accountability has examined how framing affects public and regulatory responses to technology harms. When harms are attributed to the technology itself, the response tends to be reactive and technical — build a better filter, add a warning label. When harms are attributed to design choices, the response tends to be structural — change the incentive, require different metrics, hold decision-makers accountable.

Creation and Destruction Live on the Same Spectrum

The applications that make AI genuinely useful — accelerating drug discovery, detecting cancer in early imaging, modeling climate systems, translating across languages, making education accessible to people without teachers — use the same underlying capabilities that make it dangerous. The ability to generate realistic text, images, and video at scale can serve education or serve propaganda. The ability to model complex biological systems can speed vaccine development or inform bioweapons design. This is not unique to AI. The nuclear physics that powers cities is the same physics that destroys them. The chemistry that makes fertilizer also makes explosives. The dual-use problem is ancient. What is new is the speed at which AI capabilities diffuse and the number of people who have access to them.

The Tangent That Illuminates the Point

There is a small but important body of research on how people relate to tools that have caused them harm. Psychologists studying trauma associated with car accidents have found something interesting: people who experience serious accidents rarely develop a lasting fear of cars. They develop a more nuanced relationship with driving — more alert, more aware of conditions, more deliberate about risk. The tool remains useful; the relationship with it matures. This is possibly the healthiest model for a society-wide relationship with AI. Not fear, not naive enthusiasm, but a relationship that has been tested by real consequences and emerged more clear-eyed.

Who Bears the Cost When It Destroys

The destructive applications of any technology rarely harm the people who benefit most from its creative applications. The people whose jobs are displaced by AI-driven automation are not generally the investors who profit from the productivity gains. The communities flooded with AI-generated disinformation are not the companies whose advertising models depend on engagement. Research from the Economic Policy Institute on the distributional effects of automation has documented a consistent pattern: the gains from labor-displacing technology concentrate among capital owners and highly skilled workers, while the costs spread across communities that lose employment bases without comparable alternatives arising to replace them. This is a political and economic problem, not a technical one. The technology does not distribute its own consequences. Societies make choices, through policy or the absence of it, about who benefits and who bears costs.

The Choice Is Always Being Made

There is no neutral position. Choosing not to regulate is a choice. Choosing to deploy without safeguards is a choice. Choosing to optimize for profit without modeling second-order effects is a choice. The fire analogy does not offer comfort; it offers clarity. We have always lived with powerful tools that can create or destroy. What determines the outcome is not the tool — it is the accumulated weight of decisions made by people who could have chosen differently. That weight is being accumulated right now, with consequences that will extend well beyond the people making the decisions.