The story of smong is about knowledge surviving through time. On the Indonesian island of Simeulue, the memory of a devastating 1907 tsunami was passed from one generation to another as a simple lesson: when the earth shakes and the sea suddenly withdraws, go to high ground. Nearly a century later, that knowledge saved lives during the 2004 Indian Ocean tsunami. It is an extraordinary example of a society successfully preserving what it had learned.
But there is another problem with human knowledge that may be just as important. Sometimes we have not forgotten the lesson. We possess enormous amounts of relevant knowledge, but it is scattered among different people, disciplines, governments and technologies. The problem is not remembering it. The problem is assembling it and getting it to the person who needs it before it is too late.
The recent catastrophic glacier collapse and flood in Nepal provides a troubling example.
Scientists have known for years that a warming Himalaya is changing glaciers, glacial lakes, permafrost and mountain slopes. Governments know that mountain communities face growing risks from floods, landslides and glacial events. Satellites watch the mountains. Seismographs detect movements of the earth. Meteorologists watch storms and hydrologists watch rivers. Nepal and China had even been discussing glacier hazards and better sharing of information before the disaster.
None of that means somebody knew that this particular glacier would collapse on a particular morning. That distinction is important. The event appears to have been different from the more familiar glacial-lake outburst floods for which authorities had prepared. A mass of glacier, rock and ice suddenly failed, creating a devastating cascade downstream.
It would therefore be unfair to say that experts predicted the disaster and somebody simply ignored them. They understood the class of danger without being able to predict the particular event.
But something changes the moment the glacier collapses. The unknowable future becomes an observable present.
The mountain moves. Seismic instruments register something. People upstream see unusual conditions. River levels change. Water and debris begin moving downstream. Different people and different systems start acquiring pieces of the same story.
The question is whether those pieces become knowledge quickly enough.
At one Nepalese school, apparently, they did. The headmaster received an informal warning through someone with a relative upstream. He rang the emergency bell and evacuated more than 900 students and staff. They had only minutes. The school was subsequently destroyed. The children were safe.
There is something almost startling about the contrast. Humanity possesses satellites, supercomputers, seismic networks, climate models and sophisticated government agencies, yet the crucial information travelled through one human being to another and ended with someone ringing a bell.
That is not an argument against technology. It is an argument for thinking more carefully about what technology is for.
We tend to measure our capacity to deal with problems by how much we know. Perhaps we should pay more attention to whether knowledge can move.
Modern knowledge is extraordinarily fragmented. Geologists understand mountains. Meteorologists understand weather. Hydrologists understand rivers. Emergency officials understand evacuation. Governments control information within their borders. Local people know things about their surroundings that may appear in no official database.
Nature recognizes none of these divisions. The mountain collapses into the river during the storm and the resulting flood crosses an international border.
This may be one of the most important opportunities presented by artificial intelligence.
AI does not have to predict precisely which glacier will collapse next Tuesday to be useful. Its greater value may lie in recognizing when pieces of information that normally live in different places suddenly belong together. Unusual glacier movement, heavy rainfall, a seismic event, an unexpected river reading, a satellite observation and an eyewitness report may individually be inconclusive. Together they may mean something quite different.
At some point the sensible response becomes not “we know what is happening” but “we know enough to warn people.”
AI could help us detect convergence earlier. But it cannot install the river gauge, compel governments to share information, maintain communications during a disaster, create an evacuation route or make someone trust the warning. Nor should we allow increasingly sophisticated technology to crowd out local knowledge. The person standing beside the river may know something the satellite does not.
There is also an awkward political problem with prevention. When a warning fails, we count the dead. When a warning succeeds, there may be nothing to count. People evacuate, the disaster occurs, everyone survives, and afterward someone asks whether the alarm was really necessary.
Successful prevention can look remarkably like nothing happened.
That is why the lesson from Nepal goes beyond glaciers. We need institutions prepared to connect knowledge rather than merely accumulate it, to share information rather than possess it, and occasionally to act before every uncertainty has disappeared. AI can help enormously, but only if we design human systems capable of receiving what it discovers and doing something with it.
The people of Simeulue faced one great problem of knowledge: Could an essential lesson survive for a century? The people downstream in Nepal faced another: Could an essential warning travel in minutes?
Our first task is to remember what humanity has learned. Our second is to connect what humanity already knows. Because knowledge that arrives after the disaster is history. Knowledge that arrives before it can mean survival.