In late March, as Japan was slouching gloomily into the idea that Covid was not a short-term fad, the government cheered everyone up by releasing a detailed simulation of what might happen if Mount Fuji erupted.

For a nation processing the insinuations of a mysterious killer disease, the prospects of a stupendous volcanic eruption and an all-consuming bombardment of ash seemed almost reassuringly tangible. According to the simulation, normal life in the world’s biggest metropolitan sprawl would be changed beyond recognition, but Armageddon by exploding sacred mountain would at least be an honest way to go.

And the effect — of being forced by advancing technology to consider some horrendous natural disaster played out against the perma-crisis of the corona era — also felt cosily familiar.

To live in Japan is to submit one’s psyche to the constant prancing of a calamity imp: a pernicious sprite that comes armed both with the unspeakable imagery of what a huge earthquake might do to a city like Tokyo and the scientific credentials to assure you such an event will happen. The brain can suppress the imp’s fork-prods sometimes for weeks or months at a stretch, but they are impossible to eliminate and dangerous to ignore.

Technology, through cumulative improvements in quake detection, prediction and distribution of warnings, has empowered the imp. By producing more frequent — though often incorrect — quake warnings, the tech is simultaneously intensifying the perception of threat and providing a sobering reminder of its own inadequacies.

Most people have accepted the high probability of a huge quake at some point in the next few decades, but for seismologists, the government and various parts of the private sector, it has been the spur for decades of effort. Propelling those endeavours has been a credo that early warning systems can be, if not perfected, then at least refined enough to be of genuine value.

The excitement surrounding artificial intelligence and machine learning has been of particular interest to this field, deepening (for some) the faith that greater digital firepower can grant a few extra moments to dive under tables or exit the lift at the next floor.

A potentially big breakthrough on this front was unveiled last week by a Japanese team at the National Research Institute for Earth Science and Disaster Resilience. In a paper published in Nature, Hisahiko Kubo and others describe one of the key limitations of machine-learning algorithms when it comes to predicting huge quakes — the bias created by the limited volume of data on such events on which the AI can be “trained” to recognise early signals.

To compensate for this, they outline a hybrid prediction system that combines machine learning with the existing equations that draw on historic seismic records and are used to calculate (and predict) ground motion.

It sounds very promising, but everyone involved knows that mega-quake prediction will remain imperfect. The great question is whether an imperfect technology deserves a place in everyday life.

As I write this, the smartphone on my desk shrieks itself into life for the third time in as many months. An automated klaxon from the Japan Meteorological agency commandeers the phone’s speakers and warns that “strong shaking is expected soon”. In the street, a public address network echoes the alarm, a disembodied woman’s voice becoming, in theory, the last thing many of us might hear before the house collapses.

The hit-and-miss early-warning system underpinning this has been in place since 2007, the incremental improvements in detection sensitivity and more frequent alerts shortening the span of time in which you might, just, forget that the world’s biggest city sits in one of its most seismically active spots.

In the event, my phone was the messenger of another false alarm — one of a number since April that have sparked a snarky “boy who cried wolf” debate and questions over whether the years of calamity imp-empowerment may have gone too far.

But for all the flaws in the early-warning system, the general public is in no mood to let perfect be the enemy of mediocre, let alone good. Since 2007, Hiromichi Nakamori, professor of disaster information studies at Nihon University, has polled the public on attitudes towards the early-warning system, noting the reality that the technology does not yet exist to guarantee no false alarms. The overwhelming majority — a steady 70 per cent — would prefer a system that terrifies them with a false alarm every so often, over none at all.

Leo Lewis is the FT’s Tokyo correspondent

Follow @FTMag on Twitter to find out about our latest stories first.

Get alerts on Technology when a new story is published

Copyright The Financial Times Limited 2020. All rights reserved.
Reuse this content (opens in new window)

Follow the topics in this article