Artificial Intelligence

Marianne Karplus

A quiet room of computers
blinks inquisitively,
hums faithfully.
I type the last line of code
for this machine learning algorithm,
eager to detect new patterns.

My technological family
of earthquake chasers is
trained to distinguish
earthquakes from explosions
and the rumble of fuel tankers
down the highway.

I feed them seismograms,
spaghetti trails that represent
ground motion through time,
up and down,
north and south,
east and west.

The faithful processors
offer their best solutions,
but their supervised learning
encountered only California earthquakes,
and now we’re chasing
earthquakes in southern Mexico.

Different tectonic landscapes
tremble with diverse tones
and despite all our learning,
we are vulnerable to the inductive biases
of our assumptions,
machines and scientists alike.

The Science

This poem describes a seismology application of machine learning, which is a subset of artificial intelligence, to detect earthquake signals. Seismograms are recordings of ground motions generated by earthquakes, explosions, and many other sources. In this poem, the narrator is testing a machine learning algorithm to detect earthquakes in data from southern Mexico. However, the algorithm was trained using only earthquake data from California. The narrator realises that the algorithm is not detecting earthquakes well due to inductive bias.

The Poet

Marianne Karplus, Ph.D., is a seismologist and Assistant Professor of Geological Sciences at The University of Texas at El Paso. She is an enthusiastic student of the Earth who enjoys doing field work and making new discoveries in earthquake science. She also writes poetry, especially about nature and interactions between humans and the environment. Her poems have been published in the Rio Grande Review and the California Quarterly.

Next poem: Fraud Scheme by Akul Anandur