Pattern recognition: when the human mind overcomes machines
If I asked you which one performs better, AI or the human mind, you would likely answer: the first! And, indeed, that would be quite correct.
If we consider the ability to recognize well-defined and recurrent structures in the observed objects (pattern recognition), it turns out that our mind has a performance far better than any machine. At least, so far. For the sake of completeness, this ability of humans is also the cause of many problems, as it often leads to identify patterns or links between objects that actually do not exist (and, yes, this is an example of cognitive biases). Anyway, someone has thought to apply this human ability to the study of one of the most challenging scientific problems: earthquakes.
When an earthquake occurs, the induced ground motion is recorded at several measuring stations. The track at a given station, which can be thought of as the footprint of the seismic waves passage, is a ground motion time series. Now, here is the magic: this waveform is converted into audible sound, so that seismic data can literally be heard. Such transformation is usually achieved by time-compression: by doing so, the recorded signal is shifted from the seismic to the audible frequency range.
“The human auditory system can perceive a great deal of subtlety in seismic data, including direction, earthquake source, and also the physical nature of the crust near the seismometer”, said geophysicist Ben Holtzman, director of Seismic Sound Lab.
The goal is to use the pattern recognition human abilities to distinguish complex features in seismic data, such as earthquake type, triggering mechanisms, subsurface properties. This could complement more traditional data tools or, hopefully, it could lead to new insights into the understanding of earthquake physics. According to some recent studies (e.g. 1, 2), human listeners can successfully recognize similar features in sonified seismic signals and link them to specific geophysical parameters, if trained. In other words, supervised learning of the human machine seems to work very well.
The sonified seismogram below is related to a tremor triggered in Parkfield (California) by the M7.8 Denali (Alaska) earthquake:
You should have been able to discriminate between the earthquake (the bang at the beginning) and the triggered tremor (which reminds us of a rattlesnake). These dissimilarities reflect different seismic mechanisms: the bang is related to a big earthquake, clearly distinguishable by the subsequent events and occurred over a very short time interval; the rattlesnake-like sound is related to a tectonic tremor along the San Andreas Fault, consisting of several low-frequency earthquakes, all of comparable intensity, and covering a longer time.
At this point, you may ask why we do not use the pattern recognition capabilities of the human visual system as well. That’s exactly what Ben Holtzman and his team have done, by representing the seismic signal through sound and animated images, leading to an audio-visual representation of earthquake data.
Here is one of the most striking results of this team: the conversion of the seismic waves generated by the 2011 Tohoku earthquake into sound waves and light patterns:
By visualizing the numerical simulation of the seismic waves induced by an earthquake, we can identify the main features of the seismic wavefield and make inferences about the characteristics of the Earth internal structure (seismic wave propagation is influenced by the properties of the medium).
Science is always an exciting journey: you know where you start from, but you don’t know where it will take you. Maybe the audio-visual representation of seismic data will not bring new insights about the physics of the earthquake process, but it has already opened unexpected doors: as a matter of fact, it has inspired several artists (Florian Dombois is a great example) and has been successfully used for public outreach purposes (like for the SeismoDome installment). After all, creativity is one of the traits that makes human mind different from any other machine.
***
Interesting links:
- SeisSound → IRIS (@IRIS_EPO) Audio/Video Seismic Waveform Visualization
- Earthquake Music → Seismic recordings from all around the world converted to sounds. More details here: Listen, Watch, Learn: SeisSound Video Products
- SeismicSounds → Sonified seismograms samples from Sounds of Seismic, an art-science project.
- Mt. Etna Volcano Sonification → Volcanoes trigger seismic waves as well, we can listen to them too! A project by @INFN_ and INGV.
