Planetary SETI--It's all about (geological) context
Planetary technosignatures are an innovative way to conduct SETI research that enables us to look in our own solar system for non-terrestrial artifacts. Considering the greatness interstellar distances, it is unlikely that we will find non-terrestrial technology in the confines of our solar system, unless it comes from a civilization that evolved on Earth, or perhaps ancient wet Mars or pre-greenhouse Venus, in the geologic past. Nonetheless, considering the nearness of solar system objects, it is relatively easy to search for planetary signatures. Unlike exoplanets or distant star systems, we can observe objects in our own solar system with relatively high resolution and obtain data that is both greater in volume and quality. also we will not know unless we conduct an actual search.
Conducting such a search leads to a significant issue, however. That issue is how do we know what a planetary technosignature looks like? One way around this issue is to use an unsupervised machine learning algorithm (such as an autoencoder) to reconstruct natural geology on planetary surfaces and flag what it can’t reconstruct as an anomaly for further analysis. Models could be developed for the unique geology of the Moon, Mars, and icy moons to detect potential non-terrestrial artifacts, but also evidence of weird undiscovered planetary phenomena that advances planetary exploration and resources useful to future astronauts and space settlers.
Over 400 years ago, Johannes Kepler is reported to have thought that the circular craters on the moon were fortifications because he was unfamiliar with the process of impact cratering. This shows the propensity for humans to mistake orderly natural features for artificial structures.
On the other hand, it is also possible for us to misidentify artificial structures as natural features because we assume that everyone builds the same way that we do. This is part of the reason that the Nazca lines went undetected by European explorers…until they were viewed from the air. From the surface and on a small scale, the Nazca lines might look like natural trackways in the desert, but the appropriate scale reveals their artificiality.
Because of the assumptions involved, it is simpler to instead to consider what alien artificial structures do not look like. One way to do this is to train an unsupervised machine learning model (such as an autoencoder) on natural features that occur on planetary surfaces (craters, scarps, dunes, volcanic domes, etc.) so that it will recognize when something is not natural and flag it as an anomaly.
Supervised and unsupervised machine learning approaches have been proposed for the Moon for a variety of investigations related to planetary science and space resources, not just technosignatures. The main difference between supervised and unsupervised machine learning is that supervised machine learning algorithms are trained to identify specific features and also identify false positives of that feature. A supervised ML approach to planetary technosignatures might be to train an algorithm to distinguish between artificial and natural features based on a given definition. For example, features that circular with bright rims (craters) might be defined as natural features, whereas linear features high brightness might be defined as potential artificial structures. In other words, a supervised approach results in an “duck/not-duck” app for detecting aliens. While not necessarily the wrong approach, it is limited by assumptions about the nature and appearance of artificial structures which may not hold true for a non-terrestrial civilization.
An unsupervised learning approach to ML model building would allow for the model to focus on identifying natural geology until it can identify something that does not fit with what we know of natural geologic features. Once identified, the feature can be flagged and further analyzed to determine if it is likely to be of technological origin rather than just assuming it is based on pre-determined definitions of what is artificial in the search algorithm.
Although a variety of machine learning algorithms have been applied to the Moon, the approach is yet to be widely adopted by the planetary science and technosignatures communities. Furthermore, current proposals specifically for technosignature detection mostly focus on the Moon without extending to other bodies.
There is good reason for this since to do otherwise would add complexity to the project. If you are trying to see if an approach works, it is good to have a well-behaved control. The Moon is also a common choice because 1) its surface is very old and relatively unaltered, meaning a non-terrestrial artifact could remain on the surface for millions of years, increasing the probability of detection, and 2) the Moon’s proximity to Earth could make it a promising location from which to observe life on Earth, increasing the likelihood that a spacefaring civilization would take an interest in the Moon. Nonetheless, there are bodies beyond the Moon which could also have technosignatures, including Mars and possibly icy outer solar system moons.
As a planetary geologist with experience in quantitative modeling and geospatial data analysis, I am well positioned to develop an ML model (possibly an autoencoder) leveraging what we know of the geology of planetary surfaces to enable accurate anomaly detection that could aid technosignature searches. Such a model also has applications in broader planetary science research and resource detection.
I would start with training and developing an ML model for lunar geology. Once a model has been developed which can accurately identify anomalies on the Moon, the unsupervised ML framework could be applied to other planetary bodies. A logical next step would be the planet Mars. With the geological evidence for a more habitable phase in its geologic past, Mars could be considered the most likely place in the inner solar system after Earth to be a center for the formation of life. This argument is made more compelling by the recent evidence of a potential biosignature in rocks at Jezero Crater found using NASA’s Perseverance rover.
If Mars is the most likely place outside Earth in the inner solar system to have indigenous non-terrestrial life, it also follows that it would be the most likely place to find non-terrestrial intelligence. This is not to say that intelligent life is likely to have evolved on ancient Mars, but a planet with life is more likely to have intelligence evolve than a planet with no life from the start. Alternatively, the potential past habitability of Mars also makes it a likely place for non-indigenous intelligent life as a target for exploration, prospecting, or even colonization. This point is exemplified by aliens from Earth (i.e., Homo sapiens) that currently have plans to explore and possibly colonize the red planet.
Developing an unsupervised ML model to reconstruct Mars geology would not be drastically different from developing an unsupervised ML model for lunar geology. The main difference is that the terrain and range of geologic features on Mars is greater and more complex. The Moon’s geology has been largely shaped by impact events and volcanic eruptions. Most of the Moon’s geologic features are related to these two processes. Volcanic activity and impact events have played a major role in Mars’s history, but so has wind, running water, glaciers, ice sheets, and ground ice. This make the geology of Mars much more complex than the Moon, so it will be more difficult to train a model to identify anomalies on Mars just because of the greater diversity of features.
After the Moon and Mars, the next logical place in my opinion would be icy, ocean-bearing moons in the outer solar system like Jupiter’s moon Europa or Saturn’s moon Enceladus. Using the same logic I used for Mars, icy moons with potentially habitable subsurface oceans are the most likely places to host non-terrestrial life after Mars and are there also the most likely place to find non-terrestrial intelligence. This non-terrestrial intelligence could be indigenous the icy moon or take the form of off-world visitors or colonists.
An unsupervised ML approach to technosignature detection on the surface of Europa or Enceladus is particularly helpful. This is because of the highly speculative nature of of a technologically advanced civilization that developed in the subsurface ocean of an icy moon and what form its technology would take. For example, such a civilization would not be able to use fire (i.e., combustion), which is directly or indirectly the basis of much of our civilization’s advanced technology. The evolutionary relationship between human civilization and fire is explored in depth in Stephen J. Pyne’s book The Pyrocene.
This has implications. Would we expect the same type of technological structures that one might expect from an originally land-dwelling civilization that can use combustion-based technology? This is a fascinating problem but also one that is unlikely to be solved in the time limit imposed by the average research grant. Focusing on the natural geology and flagging structures that don’t fit natural patterns saves time and allows us to avoid the pitfalls of prior assumptions.
In some ways, the geology of outer solar system icy moons is similar to that of Earth’s Moon. Impact events and volcanism, albeit cryovolcanism, where the lava is liquid water instead of molten rock, have both played a role in the geologic history of icy outer solar system moons. In other ways, the geology of icy moons is wildly different. For example, tectonic processes, including faulting and rifting, have played a significant role in in the geological evolution of the surface of Europa and possibly other geologically active icy moons.
This makes Europa and similar icy moons less like the Moon or Mars geologically and more like Earth, where geology is dominated by plate tectonics, where the crust is divided between tectonic “plates” with most of the active geology on Earth (volcanic eruptions, earthquakes, etc.) happens at the plate margins where the plates are in contact with each other. The plate tectonics of Earth is unique in the solar system and not found on other planetary bodies in exact replication. Nonetheless, many parallels are found between tectonic processes on Earth and Europa, including plate subduction.
This means that the geologic features identified as natural could be different than those identified on the Moon or Mars. Thus, features that would be anomalies on the Moon or Mars would not be anomalies on Europa or Enceladus. Making a ML model trained on the geology of each body would therefore be a complex endeavor, but it would also expand our search, not just for technosignatures but also weird undiscovered planetary phenomena and resources (water, minerals, etc.) for future astronauts and space settlers.
Thank you for reading my Substack! If you like what you read, feel free to share or subscribe and thank you again if you already have.
If you would like to support independent planetary science research, you can become a paid subscriber and double thanks if you already have.