[ExI] AI finds Dyson sphere candidates

Stuart LaForge avant at sollegro.com
Thu May 30 02:07:24 UTC 2024

A recent paper reported on a project called Hephaistos that used a 
convolutional neural network (CNN) to sift through the spectral data of 
5 million stars in the Gaia satellite database for stars with very high 
infrared emissions relative to visible light. The AI found 7 candidates 
for closer study.


The search for extraterrestrial intelligence is currently being pursued 
using multiple techniques and in different wavelength bands. Dyson 
spheres, megastructures that could be constructed by advanced 
civilizations to harness the radiation energy of their host stars, 
represent a potential technosignature, that in principle may be hiding 
in public data already collected as part of large astronomical surveys. 
In this study, we present a comprehensive search for partial Dyson 
spheres by analysing optical and infrared observations from Gaia, 2MASS, 
and WISE. We develop a pipeline that employs multiple filters to 
identify potential candidates and reject interlopers in a sample of five 
million objects, which incorporates a convolutional neural network to 
help identify confusion in WISE data. Finally, the pipeline identifies 
seven candidates deserving of further analysis. All of these objects are 
M-dwarfs, for which astrophysical phenomena cannot easily account for 
the observed infrared excess emission.

I wonder if there is any overlap between these candidates and Schmidt's 
cluster of Boyajian-like dimming stars. I hope these guys get time on 
the JWST to take a closer look.

Stuart LaForge

More information about the extropy-chat mailing list