At the heart of the millions of square kilometers of the Amazon forest could hide thousands of little-known archaeological sites, witnesses of ancient civilizations whose traces go back up to 13,000 years. But the jungle, which spans nine countries and is home to multitudes of indigenous peoples, remains so impenetrable that it is almost impossible for archaeologists to physically track down every single remains. So where to start?
Faced with this immensity, researchers are increasingly relying on cutting-edge technologies such as artificial intelligence and machine learning, hoping that they will help them detect what they cannot see with the naked eye. It is in this spirit that OpenAI, the firm responsible for ChatGPT, joined forces with two archaeologists to judge a unique competition: it invited tech enthusiasts to examine satellite images and remote sensor data in search of unknown archaeological sites, explains an article from National Geographic.
The “OpenAI to Z Challenge” competition crowned a three-person team nicknamed “Black Bean,” which identified sixty-seven distinct areas across the Amazon that could be home to major ancient sites – and therefore serve as starting points for future excavations. The judges, including Sarah Parcak, Egyptologist, and Chris Fisher, archaeologist specializing in Mesoamerica, praised the rigor of the work.
To achieve these results, members of “Black Bean” trained deep learning models on public databases such as satellite images from Google Earth Engine, LiDAR data and digital models from NASA. Then, they used OpenAI’s GPT-4o model to compare the patterns of known sites with the still unexplored regions, with the chosen coordinates mainly concentrated in Brazil.
Interestingly, the majority of areas identified by the AI seem to cluster around bodies of water. “Our results are in fact logical, at a level that is even common sense”explains Yao Zhao, team member and software engineer at Meta. “Ancient civilizations, after all, tended to thrive near accessible water sources.”
Preserve heritage
The ability to scan millions of square kilometers of geographic data in a few weeks could greatly facilitate the work of archaeologists, allowing them to identify recurring patterns before even carrying out any field work, says Yao Zhao.
Machine learning and imaging tools are not new to some explorers of these ancient times. Sarah Parcak has used LiDAR, satellite and thermal imaging for decades in her research in Egypt and Tunisia. These techniques, coupled with algorithms, allowed him to discover thousands of unsuspected sites and tombs.
“New AI models can go further, look beyond known targets and reveal entirely new areas to explore”underlines Sarah Parcak. For Chris Fisher, ““We have a limited period of time to document the Earth before it fundamentally changes.”. Deforestation, wear and tear, the threats are numerous and these sites could simply disappear without us having had the chance to discover them.
If Sarah Parcak and Chris Fisher see these competitions as a way of democratizing archeology, some specialists warn of the risks of exploitation. OpenAI did not first consult indigenous groups living near the targeted areas, which could pose problems for the preservation of their heritage. The Brazilian indigenous ministry also demanded the suspension of the competition.
For Sarah Parcak, AI is not a threat to her discipline: “I don’t think artificial intelligence is going to replace archaeologists. It simply multiplies what remote sensing specialists have been doing for fifty years”.
