Today I learned the first words unraveled for the Vesuvius Challenge.
Do things that are available in lesser quantities afford more pleasure than those available in abundance? Our author thinks not: “as too in the case of food, we do not right away believe things that are scarce to be absolutely more pleasant than those which are abundant.” However, is it easier for us naturally to do without things that are plentiful? “Such questions will be considered frequently.”
Philodemus1, of the Epicurean school, is thought to have been the philosopher-in-residence of the villa, working in the small library in which the scrolls were found. The text is what originally caught my attention because I like the sentiment, but what is most fascinating about the discovery is that the team of young men2 used artificial intelligence to unravel the burnt scroll virtually. It illustrates how machine learning can be applied to almost any analysis of data including that of CT scans of ancient scrolls.
I had assigned seating at an event last Saturday night and one of the fellas at our table started asking about artificial intelligence mostly as it relates to investing. After informing why my wife handles our money, and my less than stellar record of investments, I tried to explain how artificial intelligence is not something that requires the computing power of a large company or a massive farm of computers. I think many folks assume that AI is only something that can be done with massive computing power, which is simply not the case. It certainly helps when you're trying to train models on extremely large datasets, but a normal laptop can train and run some of the larger language models. As illustrated by this challenge, the majority of the code needed to build out artificial intelligence data analysis is readily available and easy to use on most computers.
The students and interns on this team used PyTorch3, an open-source machine-learning framework, and a series of other open-source packages to process the data4. The hardest part seems to be the processing of spatiotemporal 3D data and a good explanation of the process is available in the ThaumatoAnakalyptor repo5. Since this has been an ongoing challenge in the archeology world, there are numerous other projects. The contestants were given a good start to the process using other existing packages like Volume Cartographer6.
It's nice to see that the teams open-sourced their code for future research and collaboration. The original purpose of the contest was to save money because shipping all of the existing fragile scrolls to a particle accelerator in England would end up costing at least $30M. By using artificial intelligence, the researchers are hoping to save the majority and likewise increase the research speed.
The first time I ever saw a project like this was FoldIt7, which used a game with human participants to try and solve protein structure predictions. I was one of the 57,000+ participants. Foldit didn't use AI, but AlphaFold8 is its AI successor. Although most of the news is related to the somewhat novel technologies like driving your car or stacking boxes in warehouses, I think the most important applications will be pushing the pace within already established scientific fields. I think projects the Vesuvius Challange will help knock out the translation of all 800 scrolls lost to the volcano within a couple years9.
Footnotes
-
Philodemus - https://en.wikipedia.org/wiki/Philodemus ↩
-
Vesuvius Challenge - https://scrollprize.org/grandprize ↩
-
PyTorch - https://en.wikipedia.org/wiki/PyTorch ↩
-
Vesuvius Grandprize Winner - https://github.com/younader/Vesuvius-Grandprize-Winner ↩
-
ThaumatoAnakalyptor - https://github.com/schillij95/ThaumatoAnakalyptor ↩
-
Volume Cartographer - https://github.com/educelab/volume-cartographer ↩
-
Foldit - https://en.wikipedia.org/wiki/Foldit ↩
-
AlphaFold - https://en.wikipedia.org/wiki/AlphaFold ↩
-
Master Plan - Vesuvius Challenge - https://scrollprize.org/master_plan ↩