TED - How AI Is Decoding Ancient Scrolls | Julian Schilliger and Youssef Nader | TED
The transcript discusses the use of AI to digitally unroll and read ancient scrolls that were buried by the eruption of Mount Vesuvius. Youssef Nader, an Egyptian AI researcher, led a team that won the Vesuvius Grand Prize for their work in this area. The scrolls, believed to contain writings by the Greek philosopher Philodemus, were charred and damaged, making them unreadable by traditional means. The team used CT scan technology to create 3D models of the scrolls and developed algorithms to digitally unroll them. The challenge was to make the ink visible, as it was carbon-based and blended with the carbon-based papyrus. By training machine-learning models to recognize the faint ink patterns, the team was able to recover text from the scrolls. This process involved iterating on existing tools, improving algorithms, and using computer vision techniques to enhance the visibility of the ink. The project was open-sourced, allowing others to contribute and further the research. The team successfully recovered over 2,000 letters and 14 columns of text, providing new insights into ancient writings.
Key Points:
- AI can digitally unroll ancient scrolls, revealing lost texts.
- CT scan technology creates 3D models of scrolls for analysis.
- Machine-learning models are trained to recognize faint ink patterns.
- Open-source collaboration enhances research and development.
- Over 2,000 letters and 14 columns of text were recovered.
Details:
1. 🤖 AI's Role in Understanding History
- AI is revolutionizing the way we interpret historical data by uncovering patterns and insights that were previously unnoticed, offering new interpretations of historical events.
- For example, AI-driven analysis of large datasets from historical records can reveal trends and correlations that challenge existing historical narratives.
- This approach allows historians to construct a more nuanced understanding of the past, as AI can process vast amounts of information more efficiently than traditional methods.
- AI's application in history includes techniques like natural language processing to analyze historical texts and machine learning algorithms that can predict historical trends based on past data.
2. 🏆 Meet Youssef Nader and the Vesuvius Challenge
- Youssef Nader is an Egyptian AI researcher and a PhD student at the Free University in Berlin.
- He led the team that won the Vesuvius Grand Prize last year.
- The Vesuvius Challenge is an international competition aimed at leveraging AI to solve complex problems, attracting top talent worldwide.
- Youssef's team stood out by developing a novel approach to AI problem-solving, which was key to their success.
- Winning this challenge highlights Youssef's innovative capabilities and potential in the field of AI research.
3. 📜 The Ancient Past Preserved by Vesuvius
- Nearly 2,000 years ago, the eruption of Mount Vesuvius preserved the works of Philodemus of Gadara, a prominent Greek philosopher, by burying the Villa dei Papiri in Herculaneum under layers of ash and mud.
- Philodemus's philosophical discourses on subjects like music, pleasure, and enjoyment were meticulously transcribed onto papyrus by scribes, aimed at enlightening future generations.
- The catastrophic eruption, while devastating, inadvertently shielded these papyrus scrolls from decay, allowing them to survive through millennia.
- The Villa dei Papiri, an epicenter of Roman intellectual life, housed these invaluable texts, offering a rare glimpse into ancient philosophical thought and cultural practices.
4. 🕵️♂️ Rediscovery and Challenges of Ancient Scrolls
- The excavation in the 17th century led to the discovery of ancient scrolls among statues and frescoes, sparking significant interest in their contents.
- These scrolls appeared like pieces of charcoal due to their charring from a pyroclastic flow, which also shredded and deformed them, posing significant challenges to researchers.
- Despite the difficulties, these scrolls are believed to contain potentially groundbreaking knowledge, motivating ongoing research.
- Modern technological advancements, such as multispectral imaging, are now employed to read the scrolls without unrolling them, preserving their integrity.
- The scrolls represent a significant historical treasure, offering insights into ancient life and thought, and their study continues to evolve with new techniques.
5. 💡 Innovations in Digital Archaeology
- Historically, the lack of technology required physical unrolling of scrolls, resulting in their destruction.
- Currently, only the most damaged and deformed scrolls remain in their original rolled-up state.
- The advent of the computer age has introduced new possibilities in digital archaeology.
- Brent Seales, a researcher and professor, proposed using CT scan technology to digitize scrolls.
- The goal of using CT scans is to digitally unroll the scrolls without causing physical damage.
6. 💻 The Vesuvius Challenge and Digital Unrolling
- The CT scan process is used to visualize the tightly wound papyrus, which presents challenges for digital unrolling.
- In 2022, Nat Friedman and Brent Seales launched the Vesuvius Challenge to motivate global participation in solving the digital unrolling problem.
- The challenge offered both monetary incentives and 'eternal glory' as rewards to attract problem solvers.
- A community on Discord was formed for participants to exchange ideas and collaborate on the challenge.
- The speaker, inspired during their master's thesis at ETH Zurich, joined the challenge and collaborated with Youssef and Luke Farritor, ultimately winning the grand prize.
7. 🏅 Achievements and Algorithm Development
7.1. Global Recognition
7.2. Initial Challenges
7.3. Computer-Vision Problem
7.4. Tool Improvement
7.5. Team Collaboration and Segmentation
7.6. Algorithm Enhancement
7.7. Community Feedback
7.8. Performance Improvements
8. 🔍 Finding Hidden Ink with AI
- The algorithm was used to unroll the papyrus, but the ink remained invisible due to being carbon-based, similar to the papyrus itself.
- The University of Kentucky team tested for ink presence in CT scans by analyzing broken scroll pieces, generating 3D data for analysis.
- Infrared imaging provided a ground truth for visible ink, allowing for machine learning model training to identify letters.
- The model analyzes small data cubes to detect ink presence, improving its accuracy over time.
- The trained model began accurately identifying letters, indicating the ink's presence despite initial invisibility.
9. 🧠 Training AI to Recognize Ancient Text
- AI was specifically trained to identify faint letters in CT scans of ancient papyrus by leveraging contrast and speckle patterns that are challenging for humans to detect.
- The AI's ability to recognize letters involves analyzing specific crackle patterns within the papyrus structure, which enhances the accuracy of text identification.
- There is a significant financial incentive for correctly identifying patterns, highlighted by a historical reference to a $40,000 prize for successful identification.
- This approach demonstrates AI's superior capability in detecting minute features in complex data environments, offering an efficient alternative to human recognition.
10. 🔄 Iterative Improvements in AI Models
10.1. Initial Model Development
10.2. Iterative Training Process
10.3. Overcoming Challenges and Achieving Success
11. 📚 Success in Reading Ancient Text
- First coherent word read from an unopened papyrus sheet, marking a significant breakthrough in ancient text reading.
- Process scaled within weeks to produce columns of text and special characters, which were of particular interest to papyrologists.
- Approach was open-sourced, facilitating community involvement and competition for a grand prize.
- Achieved recovery of four paragraphs with 85% clarity, indicating high accuracy and effectiveness of the approach.
- Success attributed to perfecting data and model through numerous iterations and experiments.
- Recovered more than 14 columns of text and 2,000 letters, showcasing the extensive capability of the methodology.
12. 🔮 Future Prospects of AI in Historical Discoveries
- AI has revolutionized the process of uncovering historical artifacts, with 2,000 characters stored two millennia ago being rediscovered in just nine months using AI technology.
- This rapid rediscovery was facilitated by AI's ability to improve code and algorithm development, providing insights into the past that were previously inaccessible.
- AI's role is expanding, with expectations to unlock more historical data, thereby enhancing our understanding of history and providing new opportunities for research and education.
- Specific applications include deciphering ancient scripts, reconstructing fragmented artifacts, and predicting historical events' impacts through data analysis.