Simple Explanation (ELI5)

It's like giving a super-smart robot a library of ancient herbal medicine books and asking it to learn how to use them - but the language is tricky because Thai words run together like a continuous stream, making it hard for computers to read!

Key Advantages & Current Limits

🟢 L'IA può analizzare e prevedere trattamenti basati su secoli di conoscenza tradizionale thailandese 🔴 La lingua thailandese è complessa da elaborare perché non usa spazi tra le parole (scrittura continua), rendendo l'addestramento difficile

Future Outlook & Takeaways

In the next 3-5 years, AI could digitize and protect the traditional medical wisdom of the world, making it accessible to everyone with precision and safety, preserving cultural knowledge for future generations.

Can AI become an expert in ancient herbs?

Article

100%

Imagine having a super-intelligent assistant capable of browsing through millennia of secret recipes in a single second. It sounds like science fiction, but it is actually happening! A new project called AppHerb is teaching artificial intelligence the secrets of Traditional Thai Medicine (TTM).

Why does this matter? Often, online information about herbal remedies can be confusing or unreliable. To solve this, researchers took a very powerful AI model, called Gemma-2, and had it study ancient sacred texts of Thai medicine. It is a bit like sending a computer to university to earn a degree in herbalism!

But it wasn't easy. The Thai language is a real puzzle for computers: it doesn't use spaces between words and has a unique writing system. It is like trying to read a text without ever finding a period or a comma! Despite these challenges, the AI managed to learn how to predict treatments and even suggest herbal recipes. This is just the beginning: in the future, AI could help us protect and spread the wisdom of traditional medical traditions worldwide, making them accessible to everyone with precision and safety.

Bonus

100%

The research focused on fine-tuning a Large Language Model (LLM) using the Gemma-2 architecture with 9 billion parameters. The main challenge was Natural Language Processing (NLP) for the Thai language, which is classified as a 'low-resource' language in the context of model training. Unlike English, Thai is a 'criptio continua' language, meaning words are written without spaces, making text segmentation extremely complex for algorithms. Researchers trained the model on two fundamental historical corpora: the Wat Ratcha-orasaram Ratchaworawihan (WRO) textbook and the Tamra Osot Phra Narai (NR). The goal was twofold: treatment prediction (TrP) and herbal recipe generation (HRG). The results demonstrated that AI can compete with existing models based on Traditional Chinese Medicine, proving that integrating specific cultural knowledge into pre-trained models is a viable path to digitizing traditional medical wisdom without losing the necessary precision.