Last Week in AI AI Fails to Read PDFs, OpenAI Jukebox, and more!
May 9, 2020
This week, experts tackle the hurdles AI faces in reading PDFs, crucial for analyzing COVID-19 literature. They discuss a Kaggle competition aimed at improving data extraction from medical papers. The conversation compares Google's AI for eye disease detection with OpenAI's music-generating Jukebox, delving into practical challenges and ethical concerns. Additionally, the potential of Moxie, a social robot designed to aid children's emotional development, is explored, highlighting its role in enhancing parent-child interactions amidst increasing screen time.
AI Snips
Chapters
Transcript
Episode notes
AI's PDF Problem
- AI struggles with the lack of standardization in PDF formats, making it difficult to extract information.
- Humans easily adapt to varying PDF layouts, but AI requires significant effort.
Google's Medical AI in Thailand
- Google's medical AI, designed to detect eye disease, faced real-world challenges in Thailand.
- Trained on high-quality images, it struggled with cell phone images and poor internet connections, highlighting the gap between lab accuracy and real-world performance.
Interdisciplinary Approach to AI
- AI researchers need to be more interdisciplinary and interact with other fields.
- Understand the workflow and context before creating AI tools.