AI: The current state of play
summary
Press releases claim machines read like humans but the cited result is 82.6 vs 82.3 on Amazon Turk tasks, not human experts; deep learning excels at image, face and speech recognition and tumor detection yet is greedy, brittle, opaque and shallow, fooled by adversarial examples. Commercial projects (Watson in cancer, Facebook’s assistant) and domestic robots (Roomba vs Rosie) underdeliver; Go systems succeed because rules and data are perfect, unlike real-world tasks needing common sense and transfer learning.
AI-written summary · about
chapters (6)
- 0:00Start
- 3:56Deep Learning
- 5:11Rosie the Robot
- 6:02Downsides to Deep Learning
- 15:10Geoff Hinton
- 16:25Intuitive Physics
related talks
54mAI: Actual Intelligence
42mAI & Human Decision Making
48mThe Perils and Promise of Artificial Intelligence
13mAI: Big Expectations
04mFrom big data to human-level artificial intelligence
47mSponsored Session: Beyond the Buzz: Navigating AI's Myths and Realities
16mAI: Common Misperceptions and Uncommon Perceptions
23mHuman-Driven Intelligence: Where AI & Automation Fall Short