Google’s artificial intelligence made own AI that outperformed human-made systems at identifying objects

Briefing

Google’s artificial intelligence made own AI that outperformed human-made systems at identifying objects

December 7, 2017

Briefing

  • AI Replication – Google’s machine learning artificial intelligence, AutoML, developed own artificial intelligence, NASNet, using reinforcement learning which can identify video objects in real-time
  • NASNet Performance – Predicted images with 82.7% accuracy, 1.2% higher than human-made artificial intelligence systems, while being 4% more efficient
  • Automated Training – AutoML evaluates and improves NASNet’s performance
  • Open Source – Open-sourced AI to accelerate development of computer vision applications
  • Potential Applications – Include robots that help blind to see as well as improve computer vision of driverless cars
  • Negative Implication – Could produce AI with inherited biases

Accelerator

Sector

Information Technology, Transportation and Logistics

Organization

Google Inc.

Source

Original Publication Date

December 1, 2017

Leave a comment