Designing in the Age of Machines
summary
Team builds a Google Clips-inspired ML system for astronauts on the ISS, faces 1.5 years of regulation work and a year where models predict well but the prototype fails, then runs 6 months of reviews and testing. Architecture uses dialogue manager, process manager, OWL ontology and database; approach shifts to teaching the model not to lose the tool by training on a few mechanisms so it learns continuity and consistency and achieves near 98% clock detection for task prediction, increasing crew trust while avoiding invasiveness.
AI-written summary · about
chapters (10)
- 0:00Introduction to AI devices
- 1:16Building AI for astronauts
- 6:28Prototype challenges and lessons
- 9:14Training AI with clockwork logic
- 13:14Key principles for AI training
- 16:26Human needs in AI design
- 18:57Machine vs human objectives
- 21:26AI energy efficiency insights
- 22:31AI guidelines and data bias
- 23:15Q&A session
related talks
39mChanging Realities - The Final Frontier for Machine Learning?
32mTesting Machine Learning Systems: What it is and why you should care
18mChandrayaan Mindset for Your AI Career: Build Systems That Survive India
38mDesigning Machine Learning for Humans
48mPrinciples and Practices of Building Machine Learning Systems
46mDesigning Artificial Intelligence
38mBuilding robust machine learning systems
32mFrom Three Days to One Minute: Architecture Trade-offs in Production Enterprise AI