On the ethical challenges raised by robots powered by AI
summary
Compares an aerospace benchmark 0.0000‑something hourly fault rate to current robot control showing roughly 16% failure rate when picking objects, arguing that commercial efforts use large teleoperated datasets to train Vision-Language-Action models (GPT‑style generative nets plus video and motion traces) but still miss body‑distributed intelligence. Proposes mixing machine learning with symbolic planning, experimenting with active inference and octopus‑inspired body intelligence, and frames ethical alignment as tying those concrete architectures to human values.
AI-written summary · about
chapters (8)
- 0:00Introduction to ethical challenges
- 1:10Current state of robotics
- 3:21Vision language action models
- 6:34Research questions on robot intelligence
- 9:04Robot architectures over 60 years
- 12:11Active inference in robotics
- 13:18Key directions for future robots
- 14:43Ethical implications of robot autonomy
related talks
17mBeyond Automation: AI and the Evolution of Robotic Intelligence
19mAI: The current state of play
22mMachine Ethics - moving from Value to Values-driven insights
41mBeing Good: An Introduction to Robo- and Machine Ethics
03mMaking good robots
16mReal-World Robotics as the Next Frontier for AI?
38mBuilding Responsible AI in the Age of GPT
18mRobotics: why now?