How do AI-powered drones and robots actually decide what to do?

In episode 4 of No Fixed Lane, Rashad and Mammad explore AI in military drones, police drones, and consumer robotics. They separate two ideas that often get blended together: an AI model reasoning about a task, and reinforcement learning teaching a robot the physical actions through trial and error.

They discuss reported drone swarms in Ukraine, naval kamikaze boats, Tesla's robotaxi, and why autonomous systems fail in edge cases they weren't trained on. The episode closes with how robotaxis and robots like Optimus could reshape car ownership, work, family routines, and even insurance.

Chapters

00:00 AI in defense, policing, and robotics

01:14 How LLMs and reinforcement learning fit together

02:05 Turning a task into subtasks: cleaning a desk

03:45 Trial and error: rewards and penalties in robotics

06:02 Training military drones in simulated scenarios

07:35 Police drones and civil surveillance

08:44 Hard-coded logic vs. reinforcement learning in swarms

09:33 Why robots plan before they act

10:36 Drone swarms in Ukraine and target coordination

11:17 Naval kamikaze boats and rapid defense innovation

12:50 Why AI reasoning and reinforcement learning get confused

13:24 Making the bed: splitting chores into subtasks

14:19 Robotaxis, Grok, and driving logic vs. LLM reasoning

15:31 Why autonomous systems fail in edge cases

16:41 What the robotaxi is actually learning

17:42 Self-driving ambulances and emergency response

18:10 The data race to train better models

18:48 How robotaxis could change car ownership

20:01 Shared vehicles and the economics of robotaxis

22:08 Scheduling life around autonomous pick-ups

22:34 Optimus, household robots, and the cost of convenience

23:47 Why change may arrive faster than expected

"If they are not trained in those cases, I don't think they will be successful."

Follow No Fixed Lane for new episodes on tech, health, business, and life.