Viewpoint
Physical AI requires moving fast while shipping safely
Physical AI requires moving fast while shipping safely
Dmitri Dolgov says that in physical AI, breaking things is not acceptable, so the principle is to move fast and ship safely rather than move fast and break things.
- Speaker
- Dmitri Dolgov
- Source timestamp
- 2:21
More from this interview
- Physical AI faces life-cost errors, extreme latency, no digitized internet, and pre-deployment safety validation.
- An 18-month demo versus a 15-year scalable service with over 200 million miles.
- Reliability follows an exponential ladder of nines; at fleet scale rare events become daily.
- Complementary camera, LiDAR, and radar with active sensors provide redundant, superhuman safety.
- Do not anchor to today's component costs
- Waymo rebuilt its stack around AI breakthroughs; production without regressions is harder than prototyping.
- Waymo's foundation model is a multimodal world-action language model using encoder-decoder and fast/slow reasoning.
- Structured approaches should channel scale, not fight it
- Waymo's structure-augmented models improve safety validation
- Waymo's simulator is a large AI model for closed-loop evaluation and training.
- Generative models enable training on rare edge-case scenarios
- Agent, simulator, and critic AIs share a foundation world model and data flywheel.
- Metrics and evaluation build the strategic moat
- Autonomy trust is earned through safety frameworks and data
- Physical AI mirrors digital AI's earlier stage; next decade unfolds in physical world.