Designing for Autonomous Vehicles at Scale With Hussein Mehanna of Cruise

Hussein Mehanna, Head of Artificial Intelligence at Cruise, discusses the challenges of building safe autonomous vehicles and the importance of being 'AI-native' in product development.

• AI-native products require a different methodology of building them that needs to go beyond the lean startup approach.
• Continuous learning machines that are automated are necessary for AI-native products to operate at a human level or surpass it.
• AI-first turns the problem into an AI hammer that makes everything else look like an AI nail, whereas AI transformation is a better concept.
• Academia operates off of standardized data sets, whereas the industry requires a continuous stream of data.
• Cruise has built a mindset of a continuous learning machine to break silos between teams and build safe autonomous vehicles.
• Cruise's modeling engineers collaborate with the data team to build a modeling solution that assumes a constant stream of data.
• AI cannot turn a bad product into a good one, and companies should focus on their product first before considering AI as a tool.

At Cruise, it is important for their autonomous vehicles to turn every mile into some gradient that their neural networks can learn from. To achieve this, modeling engineers build a modeling solution that assumes a continuous stream of data and collaborate with the data team to ensure that every iteration of the model does not regress due to bad data.

Hussein Mehanna, Head of Artificial Intelligence at Cruise, joins Scale AI CTO, Brad Porter, for a fireside chat to discuss how Cruise has approached the challenge of building safe autonomous vehicles that are ready for the real world. Together they explore what it means to be 'AI-native' as opposed to 'AI-first' and how one can organize broad multi-disciplinary teams to build AI applications in a continuously iterative way. In this discussion, Hussein explores the key questions that need to be answered when building safe autonomous vehicles. When creating AI models, how do you evaluate progress and ensure it's the correct type of progress? How do you design for safety when AV vehicles get safer, and potential learning examples become rarer? Is it possible to develop an AI model that can safely handle scenarios it has never seen before? Join this session to hear the key experiences and learnings from one of the leaders of autonomous vehicle development.