From: lexfridman
Artificial intelligence (AI) plays a pivotal role in enhancing vehicle safety, especially in the context of autonomous driving. Chris Gertie, a professor at Stanford University, has extensively studied how AI can be integrated into both race cars and public road vehicles to perform at or beyond human levels of driving capability [00:00:07].
Technological Innovations
One prominent example of AI-driven vehicle innovation is the autonomous race car, Shelly, developed at Stanford. Shelly has been fine-tuned to achieve speeds of up to 120 miles per hour on a racetrack, exceeding the capabilities of even expert human drivers [00:03:13]. The approach used combines physics calculations with machine learning algorithms to maximize the use of tire friction and adapt to changing conditions like tire temperature [00:05:01].
Policy and Regulatory Challenges
Gertie highlights the regulatory challenges posed by AI and automated vehicles. In the United States, vehicle safety regulations are based on a system of federal motor vehicle safety standards which currently do not directly address automation [00:07:09]. Manufacturers self-certify their compliance with safety standards, contrasting with pre-market certification required in other industries like aviation [00:08:11].
To address these challenges, the US Department of Transportation rolled out a federal automated vehicle policy encouraging manufacturers to follow a voluntary safety assessment [00:15:01]. This assessment includes the definition of an “operational design domain” that specifies where and under what conditions an automated vehicle is intended to operate [00:17:00].
Ethical Considerations
The ethical dilemmas surrounding AI in vehicle safety include deciding the appropriate course of action in potential crash scenarios, commonly illustrated by the “trolley problem” [00:23:00]. Gertie argues that such decisions will likely be governed by engineering principles focused on reducing risk rather than philosophical debates on morality [00:24:13].
Data Sharing
Data sharing is crucial for the advancement of AI in vehicle safety. Sharing information on edge-case scenarios and anomalies can enhance the training of AI systems [00:34:19]. However, this must be balanced with privacy and intellectual property concerns. Successful examples of data sharing in other sectors, such as aviation, provide a model for what might be achieved in the automotive field [00:36:13].
Data Sharing for Safety
The aviation industry’s Esaias system shows a successful model of anonymized data sharing that improved safety. This concept can be adapted for vehicle safety to accelerate AI advancements [00:36:11].
Future Opportunities
The ongoing development of AI in vehicle safety represents a significant opportunity not only for improving the safety of automated and autonomous systems but also for maintaining the United States’ leadership in this transformative technology [00:39:01]. As AI-driven vehicles become more prevalent, addressing the policy, ethical, and data-sharing aspects will be critical in ensuring these technologies are safely integrated into society.