Autonomous cars: potential lifesavers but with new risks

Episode Summary

Executive Summary: The episode examines autonomous vehicle safety from two angles: functional safety and cybersecurity. WMG’s Siddharth Kastgeer argues no fully driverless car is safe today, and that public expectations should shift from “absolute safety” to informed, domain-limited safety. Simon Parkinson adds that connected vehicles expand attack surfaces, making hacking a real concern. Both stress that regulation, simulation, and proper user understanding are essential before widespread deployment.

Main Topics: Autonomous vehicles are not fully safe today (Priority: 5/5): Kastgeer argues that current driverless systems cannot be considered safe without a safety driver or remote operator. He emphasizes that autonomy must be limited to known domains where performance can be validated. From absolute safety to informed safety (Priority: 5/5): A central theme is the need to replace the unrealistic expectation of perfect safety with informed safety: users must know what the vehicle can and cannot do, and use it only within those limits. Different autonomy levels and use cases (Priority: 4/5): The episode distinguishes low-speed shuttles and last-mile transport from high-speed fully autonomous driving. These uses have different sensors, environments, timelines, and safety challenges. Human error as the main road-safety problem (Priority: 5/5): Autonomous systems are framed as safety tools because many crashes are caused by human factors such as drowsiness, distraction, and drunk driving. Automation can assist rather than fully replace humans in the near term. Simulation and validation in research (Priority: 4/5): Kastgeer explains how autonomous systems are tested in simulated environments with sensors and vehicle brains in the loop to evaluate responses to scenarios such as pedestrians stepping into a zebra crossing. Cybersecurity and attack surfaces in connected cars (Priority: 5/5): Parkinson explains that vehicle connectivity creates opportunities for attackers to exploit sensors, communication links, and software-hardware integrations, potentially affecting braking or steering-related functions. Standards, supply chains, and future-proofing (Priority: 4/5): The discussion highlights the importance of security ownership at board level, supply-chain accountability, secure updates, and designing vehicles that remain resilient over long lifespans.

Key Arguments: No autonomous vehicle without a safety driver or remote operator should be described as safe; autonomy must be constrained and validated. Public marketing often overstates capability, as with Tesla’s Autopilot, which is an assistance feature rather than true autonomy. Autonomous vehicles can reduce accidents by helping with drowsiness, distraction, and other human-error scenarios, but they will not eliminate risk entirely. Simulation is necessary because testing risky scenarios in the real world could harm people; virtual testing allows safe validation of edge cases. The biggest obstacle may be human behavior and misunderstanding, not the underlying technology alone. Connected vehicles are vulnerable because their sensors, wireless links, and embedded software create multiple possible attack paths. Security must be built into vehicle design and supply chains from the start, not patched in after deployment. Updating vehicles securely over their lifespan is a major unresolved challenge because cars remain in use for many years. Even if attacks are hard, researchers have demonstrated that some vehicle systems can be manipulated to affect acceleration or braking. The goal is not perfect safety, but meaningful risk reduction and careful domain-specific deployment.

Data Points: Global annual road deaths: 1.25 million - Kastgeer cites worldwide road fatalities each year to motivate autonomous vehicle safety research. UK annual road fatalities: 1,700 per year - He says the UK has had roughly this number of road deaths every year for the last six years. Share of accidents due to human error: 90% - Kastgeer attributes most road accidents to human error rather than vehicle technology. Estimated UK societal cost of road accidents: £35 billion - He includes healthcare and infrastructure costs associated with accidents. Potential safety improvement threshold: 10% better than human drivers - He argues that even modest improvements over human driving could save many lives. Potential lives saved over 30 years: 100,000 lives - Kastgeer estimates the long-term impact of autonomous vehicle improvements. Low-speed shuttle deployment timeline: 5 to 6 years - Expected commercial deployment window for last-mile autonomous shuttle services. Full autonomous vehicle timeline: 15 to 20 years - Estimated time before fully autonomous home-to-office vehicles become feasible commercially. Vehicle lifespan: 15 years - Parkinson notes that vehicles remain in the field long enough for security assumptions to become outdated.

Pivotal Quotes: "Autonomous cars today are not safe. They will never be 100% safe but if we know what they can and cannot do we can still benefit from them." — Siddharth Kastgeer: He defines the core safety philosophy behind his work on verification and validation. "Absolute safety is a myth" — Siddharth Kastgeer: He uses this phrase to argue that public expectations should shift toward informed safety and constrained use. "If we can create autonomous vehicles which are even 10% better than human drivers in the next 30 years we can save 100,000 lives." — Siddharth Kastgeer: He concludes by quantifying the potential public-health benefit of safer automation.

Implications: Listeners should expect autonomous vehicles to arrive first in limited, low-speed roles, not as universally safe robot cars. Industry must combine safety engineering, cybersecurity, regulation, and user education to reduce risk and build trust.