The AV Dilemma, Automotive News, ET Auto

While the dream of fully self-driving cars is still in the future, autonomous vehicles (AVs) are already part of our world. As with any form of AI, we must weigh the pros and cons before incorporating this technology into our daily lives. One of the main benefits of AVs is their potential to promote sustainable transportation; they can reduce traffic congestion and reduce our reliance on fossil fuels. Additionally, AVs can increase road safety and provide accessible transportation to hard-to-reach communities, such as people without driver's licenses.

But despite these advantages, many people remain wary of fully self-driving AVs.

Continued below

In an Australian survey conducted by Shaan Coppel of Monash University, 42% of participants said they would “definitely” not use a self-driving car to transport an unaccompanied child, compared with just 7% who said they “definitely” would do so.

Mistrust of AI seems to stem from fears that machines may make mistakes or make decisions that don’t align with human values. The concerns are reminiscent of the 1983 Stephen King horror film adaptation of “Christine,” in which a car turns killer. People worry that they will be increasingly removed from machines’ decision-making loops.

Vehicle automation is categorized into levels, with level 0 indicating “no automation” and level 5 indicating “full driving automation,” in which the human is merely a passenger.

Currently, levels 0 through 2 are available to consumers, with level 3, which offers “conditional automation,” being available on a limited basis. The second-highest level, level 4 or “high automation,” is undergoing testing. Today's AVs require a driver to monitor and intervene when automation is insufficient.

To prevent AVs from getting out of control, AI programmers use a technique called value balancing, which will be especially important as vehicles with greater degrees of autonomy are developed and tested.

Value alignment involves programming AI to behave in ways that are consistent with human goals, which can be done explicitly, in the case of knowledge-based systems, or implicitly through learning in neural networks.

For AVs, value adjustments will vary depending on the vehicle's purpose and location. They will likely take cultural values ​​into account and comply with local laws and regulations, such as stopping for ambulances.

The “trolley problem” poses a major challenge for AV coordination.

The trolley problem, first posed by philosopher Philippa Foot in 1967, explores human morality and ethics, and when applied to self-driving cars it helps us understand the complexities of aligning AI with human values.

Imagine a self-driving car heading towards a collision: it might swerve right to avoid hitting five people, putting one at risk, or it might swerve left to avoid one, putting five at risk.

What should AVs do? Which choices best reflect human values?

Now consider a scenario where the AV is a Level 1 or 2 vehicle and the driver has control: When the AV issues a warning, which way do you steer?

If you had to choose between five adults and one child, would that change your decision?

What if one of those people is a close family member like your mom or dad?

These questions highlight that the Trolley Problem was never meant to have a clear-cut answer.

This dilemma shows that aligning AVs with human values ​​is complicated.

Consider Google's Gemini language model debacle: attempts to reduce racism and gender stereotypes led to misinformation and absurd results, such as portraying Nazi-era soldiers as people of color. Achieving alignment is complex, and deciding whose values ​​should be reflected can be equally difficult.

Despite these complexities, attempts to align AVs with human values ​​are promising.

When autonomous vehicles work together, driving will become safer. Human drivers tend to overestimate their driving skills. Most car accidents are caused by human errors such as speeding, inattention, and fatigue.

Will AVs help make driving safer and more reliable? Technologies such as lane-keeping assist and adaptive cruise control in Level 1 AVs already make driving safer.

As more AVs appear on our roads, it will be important to work alongside this technology to reinforce responsible driving.

The ability to make effective decisions and drive safely, even with the assistance of an AV, is critical. Research shows that humans often rely too heavily on automated systems, a phenomenon known as automation bias. We tend to view technology as infallible.

The term “death by GPS” has become popular due to instances of people blindly following navigation systems despite clear evidence that the technology is flawed.

One notable example was when a tourist in Queensland ended up in the bay after trying to use his GPS to navigate to North Stradbroke Island.

The trolley problem shows us that technology is as fallible as humans, and even more so because it lacks embodied cognition.

Dystopian fears of AI taking over self-driving cars may not be as dramatic as imagined. A more immediate threat to the safety of self-driving cars may be humans' readiness to cede control to AI.

Uncritical use of AI can affect cognitive functions, including our sense of direction, meaning our driving skills may deteriorate as we become more reliant on the technology.

Level 5 self-driving cars may be on the horizon, but for now they rely on human decision-making and innate skepticism.

Experiencing AV breakdowns can counteract automation bias, and increasing transparency into AI decision-making processes can help AVs augment and further improve human-led road safety.

(Source: PTI)

Most Read Automotive Tech Articles