Roads must be safe and accessible to all. But according to WHOAlmost 1.35 million people die in traffic accidentseach year. Traffic accidents are still the leading cause of death for people aged five to 29 years.
During a global ministerial conference earlier this year, the Stockholm Declaration set the new target of halving the number of deaths and injuries from road accidents by 2030.
Autonomous cars have been touted as a solution that can make mobility safe, economical, sustainable and accessible to all. But can autonomous driving systems be trusted to make real-time decisions about life and death? And is the world ready to be driven in these vehicles?
These burning questions were addressedat a panel discussionduring the AI for Good Global Summit 2020.
Technology that encourages focus
The auto industry has long touted self-driving capabilities to customers for safety and comfort on the road.
Tesla recentlysayingA beta version of its "fully autonomous driving" software will be released to some "experienced and cautious drivers". Five years after its prototype debuted on the streets, Alphabet's WaymoOpenits "fully driverless" taxi for the public in Phoenix, USA.
Road safety is regulated internationally by the Road Traffic Conventions of 1949 and 1968.
While the technology shows promise in preventing accidents caused by human error, there are concerns about whether self-driving cars are designed to adapt to changing traffic conditions.
(Video) Chris Urmson: How a driverless car sees the road
"There is now too much focus on technology and too little on users and their traffic environments," said Luciana Iorio, chair of the UNECE Global Forum on Road Safety (Working Group 1), guardian of the conventions.
A technology-driven approach ignores the fact that more than half of all road fatalities are pedestrians, cyclists and motorcyclists. Protecting these vulnerable road users requires well-designed cities and infrastructure.
"Autonomous technology should not create a digital divide and should be a transformative opportunity for everyone around the world," added Iorio.
The danger also arises when automakers overstate their technology as more autonomous than it actually is. Liza Dixon, PhD student in the field of human-machine interaction in automated driving, coined the term "autonowashing" for this phenomenon.
international SAEdefines that vehicles have six levels of automation depending onthe amount of attention required of a human driver. At levels zero through two, humans control and monitor the traffic environment; At levels three through five, automated systems manage and monitor the environment.
Reality versus exaggeration
Advanced driver assistance systems (ADAS) in cars today feature partial automation at SAE Level 2. Many consider SAE Level 5, where the automated driving system can function in all types of traffic and conditions, out of their reach, focusing on SAE Level 4 with limited Driving conditions such as small geographic areas and good weather.
Consumers may not be aware of these distinctions. Videos showing people sleeping in their cars or watching movies give a false and dangerous impression of technology, Dixon said.
“All consumer vehicles on the road today require the driver to be ready to take control at all times. It's important for the driver to know that this is a cooperative support system that enhances their skills and not something that takes their driver's role,” said Dixon.
Language and perception are also important. FORStudy by the American Automobile Association (AAA) Foundation.two groups of participants presented the same driver assistance system, but with different names. One group was informed about the possibilities of the "AutonoDrive" system and the other about the limitations of the "DriveAssist" system. According to the study, the former rather mistakenly believed in the system's ability to recognize dangers and react to them.
Open structures for citizen security
Efforts are being made to correct the lack of industry-wide standards for safe autonomous driving.
World Economic ForumSafe driving initiativeShe wants to create new governance structures that will influence the safety practices and policies of the self-driving car industry. The proposed framework focuses on a scenario-based approach to ensure security.
"This is based on the premise that the way autonomous vehicles are handled today has been tested by exemptions and non-binding government regulations that are insufficient in the long term," said Tim Dawkins, WEF's director of automotive and autonomous mobility. . .
The safety of autonomous vehicles can only be determined in the context of their environment, he added.
IsITU Focus Group on AI for Autonomous and Assisted Drivingworks to establish international standards for monitoring and evaluating the behavioral performance of "AI Driver" vehicles with automated driving. suggested ainternational driver's license testfor the AI driver who requires a demonstration of satisfactory driving behavior.
The UNECE Working Group 29, responsible for the harmonization of global vehicle regulations, said: "Automated vehicle systems must not, within their operating domain (OD), cause traffic accidents that result in reasonably foreseeable and avoidable injuries or fatalities.
According to Bryn Balcombe, president of the ITU Focus Group and founder of the Autonomous Drivers Alliance (ADA), terms like "reasonably foreseeable" and "avoidable" still need to be revised.defined and agreed to ensure they meet public expectations.
IsIEEE Global Initiative on the Ethics of Intelligent and Autonomous SystemsIt has also worked on comprehensive and specific safety standards across all industries, from healthcare and agriculture to autonomous driving.
“We have taken a general approach to establishing ground rules to build trust. We applied different considerations to different case studies and tested them,” said Danit Gal, member of the Executive Committee and Technology Advisor to the Office of the Deputy Secretary-General of the United Nations.
Frameworks need to take into account the ever-changing nature of autonomous machines, he said, also noting the challenge of aligning international standards with national regulations.
Gal proposed a clearinghouse for various AI-powered mobility initiatives to share their insights. Balcombe createddie Global Initiative on AI and Common Datawhich promotes the exchange of open datasets and results. "This is to ensure that the technology can be developed in a way that allows it to be widely adopted and not create and exacerbate some of the divisions that we currently have," he said.
The "Molly Problem" of Ethical Decisions
According to Matthias Uhl and Sebastian Krügel from the Technical University of Munich, public trust and the perception of AI and its ability to explain decisions in road traffic are decisive for the future of AI-supported safe mobility.
Along with ADA, Uhl and Krügelcreated the Molly's Problem survey to support the requirements gathering phase for the ITU Focus Group.
An alternative version of the tram problem thought experiment, the Molly problem, addresses the ethical challenges that need to be considered when autonomous vehicle systems fail to prevent an accident. The premise is simple: A girl, Molly, is crossing the street and is hit by an unoccupied autonomous vehicle. There are no eyewitnesses.What the audience expects next is reflected in a series of questions that define the "Molly Problem".
ADistributed poll before panel discussionattempted to collect public feedback and received 300 replies at the time of the event.
Uhl and Krügel noted that most respondents wanted the AI system to be able to store and retrieve information about Molly's accident. They also wanted to know the condition of the vehicle from the time and place of the accident.Speed at the time of the collision, when the risk of a collision was identified and what action was taken. Respondents also believe the software should be able to explain if and when Molly was recognized by the system and whether she was recognized as a human.
“Around 73% of respondents said that while they are very excited about the future of autonomous vehicles, they believe vehicles should not be allowed on the road if they cannot remember this information. Only 12% said it might be on the way,” Kruegel said.
An even greater challenge for the industry is that 88% of respondents believe similar near miss data should be stored and retrieved.
The results of this research will help identify the data and metrics requirements to shape global regulatory frameworks and safety standards that meet public expectations for autonomous driving software.
A step towards explainable AI
Existing event data loggers focus on collecting collision information, Balcombe said. Currently there is no common approach or system capable of detecting near misses, he added.. “Black box” recorders for self-driving cars only indicate whether a human or system was in control of the vehicle, or a request to hand over control was made.
"It's about explainability. When there has been a death, whether it be an accident or surgery, post-event explanations help build trust and work towards a better future,” Balcombe said.
Krügel believes these concerns about autonomous driving require input from social scientists who can work with engineers to make algorithms ethical to society.
Gal recognizes that different constraints and ethical approaches raise challenging questions for promising solutions. “What happens when we find a system that works best for a specific audience and someone else isn't happy with a decision made during an accident? To what extent can they really challenge it, and to what extent can the machine reflect these ethical decisions and act in split seconds? She asked.
In 2019, a United Nations high-level digital cooperation body that promotes global multi-stakeholder dialogue on the use of digital technologies for human well-being issued this recommendation:
"We believe that autonomous intelligent systems should be designed in such a way that their decisions can be explained and humans are responsible for their use."
Discussions at this event will feed into the ongoing consultation process of the UN Artificial Intelligence Panel.