含羞草社区

Taiwan’s Unmanned Vehicles Technology Innovative Experimentation Act (also known as the Unmanned Vehicle Sandbox Act) took effect on 1 June 2019, establishing a “sandbox” for research purposes allowing autonomous vehicle testing on public roads within certain controlled and safe areas.

Under the remit of the Ministry of Economic Affairs, the primary objective promotes development of industrial technology and innovative services, rather than regulating traffic supervision or control.

In addition, article 20 of the Regulations Governing Road Traffic Safety provides further regulatory guidance for testing the vehicles. However, these regulations do not extend to general public use of autonomous vehicles on public roads. While multiple Level 3 and 4 autonomous driving tests have been conducted, general public road use for Level 3 autonomous vehicles is not yet permitted.

As a result, technology verification continues within the sandbox framework. According to classification of the Society of Automotive Engineers (SAE), Level 5 refers to fully autonomous vehicles that require no driver intervention – which are not yet available for general public road use.

On ultimate conclusion of the sandbox experiment, if an autonomous vehicle is to be legally deployed on public roads, it must comply with all relevant existing traffic laws and regulations. Supervision of such vehicles would then be transferred to the Ministry of Transportation and Communications, serving as the new competent authority for the matter.

Safety certification

Taiwan’s vehicle safety laws and regulations are inclined to align with international standards, specif-ically those of the United Nations Economic Commission for Europe (UN/ECE). However, relevant vehicle safety regulations are still under review and development.

Who is responsible?

In recent years, vehicles equipped with various types of artificial intelligence functionalities within vehicle systems (ADAS) have become increasingly common. Reports have emerged of drivers operating ADAS-equipped vehicles without maintaining proper control for extended periods, or engaging in other forms of dangerous driving.

However, under current law, AI is not recognised as a legal person and it is generally accepted that AI cannot directly bear civil liability. Furthermore, as there is no specific legislation governing legal liability of AI (the Artificial Intelligence Fundamental Act remains in draft form), issues of liability related to artificial intelligence are still addressed under the Civil Code, the Consumer Protection Act (CPA), and general legal principles.

Civil tort liability arising from using ADAS vehicles is primarily determined by the tort liability of the driver under the Civil Code, and/or liability of the automobile manufacturer for product defects under the CPA. Specifically, an injured party may argue that an ADAS vehicle qualifies as a “motor vehicle” within the meaning of article 191-2 of the Civil Code – which stipulates the liability of motor vehicle drivers – and that the user of the vehicle is the “driver” as defined under such provision.

If the injured party seeks damages under article 184 of the Civil Code, which sets out the elements and legal effects of tort liability, such party must prove the user was negligent in operating the ADAS vehicle.

Recently, a Taiwan court held that a defendant (driver) who relied on ADAS and failed to pay adequate attention to road conditions, resulting in a collision with a facility on the highway, was fully liable for damages under articles 184 and 191-2 of the Civil Code.

Additionally, with respect to the liability of automobile manufacturers, a party engaged in the design, production and manufacture of an ADAS-equipped vehicle (the “provider”) might be held liable under article 7 of the CPA if the court finds the provider failed to demonstrate its product met contemporary technical and professional standards of reasonably expected safety before the product was released to the public; or failed to post safety warnings and critical instructions for handling potential danger in a conspicuous place on a product that may cause harm to consumers’ lives, bodies, health or property.

Criminal issues

Regarding criminal liability, recent court precedents held that article 7 of the CPA might serve as the legal basis for holding the provider responsible for preventing harmful outcomes. However, this does not mean the provider is absolutely responsible for eliminating every possible harmful outcome.

Instead, the courts emphasise the importance of the provider’s duty of care. Specifically, when deciding whether to release ADAS-equipped vehicles on the market, the provider must exercise appropriate caution and judgement.

The key question is whether the provider breaches such duty of care during the decision-making process. Ultimately, the determination of criminal liability hinges on whether the provider fulfilled its duty of care in preventing harm when making such decisions, which is very different from the no-fault principle adopted by the CPA with respect to the civil liability of the provider.

Take a Level 3 autonomous vehicle as an example. Once the ADAS system is activated, the driver is still required to remain alert and ready to respond to the system’s request to take over control at any moment, whether actively or passively.

While such autonomous vehicles are designed to alleviate the driver’s burden, the allocation of legal responsibility between the driver, AI and the provider remains complex and evolving. How liability is assigned after an activation of an autonomous vehicle will significantly shape future legislative and regulatory developments.

Current laws require that a perpetrator possess either an “intentional” or “negligent” mens rea (guilty state of mind), depending on the nature of the criminal offence. There is currently no criminal statute that has been specifically promulgated or amended to address the unique challenges posed by the development of artificial intelligence. As a result, under existing law, AI cannot possess the requisite mens rea and, therefore, cannot itself be held criminally liable.

As such, the primary parties with potential criminal liability could be the driver and the provider. If the provider is deemed the subject of criminal liability, responsibility may also extend to individuals directly involved in the design, manufacture and distribution of the vehicle.

Conversely, if the driver is considered the liable party when assessing whether the negligent operation of a vehicle equipped with an ADAS gives rise to criminal liability for injury to another person, the prosecutor must establish both driver negligence and the existence of “causation” between the driver’s conduct and resulting harm, similar to requirements relevant to determining tort liability.

Recent court decisions have clarified that ADAS is merely a supplemental tool. If a driver fails to comply with the rules for using ADAS – such as neglecting to maintain a safe braking distance, falling asleep at the wheel, or otherwise abdicating control to the system without remaining alert and prepared to intervene – the driver remains subject to criminal liability for any resulting injuries.

Administrative penalty issues

Administrative penalties for traffic violations can only be imposed on drivers in accordance with the Highway Act, the Road Traffic Management Penalty Act and other relevant laws. The legislative framework of these two statutes places full responsibility for vehicle control on the driver.

Although autonomous driving technology serves as an aid to the driver’s judgement, the driver remains responsible for any accidents that occur. For instance, if a driver violates traffic regulations or engages in dangerous driving behaviour, such as zigzagging or operating the vehicle in a hazardous manner, penalties may be imposed under article 43 of the Road Traffic Management Penalty Act.

Conclusion

Currently, ADAS vehicles available on the market are not fully autonomous. However, as AI technology continues to advance and vehicles become capable of independent operation, determining civil, criminal and administrative liability for incidents involving them will become increasingly complex.

Hence, it is essential for legislators to re-examine existing laws and regulations. Clear definitions are needed regarding legal liability for human activities involving artificial intelligence, as well as the allocation of responsibility and risk among humans and AI-controlled products.

LEE AND LI, ATTORNEYS-AT-LAW
8F, No 555, Sec. 4, Zhongxiao E Rd, Taipei 110, Taiwan, ROC
5F, Science Park Life Hub, No 1, Industry E 2nd Rd,
Hsinchu Science Park, Hsinchu 300, Taiwan, ROC
Tel: +886 2 2763 8000; +886 3 579 9911
Email: attorneys@leeandli.com