Tort Liability for Autonomous Vehicles in the Human-Computer Collaboration Scenario

Wang Na, He Chen

Abstract： The handling of autonomous vehicles traffic accidents involves the differentiation and coordination of product liability and motor vehicle traffic accident liability. With the continuous development of the level of automation of driving technology， the basis of liability for damage caused by autonomous vehicles will shift from the compliance of driving behavior to the safety of the driving system. Correspondingly， the liability subject gradually moves from the driving system's driver to the driving system's producer. In the intermediate technical stage where human-computer must cooperate， tort liability for damage caused by autonomous vehicles can still be carried out under the framework of motor vehicle traffic accident liability as a whole to ensure that victims can obtain adequate relief in time. On the internal apportionment of liability， the separate view of the faults of the motor vehicle driver and the defects of the automatic driving system does not conform to the technical structure of "Shared Autonomy"; there are also difficulties in differential evaluation. In this regard， an integrated assessment should be carried out based on the "reasonable expectation standard，" with risk assessment as the core， to reasonably delineate the responsibilities.

Keywords： artificial intelligence; automatic driving; motor vehicle liability; product liability

CLC： D 912                   DC： A           Article： 2096?9783（2022）02⁃0116⁃10

1 Statement of Problem

Autonomous driving is a product of the deep integration of the automotive industry and artificial intelligence technology， and it is also the most eye-catching application scenario of artificial intelligence. According to the Technology Roadmap for Energy Saving and New Energy Vehicle， China has initially formed an independent innovation system for intelligent connected vehicles in 2020 and started constructing smart cities. It can be predicted that autonomous vehicles will gradually embark on high-speed industrialization， marketization， and routinization[1].

However， while maintaining high expectations for automated driving technology， society will also face systemic risks caused by uncertainty about the development of artificial intelligence. In May 2016， an electric car produced by Tesla Inc. in the auto-driving mode caused a severe accident， which has raised questions about the safety of auto-driving technology and brought to the surface the issue of responsibility for auto-driving car accidents. In 2021， the Tesla brake failure incident at the Shanghai Auto Show once again drew attention from all walks of life and put forward an urgent test of China's current legal liability system.

Since 2017， the academic circles in our country have conducted heated discussions on the liability for damage caused by autonomous driving. At this stage， a preliminary consensus has been reached in the denial of the legal status of autonomous driving vehicles[2]， the affirmation of product liability applicable to independent vehicle manufacturers[3] and the including of motor vehicle liability insurance package[4]. However， there is still a dispute about how the current motor vehicle accident liability and product liability rules apply in the autonomous driving scenario; the specific application relationship between the two responsibilities under different automated driving technology stages is still lacking in a particular discussion.

Therefore， based on the existing community consensus and the practical needs， it is necessary to comb the liability structure of the damage caused by autonomous vehicles in the current technical stage and make specific discussions on the applicable methods of the current liability regulations.

2 The "Three-Stage" Liability Structure for the Harm Caused by Autonomous Driving

2.1 Three-Stage System of Autonomous Driving Technology

A relatively unified understanding of autonomous vehicles' technical principles and development levels has been formed internationally. Currently， a rather influential classification definition system includes： the Automated and Connected Driving Strategy compiled by the German Federal Ministry of Transport and Digital Infrastructure （BMVI） in 20151; the Autonomous Vehicle Policy Guidelines released by the National Highway Traffic Safety Administration （NHTSA） in 20162， and the Driving Road Map formulated by the European Road Traffic Research Advisory Committee （ERTRAC） in 20173. In a comprehensive comparison， the difference between the grading criteria is not evident in the definition of the degree of automation. In addition， China's Ministry of Industry and Information Technology issued the recommended national standards for Automotive Driving Automation Classification in March 2020. See Figure 1 for specific comparison and description.

Of course， the distinction of pure technology has limited significance in the field of law， so different disciplinary perspectives must be adopted for analysis and screening. Based on the above table， it can be found that the real sense of autonomous driving technology can be divided into three stages. The first is the assisted driving stage， which includes DA and PA modes. Humans still control the driving operation， and the system only completes auxiliary works in different ranges. The second is the autonomous driving stage， including CA and HA modes， where the system meets all driving operations while human drivers can choose to intervene in different degrees. The third is the pilotless stage， which is the FA mode. The system is responsible for the entire driving， and there is no possibility of human drivers participating4.

Among these three stages， the leader of the driving operation gradually transitioned from a human driver to a computer system， and the degree of involvement of human decision-making and behavior in the process of the car showed a diminishing change. These differences will lead to changes in the corresponding evaluation standards in the law， which will affect the composition and assumption of responsibilities. Based on this， the following will analyze the liability structure of the damage caused by autonomous vehicles one by one by this technology division system.

2.2 Responsibility Framework for Each Technical Stage

2.2.1 The Assisted Driving Stage

Assisted driving technologies such as Adaptive Cruise Control and Automatic Parking have been widely used in reality. In assisted driving mode， the human driver is still responsible for continuously monitoring the driving environment and assuming most driving tasks. Even in the unique process when the automated system takes over the vehicle's operation， the driver should still monitor the driving situation and be ready to intervene at any time. At this technological stage， the process of the car still depends entirely on the free will and autonomous behavior of the driver. The independent driving system does not have space to control the car's operation independently， so it will not substantially impact the current responsibility structure.

Specifically， the motor vehicle accident liability is still applied first to the damage caused by a traffic accident in a car with a driving assistance system. According to Article 1208 of the Civil Code and Article 76 of China's Road Traffic Safety Law， fault liability applies to traffic accidents between motor vehicles and motor vehicles. As for traffic accidents between motor vehicles and non-motor vehicles or pedestrians， there are multiple theories： fault liability， the presumption of fault liability， non-fault liability， and indemnity liability[5]. Of course， no matter what approach is adopted， whether the driver of a motor vehicle is at fault has substantial meaning in terms of imputation.

At this technical stage， the driver assistance system's intervention during the vehicle's operation will have a particular impact on determining the driver's fault， including whether the driver misuses the driver assistance system and whether the driver assistance system is intervened correctly during operation. However， based on the technical description and practical orientation of the driving assistance system， that is， taking over the individual processes under human autonomous decision-making and monitoring， it is not difficult to judge the driver's fault at this stage. When the driver reasonably fulfills the obligation of monitoring the driving environment and controlling the driving process， but the accident occurs due to the failure of the driving assistance system， it will enter the field of product liability. If the malfunction meets the definition of "defect" in Article 46 of the Product Quality Law， the victim of the accident shall claim damages from the vehicle manufacturer.

According to the different links in which the defects occur， product defects can be divided into manufacturing defects， design defects， and warning defects[6]. Manufacturing defects are typical product defects and result in the strictest product liability. The operating program written by the manufacturer is stored in the driving assistance driving system. When individual "off-line products" appear in the same batch of cars， the operation program loaded on it is not the same as the version prepared by the manufacturer. It is why traffic accidents happen afterward and constitute the identification of manufacturing defects. Regardless of whether the defect originates from the program input link or the test link， the producer cannot claim to be exempted from liability because "all possible preventive measures have been taken"[7].

When the function of the driving assistance system is to provide additional safety guarantees， such as emergency braking technology， when the technology fails to operate normally and causes an accident， it should be a design defect. As a result， manufacturers may try to evade liability by providing more complicated warning instructions. For example， Mercedes-Benz made a special reminder for its DISTRONIC PLUS System （distance-speed automatic adjustment system）： "Drivers should pay attention to traffic conditions even when DISTRONIC PLUS is activated. Otherwise， you may not be able to recognize the danger in time， cause accidents and harm yourself and others." In this case， "The Restatement of the Law Third， Torts Products Liability" of the United States believes that warnings are not a simple way to circumvent manufacturing defects or design defects， nor are they a basis for identifying manufacturing defects and design defects. In other words， if a product can be designed to be safer， the producer should adopt this alternative design and not evade responsibility only by warning of the existence of risks[8].

2.2.2 The Autonomous Driving Stage

In the stage of autonomous driving， the driving operation and environmental monitoring of the car will all be completed by the system， and the human driver will only intervene to varying degrees. Whether the responsibility structure has completely changed is the core of the current academic dispute.

On an objective level， human drivers no longer have complete de facto control over the car， making the autonomous driving system seem more in line with the traditional definition of the driver. Therefore， the "general tort liability theory" believes that the identity of the automatic driving car's driver has been changed to the passenger. The vehicle manufacturer should be responsible for the accident， and the driver should only bear the general tort liability when it is at fault[9]. According to the technical description of the automatic driving stage， human drivers may still need to respond and take over in due course while the system is running， which leaves the latter still retaining the possibility of intervention. Therefore， at this technological stage， the operational control ability of the human driver has not been wholly lost， and it is not appropriate to equate its status with the passenger. It may cause the driver's system monitoring and necessary takeover obligations a mere figurehead.

In addition， when manufacturers of automatic driving cars are required to take responsibility for motor vehicle accidents， there are questions about how they should understand the faults of the producers. Specifically， the driver's fault refers to his violation of the necessary duty of care in traffic. That is the judgment of the existence and magnitude of the fault-based on whether the driving behavior violates Road Traffic Safety Laws， regulations， and related rules. Its essence lies in the evaluation of driving behavior compliance. The fault of the producer refers to the violation of the producer's safety communication obligation， that is， judging whether the applicability and safety of the product have reached the necessary level based on the general transaction view in the self-driving field， and its essence lies in the evaluation of the appropriateness of the production behavior[10]. The specific content of the two kinds of duty of care is different in the particular range and the judgment standards adopted.

A different point of view is that better to relieve the victims of automatic driving car accidents and solve the legal evaluation dilemma caused by artificial intelligence technology. It should be based on the risk attributes of the motor vehicle and let Halter assume no-fault liability， which can be described as "the theory of strict liability for Halter"[11]. In this view， some scholars further proposed that the detriment of the motor vehicle party includes Halter's liability and the driver's liability. The former applies the principle of no-fault liability， and the latter uses the principle of presumption of fault[12]. However， China's Civil Code， Road Traffic Safety Law， and other current legal norms do not have the concept of Halter. It requires further discussion on whether and how to introduce strict liability for Halter from comparative law.

Given that automatic vehicles still belong to motor vehicles， and there is still the possibility of human drivers intervening in the "autonomous driving stage"， the tort liability for damage caused by automatic vehicles can still be carried out under the framework of the compulsory motor vehicle accident liability as a whole. This will solve the problem of assuming external responsibilities to ensure that the victims promptly obtain adequate relief.

From the perspective of comparative law， countries have also shown a similar essential position on this. In May 2017， Germany's amendment to the Road Traffic Law for autonomous driving technology first clearly stipulated that there must be a driver on a motor vehicle in operation. Furthermore， the revised Article 1a paragraph 2 specifies that the system user is still the driver. Therefore， the system user （not the system provider） should still take the driver's responsibility independently in the autonomous driving stage5. At the same time， Article 1b， paragraph 2 of the amendment also sets special monitoring and takeover obligations on drivers while the automatic driving system operates. If an accident occurs in an automated driving mode， damages still depend directly on the driver rather than the producer6. Of course， if a malfunction of the autonomous vehicle causes the accident， the driver can claim compensation from the manufacturer after assuming the responsibility.

In the Vehicle Technology and Aviation Bill announced in February 2017 in the United Kingdom， new compulsory car insurance regulations were introduced for autonomous vehicles. Victims of self-driving car accidents can claim compensation from insurance companies first， whether in driver or self-driving mode. In compliance with the current product liability regulations， the insurance company has the right to recover from the producer.

Therefore， under the framework of the current law norms， it is reasonable to insist on "the tort liability of motor vehicle traffic accidents". In this regard， some scholars argue that the causes of accidents should be distinguished， and product liability and motor vehicle accident liability should be applied separately. In other words， if it is the autonomous vehicle problem， product liability should be used. The current penalty for motor vehicle accidents can still be applied if it is the driver's problem. But the problem with this view is how to use the existing liability rules in detail. For example， when humans and autonomous driving systems interact in decision-making， what is the difference between the fault of the drivers （motor vehicle side） and the product defect of the autonomous vehicle， and how to determine them separately. In the case of complex causality， the proof of these two elements of responsibility will become the key to whether the victim can obtain relief.

2.2.3 The Pilotless Stage

In the pilotless stage， the vehicle is not equipped with a driving control device， and the occupants are all passengers. The automatic driving system will take overall environmental monitoring and driving operations. The system user has no obligation to take over and no possibility of intervention. Therefore， views based on shared autonomy were all lost on a factual basis. At this time， regardless of the nature of liability， the producer of the car will become the sole subject of liability for damages. Under the framework of the current law， the analysis of autonomous vehicles liability in traffic accidents will thoroughly enter the institutional framework of product liability[13]. However， it is worth noting that the current system of product tort liability in comparative law is based on the two core elements of producer fault and product defect as a whole. There is a substantial difference between the two based on liability， which will lead to applying different substantive law norms. Therefore， this part will be investigated separately according to this dualistic system.

In civil law countries or regions， the product tort liability in the primary civil law adopts the principle of fault liability. However， given its indirect tort of omission， through the transformation of the security obligation of communication， in theory， the producer's violation of the special duty of care is regarded as an essential basis for judging the illegality of the tortious act and the cause of imputation[10]. Based on the content of the safety obligations of different producers， different types of producer responsibilities have been gradually developed in practice， including responsibilities for product design defects， faults for manufacturing deficiencies， duties for lack of explanations， and duties for after-sales deficiencies.

Specifically， the product's design should meet the transaction's requirements， including the theoretical design and the experiment or test under the expected conditions. The manufacturer who fails to pay due attention and results in design defects shall bear the tort liability. Similarly， producers also need to perform the necessary care in the manufacturing process of products. Not only to ensure the safety and reliability of the product production process but also to adopt appropriate quality control measures to prevent consequences with quality problems from entering the market. When the product is put into circulation， the producer is also obliged to give appropriate instructions and prompts to the purchaser， including the particular use of the product and the specific hazards that may occur. After that， the producer should still monitor the safety of the products sold and take necessary measures to prevent and remedy it when necessary. This is an affirmative duty， including the obligation to warn， recall， and improve[14].

Based on the different types of products， the specific content and requirements of the objective duty of care borne by producers are also different[15]. As far as autonomous driving cars are concerned， the producers， of course， should be obliged to take care not to endanger drivers' lives and health.

However， the difficulties that may exist under this analytical framework are： firstly， it is worth discussing the relationship between this duty of care and the responsibility to ensure the running of vehicles does not violate relevant road traffic regulations， especially in situations where self-driving cars violate traffic rules to avoid personal injury. Secondly， the producer's duty of safety care is restricted by predictability. In other words， only the danger that can be reasonably foreseeable to the producer will lead to a corresponding task of result avoidance. The most prominent feature that distinguishes driverless technology from other stages of autonomous driving technology is that the operating system itself can self-learn and adapt based on experience. As scholars have pointed out， although humans pre-set the operating rules of the autonomous driving system， they may not be able to foresee the system's specific decisions in specific situations[16]. As a result， the complete autonomy of the unmanned driving system will change the producer's ability to foresee danger， thereby affecting the judgment of reasonable duty of care. Finally， whether the producer adequately fulfilled the obligation of result avoidance is determined based on the relative trade-off between the cost of safety prevention that the defendant has not adopted and the safety benefits that the prevention should have achieved. In practice， discussions are usually conducted through the possibility of alternative solutions7. Therefore， producers may try to evade responsibility by providing more complex warnings to buyers or simply reducing the speed of the car， but this will be at the expense of the expected utility of driverless technology.

As a particular rule of general tort liability， the structure of product defect liability is the damage caused by product defects. And its particularity lies in the change of the elements of liability from "producer fault" to "product defect". Therefore， at the establishment stage of responsibility， the producer won't be held whether there is subjective intention or negligence in the occurrence of the damage but only makes an objective judgment on whether the product's properties have defects[17].

The so-called defect refers to the lack of reasonable safety of the product， which often adopts the consumer's good expectation standard as the judgment. The common expressions are "common expectations of ordinary consumers"， "general recognition of the public"， and so on[18]. The decision of the reasonable degree is a comprehensive evaluation， and the consideration factors include the characteristics of the product itself， the regular use， the period of circulation， and all related matters. When a product is abnormally dangerous under regular use， it can generally be determined that it does not meet consumers' reasonable expectations for product safety. A preliminary determination of "defect" can be reached[19]. Through the rule of primary-face proof， the court can directly infer the causation between the defect and the damage based on the evidence that the victim used the product involved and suffered an injury. Therefore， when applying product defect liability in the driverless stage， although the victim still needs to take the initial burden of proof for the defect and causality， it is easier than proving the producers fault.

3 Responsibility Determination in the Human-Computer Interaction Circumstance

The current automatic system represented by Tesla Autopilot gives the public a chance to see the future of automated driving. However， it is undeniable that it is still far from accurate "autopilot". Before the turning point of artificial intelligence technology arrives， we will be in the technical stage of "shared autonomy" for a long time， which is also the regulatory circumstance where academic controversy is concentrated. To cope with the upcoming CA technical stage， detailed interpretation work should be carried out under the current legal liability framework to clarify the specific liability determination rules， which have urgent practical significance.

3.1 The Technical Characteristics and Burden Sharing Problems of "Shared Autonomy"

Shared autonomy is a mode in which both the driver and the intelligent car control system can control the autonomous driving car under the condition of not fully autopilot， which corresponds to the technical stage of CA and HA. In the shared autonomy environment， the autopilot system and the driver share the right of decision-making and control of the car. The dynamic driving task is transformed from a traditional continuous process to an alternative automatic and manual driving method[20]. See Figure 2 for the switching process of control right.

<H：＼飛翔打包文件＼2022年科技与法律第二期＼3.2-科大科技与法律杂志2022年第2期（英文部分）＼Image＼image2_1.png>

Figure 2   Switching process of control right in the shared autonomy environment

The technical difficulty at this stage is the switching of control right. First of all， in the process of switching control from autonomous driving car to human， whether the driver can effectively recognize and evaluate the current driving state， then take over the operation of the vehicle， and ultimately avoid risks. Furthermore， how to make reasonable performance evaluation on the process of control proper switching and select the appropriate switching request timing to optimize the effectiveness of human-computer interaction to avoid accidents due to "automation bias".

The technical "hand-off problem" will directly lead to the complexity of sharing legal responsibilities. After the driver of a motor vehicle assumes external responsibilities by Article 76 of the Road Traffic Safety Law， whether it can recover compensation from the producer of the autonomous driving car， and the amount of compensation lies in the fault of the motor vehicle and the product defect of the self-driving vehicle exist or not. Under the autopilot mode， the human driver and the automatic driving system interact in decision-making， making it challenging to apply the traditional judgment methods of the two responsibilities.

According to the description of different automated driving technology stages， human drivers have other intervention spaces for driving operation， and their level of attention as a driver should be different. Suppose the regulations insist on applying the fault standards of traditional motor vehicle drivers. In that case， it does not meet the significance of the development and application of autonomous driving technology. It is more likely to transfer the responsibilities of the car manufacturer to the driver. In the circumstance of deep human-computer interaction， the human driver's "foreseeable possibility" and "avoidable possibility" of damage are substantially affected by the operating state of the automatic driving system. For the fault judgment of the motor vehicle driver， the appropriateness of the design and operation of the automated driving system still needs to be considered. Correspondingly， the human driver's decision will react to the operation effect of the automated driving system， which makes it difficult to deviate from the appropriate evaluation of the specific behavior of the motor vehicle driver in the identification of product defects. As a result， in the determination of two different responsibilities， there is a situation where the requirements are a coincidence.

3.2 The Foundation and Construction Path of the "Reasonable Expectation" Standard

The interaction between the appropriateness of the motor vehicle drivers' behavior and the safety of motor vehicle products originates from the technical interrelationships and has an everyday legal basis.

On the one hand， the motor vehicle driver's duty of care comes from the reasonable expectations of other traffic participants for the safety of motor vehicle operation. When the operation of a motor vehicle is entirely based on the driver， this reasonable expectation is embodied as the driver having the legal driving qualification， good driving ability， maintaining the necessary monitoring and vigilance， complying with road traffic laws， etc. Drivers who fail to meet these requirements are deemed unable to perform the duty of care properly and are at fault for the damage. On the other hand， the product must have safety that meets the general expectations of consumers， and the producer is obliged to limit the risk of injury or damage to an acceptable level. Therefore， the essential connotation of product defects lies in an unreasonable danger， and the main criterion is consumers' reasonable expectations.

In the context of the application of autonomous driving technology， motor vehicle drivers have gradually given way to the automatic driving system， but society's reasonable expectations for the safe operation of motor vehicles have not decreased. Therefore， in terms of external liability， applying liability for the compulsory of vehicle traffic accidents is still realistic and reasonable. And in the control of autonomous vehicles， the vehicle driver and the autonomous driving system must cooperate to ensure that the vehicle's operation meets reasonable expectations. As a result， motor vehicle manufacturers may also become the main body of responsibility and take responsibility for the safety defects of the automatic driving system. Under the standard of reasonable expectation， the faults of motor vehicle drivers and the weaknesses of the automated driving system can be evaluated as a whole， and they can act on the composition and assumption of different types of responsibilities， respectively， to solve the problem of internal responsibility allocation based on the determination of causation.

The occurrence of damage is a concrete realization of risk. The purpose of motor vehicle drivers is to fulfill the duty of reasonable care. The requirement that the product is free of defects is to control the risk level of motor vehicle operation within an excellent range to meet the reasonable expectations of others for the safety of others motor vehicle operation. Regardless of whether the specific safety problem comes from the improper operation of the human driver or the defect of the automatic driving system， it will eventually cause the safety risk of the self-driving car， leading to traffic accidents. Therefore， under the unified concept of liability， the key to applying the reasonable expectation standard lies in the assessment and delineation of risks.

To improve the safety of autonomous vehicles， risk assessment methods based on different motion models have been developed technically， including risk assessment based on physical motion models， risk assessment based on behavioral motion models， and risk assessment based on perception/consciousness interactive motion models， and so on[21]. However， these methods are mostly centered on automated driving systems and focused on technological improvements in maps， perception， and planning， lacking consideration of human decision-making and behavior.

The influence of humans as an uncertain factor fundamentally leads to the complexity of risk analysis in shared autonomy circumstances. In the risk analysis， the human condition， driving style， and previous human-machine cooperation experience should be fully considered. Friedman of the Massachusetts Institute of Technology Computer Science and Artificial Intelligence Laboratory defined an engineering paradigm for developing a "shared autonomous autonomy" system in a human-centered manner. Based on this， he proposed seven principles for the operation of autonomous vehicles， emphasizing that the perception should be shared based on understanding human drivers and that attention should be paid to the effectiveness and safety of autonomous vehicles as a whole， rather than individual components[22]. This risk assessment method based on the technical framework aligns with integrated assessment thinking. It also has practical significance for risk division under the shared autonomy model.

4 Conclusion

Responsibility and legislative issues need to be addressed before automated driving technology is introduced to the public. Automated and networked driving are the inevitable trends of technological progress， but its safe， reliable， and controllable development cannot be separated from the support of the legal system. The handling of responsibilities for driving car traffic accidents involves the complex issue of distributing responsibilities between drivers of self-driving systems and providers of autonomous vehicle technology. Currently， autonomous driving technology has not shaken the existing liability framework system. In the long run， as the space for human drivers to intervene in vehicle operations continues to shrink， the ultimate responsibility will gradually shift from the driver to the producer of the automated driving system.

In the intermediate technical stage of human-computer interaction， it is difficult to identify the faults of motor vehicles drivers and the product defects of the automatic driving system， leading to difficulties in sharing responsibilities. From the theory of interpretation， with reasonable expectations for the safe operation of autonomous vehicles as the connection point， it is expected to build an integrated liability determination criterion and provide a legal basis for the reasonable delineation of responsibilities. In terms of application， the concrete construction of appropriate expectation standards still mainly depends on technical risk assessment methods but should pay attention to the influence of human drivers' decision-making behavior in specific scenarios.

References：

[1] Energy Saving and New Energy Vehicle Technology Roadmap Strategic Advisory Committee. China society of automotive engineering： energy saving and new energy vehicle vechnology roadmap[M]. Beijing： China Machine Press，2016： 206⁃207.

[2] ZHU Y H. Criticism of artificial intelligence legal personality viewpoint and rational responses[J]. Law Science Magazine， 2020， 41（3）： 132⁃140.

[3] FENG Y. Tort liability for damage caused by automatic driving vehicle[J]. China Legal Science， 2018， 4（6）： 109⁃132.

[4] FENG J Y. Artificial intelligence technology and change of liability law： autopilot as an object of investigation[J]. Journal of Comparative Law， 2018， 4（2）： 143⁃155.

[5] The Research Group on Tort Law of the Supreme People's Court. Understanding and application of the provisions of the tort law of the People's Republic of China[M]. Beijing： People's Court Press， 2010： 349.

[6] American Academy of Law. Tort law restatement third edition： product liability[M]. Translated by XIAO Y P， et al.Beijing： Law Press， 2006： 273.

[7] HE C. Reflection and reconstruction of developing risk defense system in China's product liability law[J]. Science of Law， 2016， 34（3）： 135⁃144.

[8] VICTOR E， SCHWARTZ. The restatement， third， torts Products liability a model of fairness and balance[J]. KAN J L. & Pub. Pol'y， 2000， 10： 41.

[9] ZHANG T. Study on the civil liability for damage caused by artificial intelligence products[J]. Journal of Social Science， 2018， 4（4）： 103⁃112.

[10] GUO L Z. Introduction to the development of China's product liability law in the past ten years[J]. Taiwan Law Review， 2004， 110（7）： 30.

[11] YIN Q S. Problems and countermeasures of tort law of intelligent vehicle [J]. Science of Law， 2018， 36（5）： 42?51.

[12] LIU Z C. The construction of tort liability of the damage caused by autopilot vehicles[J]. Northern Legal Science， 2020，14（4）： 5⁃17.

[13] GARY E M， RACHEL A L. The coming collision between autonomous vehicles and the liability system[J]. Santa Clara L. Rev， 2012， 52： 1321.

[14] ERWIN D ， HANS J A. Unerlaubte handlungen·schadensersatz·schmerzensgeld （5. auflage） [M]. Translated by YE M Y & WEN D J. Beijing： China Renmin University Press， 2016： 132⁃134.

[15] TERUAKI T. The Japanese tort law[M]. Translated by  GU Z X & DING X S. Beijing： Peking University Press， 2011： 198⁃199.

[16] SI X， CAO J F. On the civil liability of artificial intelligence：taking automatic driving vehicle and intelligent robot as the breakthrough point[J]. Science of Law， 2017， 35（5）： 166⁃173.

[17] YU M. The Japanese tort law[M]. Beijing： Law Press， 2015： 529⁃530.

[18] Civil Law Office of the Legal Work Committee of the Standing Committee of the National People's Congress. Explanation， legislative reasons and relevant provisions of the tort liability law of the People's Republic of China[M]. Beijing： Peking University Press， 2010： 173.

[19] HE C. The judicial effect of compulsory standard in Tort Law—focus on product responsibility[J]. Seeker， 2016， 4（4）： 33⁃38.

[20] WU C Z， WU H R， LYU N C. Review of control switch and safety of human-computer driving intelligent vehicle[J]. Journal of Traffic and Transportation Engineering， 2018， 18（6）： 131⁃141.

[21] STEPHANIE L， DIZAN V， CHRISTIAN L. A survey on motion prediction and risk assessment for intelligent vehicles[J]. Robomech Journal， 2014， 1（1）： 1.

[22] LEX F. Human-centered autonomous vehicle systems：principles of effective shared autonomy[EB/OL]. [2021-7-10]. https：//arxiv. org/abs/1810. 01835.

人機协作场景下自动驾驶汽车致害的侵权责任

王  娜，贺  琛

（安徽大学  法学院，合肥 230601）

摘    要：在自动驾驶汽车交通事故的处理上，涉及产品责任与机动车交通事故责任的界分与衔接问题。随着驾驶技术自动化水平的持续提高，自动驾驶汽车致害的归责基础将由驾驶行为的合规性转向驾驶系统的安全性。与之相对应，责任主体逐渐从驾驶系统的使用者向驾驶系统的生产者转移。在人机必须合作的中间技术阶段，自动驾驶汽车致害的侵权责任仍可整体在机动车交通事故责任的框架下展开，以确保受害人能够及时获得充分的救济。而在责任内部分担上，割裂性看待机动车使用人过错与自动驾驶系统缺陷，并不符合“人车共驾”的技术结构，亦存在区别评价上的困难。对此，应基于“合理期待标准”进行一体化评价，以风险评估为核心，合理划定责任。

关键词：人工智能;自动驾驶;机动车责任;产品责任