On Robots and Insurance

Insurance is a contract aimed at shielding the insured party from the adverse economic consequences of a future and possible risk, should that risk materialize. The risk can be of any type, depending on a negligent behaviour of the insured, or of third parties, causing harm to the same party entering the contract (first party insurance) or to others (third party insurance). The losses due to its materialization instead, may pertain both to the health and estate of the contracting party. Pursuant to art. 1882 of the Italian civil code (henceforth c.c.), insurance is defined as:

The contract whereby the insurer, against payment of a premium, undertakes to retaliate the insured, within the agreed limits, the damage caused by an accident [...]²

According to the definition above, the prerequisites of this contract are: (i) an agreement between the parties, (ii) the existence of a risk to the insured party or potential third parties, (iii) the payment of a premium.

The latter is a function of the risk; hence it varies according to both the likelihood of its occurrence, and the severity of the consequences that may arise once it materializes. Therefore, the very possibility to enter such a contract rests on the availability of data related to those parameters. In the absence of adequate statistical information, the insurance company will not be able to determine the premium and most likely will refuse to contract for the specific risk.

Insurance is one of the possible legal tools that parties can resort to in order to manage the risks they face with their activities, both professional and not (e.g.: driving). The contract will thence be entered into in order to avoid (shift) the negative economic consequences the party would otherwise face because of the application of liabilities rules set forth by existing legislation that—for the issue here addressed—may range from professional to enterprise liability as well as product liability.

Normally, the parties are left free to decide whether to enter an insurance contract or not, in order to manage a risk they are exposed to. Their personal preferences, their risk aversion and ultimately their utility functions will determine whether a contract will be signed, and at which conditions, as for any other asset-management-related choice.

However, in some cases, the legislator deeming that the risk associated to a certain activity is too high, or that moral hazardous behaviour and adverse selection may negatively affect the pooling and spreading of the damages that arise, may legislatively impose a duty to purchase insurance (normally third party insurance) [5‐7]. This is the case with compulsory traffic insurance [8] as well as with professional liability insurance (typically medical doctors and lawyers), and more recently also with drones [8].

So briefly sketched the notion of insurance, and its purpose, we can attempt to address the questions of whether (i) robotic applications do pose novel issues in this perspective, (ii) whether the risks they pose can be insured, and (iii) who eventually is better suited to bear the burden associated therewith, hence to purchase coverage.

Given the variety of applications (from surgery to domestic chores) and the differences in national legislation, in this paper we will refer mainly to Italian and European law, when applicable. The considerations drawn are however of a general nature and can be extended to other legal systems.

It is possible to distinguish two main typologies of industrial robots: robots operating in isolation from human beings, usually constrained inside protective cages; and “collaborative” robots, which are designed to interact physically with workers, such as Baxter by Rethinking Robotics [13] or UR5 by Universal Robots [14]. While conventional industrial robots are already market products, with EC mark, Baxter and other similar robots designed for physical collaboration with workers are not yet commercialised in Europe. As to insurance, it seems possible to argue that there are no specific requirements neither for producers, nor owners, nor users of conventional industrial robots. As a matter of fact, they are very well established in manufacturing and their production and use fully covered by regulations [15] and standards [16]. On the contrary, safety standards for “collaborative” robots are still under development [17]. Therefore, it remains uncertain whether insurance companies are already prepared for covering the risks associated with the deployment of collaborative robots such as Baxter, considering that workers insurance is compulsory in many countries, such as Italy.

Parenthetically, the conventional assumption that automation increases the overall safety of an activity has led some scholar to consider the consequences of a safer workplace. By reducing the risks to which workers are exposed to in factories, it is likely that workers’ compensation coverage can be reduced too [18]. Whether collaborative robots will increase or further reduce the safety of the workers interacting with them is not clear yet.

A service robot ‘is a robot that performs useful tasks for humans or equipment excluding industrial automation application’ [19]. An example of service robot for non professional use is Roomba by iRobot, which was introduced in the market since 2002 [20]. Roomba is an autonomous robot designed to vacuum clean floors. There are many other robots similar to Roomba that are now commercialised, such as such as pool, window or gutter cleaners. For these kinds of “chore” robots there are no specific insurance requirements for production, ownership or use. This is due to the fact that the risks are easily identifiable and do not present significant physical hazards for people or things, given the small sizes and limited capabilities of these robots. The same is true for some of the most popular entertainment and educational robots sold in the market, such the robotic dog AIBO by Sony (model ERS-7) and the small humanoid robot Nao Evolution by Aldebaran.

According to the regulatory information provided by the manufacturers, the products are compliant with European Directives as well as European and international standards.³ Nevertheless, saying that there are no insurance requirements or that robots are available in the market does not imply that they might not bring about risks, in particular to human safety and security. However, since these robots are relatively new, there are no long term studies showing evidence of such risks. The problem has been succinctly described by the Collingridge dilemma: the impact of a new technology cannot be easily predicted until the technology is extensively developed and widely used. However, at that time, it would become difficult to remediate to eventual problem [21].

Still in the field of service applications, a special case is deserved by civilian drones. A drone is intended by the International Civil Aviation Organisation (ICAO)⁴ as ‘an unmanned aerial vehicle, which is a pilotless aircraft, in the sense of Article 8 of the Convention on International Civil Aviation, which is flown without a pilot-in-command on-board and is either remotely and fully controlled from another place (ground, another aircraft, space) or programmed and fully autonomous’ [22].⁵ Therefore, with respect to human control, drones can be divided into manned and unmanned aircraft, that is, devices remotely operated or autonomous, respectively. Manned drones are also known as ‘Remotely Piloted Aviation Systems (RPAS)’, namely: ‘A set of configurable elements consisting of a remotely-piloted aircraft, its associated remote pilot station(s), the required command and control links and any other system elements as may be required, at any point during flight operation’ [22]. RPAS are currently the subject of new regulations for use in airspace. As pointed out by ICAO: ‘Fully autonomous aircraft operations are not being considered in this effort, nor are unmanned free balloons nor other types of aircraft which cannot be managed on a real-time basis during flight.’ [22]. Indeed, the Italian Civil Aviation Authority has issued the ‘Remotely Piloted Aerial Vehicles Regulation’, which entered into force on April 30th, 2014 [8]. As far as insurance is concerned, according to Art. 20 of the Italian regulation: ‘A RPAS cannot be operated without valid, adequate third party insurance, not less than the minimum insurance coverage of the table in Article 7 of Regulation (EC) No 785/2004’ [8]. However, as pointed out in the literature, insuring drones can be complicated given the fact that there is no sufficient data on the hazard that can be caused by RPAS [23].

Going back to unmanned drones, that is, drones that can fly autonomously, without any human intervention, although they are not yet authorised for use, either by ICAO or under EU regulations, nevertheless, it is worth pointing out that they are already available in the market. An example is Hexo Plus, a drone with a camera that can follow and film a person autonomously [24].

Self-driving cars are not yet on the market, although the technology is currently being experimented on many public roads, also in Europe [25], thanks to special legal permissions [26]. As pointed out by [27]: ‘The problem of driverless cars has shifted from one of engineering and technology, to one of legislation and insurance’. Outside of the medical field, self-driving cars are one of the most discussed topics with respect to robotics and autonomy [28]. Concerning insurance, according to [29], for the management of risks deriving from autonomous vehicles new business models are needed. Likewise surgical robots, autonomous vehicles could increase safety and therefore may determine a decrease of insurance costs [30]. The conclusion seems however implausible and needs to be further specified. No matter how safe driverless vehicles will be, accidents will still occur, due to causes and malfunctions that may be hard to foresee at the time being. Accidents due to human lack of attention and reckless behaviour on the street will clearly become negligible, but new vulnerabilities will certainly emerge. For instance, the fact that driverless vehicles will make use of numerous sensors and connections will give raise to the relevant issue of their cyber security and vulnerability to attacks brought about from a distance. The role of policy makers in fostering available insurance product for autonomous vehicles will be determinant.⁶

As far as insurance issues are concerned, the central point seems to be the human action (or inaction), in terms of either choosing to override the autonomous robot’s actions in the name of safety, or simply letting the robot continue its intended actions. If an accident occurs, the legal debate will be whether it is the manufacturer’s fault, or the individual’s fault for taking over [31]. With a fully autonomous vehicle the responsibility for avoiding an accident shifts entirely to the vehicle and the components of its accident avoidance systems [32]. Some might argue that if cars really do become safer, there might not even be a need for motor insurance [33] with evident negative consequences for the insurance market. However, as stated in Lloyd’s report on Autonomous vehicles [34], some element of risk would be retained by the owner of a car (e.g. damage or theft can still occur when a car is parked in a driveway).

Finally, another special kind of service that robots can perform is in the medical field. A definition of medical robots within standards organisations doesn’t exist yet, however the phrase is very popular in the literature.⁷ Most of “medical robots” are used by professional users, such as Da Vinci surgical robot by from Intuitive Surgical [36]. However, there are also medical robots designed for patients, such as DEKA Arm System by DEKA [37], a bionic arm for amputees, which has been approved by U.S. Food and Drug Administration (FDA) for commercialisation or the robot suits HAL by Cyberdyne [38], an exoskeleton designed to assist people with reduced mobility, which is under FDA approval. Given the variety of applications, we will consider here the robot Da Vinci, which is already sold in the European market. The robot is a tele-operated robot currently used for minimally invasive surgery [36]. Since it is difficult to identify a common attitude towards insurance coverage among European and non-EU countries, we will report here the statement taken from the Da Vinci factory: ‘da Vinci Surgery is categorized as robot-assisted minimally invasive surgery, so any insurance that covers minimally invasive surgery generally covers da Vinci Surgery’ [36]. Searching in the literature, the most recurrent issues concerning surgical robots and insurance are about the coverage of specific surgical interventions by the national health insurance service [39, 40] and about the reduced health insurance costs resulting from robotic assisted surgery [41]. These issues are relevant also for other medical applications of robots, such as prosthetic devices, as reported by [2]. Finally, a few scholars focus on the need for hospitals and doctors to have insurance that covers robotic assisted surgery [42, 43]. Indeed, it is questionable whether the robot fits in the existing contracts for medical doctors or as a consequence of new risks, new contracts should be devised.

Although robots are intended to perform the tasks they are designed for in a safer way than the human being they replace or assist, it is not possible to exclude the occurrence of a dangerous event. Each robot presents different kinds and levels of risk, which depend on how the robot is made of (its nature, including size, shape, weight, material etc.), what the robot is doing (its function or task), where the robot is employed (its operative environment) and whom it interacts with (its relationship with human beings or objects). The fact that nowadays robots are deployed in several public environments in physical contact or proximity with users or bystanders, reduces the possibility to devise extrinsic protective measures, such as fences or helmets, as it was with early industrial robots in factories, and highlights the need to design intrinsic safety solutions.

Let us now consider the kinds of risks that a robot may cause. Likewise any other artefact, the risks generated by a robot can be divided into damage to a person and to things and property.

As far as personal damage is concerned, we propose to consider two kinds of damage which may happen using a robot:The distinction is meant to take into account the current evolution in the field of human robot interaction from merely physical to cognitive and affective.

Hard damages are physical, that is, related to the body of a person and caused by physical human–robot interaction [44]. Physical damages can be caused by defects in the correct working of a robot and therefore they are ascribed to the producer. However, they can also be caused by users to third persons or objects as a consequence of wrong use, software or hardware alteration, lack of maintenance.

On the contrary, we propose to call “soft damages”, mental or psychological damages, which are related to the mind and can affect the cognitive, social and even emotional capabilities of a person. This kind of damage may derive from prolonged interactions with special kinds of robots, such as personal care robots. Psychologist Sherry Turkle defines “nurturing machines” robots that are eliciting in their users a compulsion to do something for them, e.g. feeding or interacting [45]. This kind of danger may be compared to addiction to video-poker machines or the Internet, which are new entries in the Diagnostic and Statistical Manual of Mental Disorders edited by the American Psychiatric Association [46]. Another example of soft hazard could be caused by the so-called “environmental generational amnesia” [47], that is, to get used to and prefer the artificial copy or surrogate to the natural as a consequence of the increasing replacement of reality with technological surrogates.

Related and preliminary to the presence and definition of risks is their identification and evaluation. The evaluation of a risk, which is at the basis of the definition of the insurance premium, is based on the definition of its frequency and its severity.

The difficulty in assessing risks with respect to robots is due to (i) the technical complexity of robotic devices, (ii) the lack of sufficient data with respect to the potential risks and the accidents they may cause, (iii) the uncertainties with respect to liabilities that producers and users may face.

With respect to the difficulties in assessing technological risks due to robot complexity, the close interaction of insurance companies and robot designers and producers may prove useful, despite non-sufficient. The issue concerning the lack of substantial data about potential risks and the accidents, may be partially addressed through extensive testing in real life environments, which do prove useful information when interactions of the device in the environment are closer to reality. Indeed, the possibility to test robots in their operating fields rather than in laboratories would allow situating robots in their environments, namely in designing socially as well as legally ‘embedded robots that are structurally coupled with their surroundings’ [48]. As regards to this, deregulation areas [49] or robotics innovation facilities (RIF) [50] may prove useful. However, in the case of devices to be used in very unstructured and extremely variable and diverse scenarios—like robotic prostheses and exoskeletons—only the diffusion and real use of the device—and subsequent accidents caused—will provide more reliable data [33]. In this case, ethical considerations should be carefully evaluated since the experiments would involve human subjects. Although valuable at research level, the argument in favour of “real use” as the best way for risk assessment, to be a viable solution should receive approval from the national Ethical Committees and together with the informed consent, be in line with regulations on human clinical trials and with the Helsinki Declaration. Finally, as far as the uncertainty in liability is concerned, this issue may only be partially addressed. Indeed, all existing and forthcoming robotic applications are regulated by existing liability rules, which, unless provided otherwise, do apply. Therefore the problem of the “legal gap” is a false problem, since anything that exists, is also regulated by the legal system where it rests. The real issue instead, is whether the existing legal framework is the best and desirable one for that kind of technology, given its social relevance, desirability and ethical and economical importance. In this perspective, sound policy arguments can often be formulated both to ensure high levels of product safety and adequate compensation to the victim, while allowing the growth of a market for these applications.

In the next section, we will provide a brief discussion of the problems associated with the application of liability rules to robotic devices.

Robots are products [51],⁸ hence product liability rules do apply. Product liability rules make the producer responsible for all damages caused by the user and third parties by the functioning of the device. In particular, when damage occurs as a consequence of the use of a robot, its design may be deemed defective and not sufficiently safe. What kind of design may be deemed defective is however a matter of fact, decided ex post by a judge in a trial. In some cases, the novelty of robotic technologies may cause the outcome of a judicial decision to be highly uncertain. Moreover, the absence of sufficiently narrow-tailored technical standards [52] may increase this uncertainty. To better illustrate this concept, the case of robotic prostheses may be shortly considered. Prostheses, pursuant to the European directive on active implantable medical devices [52], need to abide those technical regulations in order to receive the EC marking, which is required for their commercialization in Europe. However, that same directive applies to anything ranging from a pace maker to an industrial robot, intuitively very distinct devices. At the same time, pursuant to existing legislation even if the device complied with the safety standards set forth by the directive, liability would not be excluded if some accident occurred while the product was being used.

Other existing liability rules may be considered at a theoretical level in order to determine who may be called to pay for damages arising from the use of robots. Rules such as that establishing the liability of the person for the damage caused by a thing in custody or by an animal [53], as well as a general negligence standard (such as that set forth by art. 2043 c.c.) can be considered [54, 55], and [51]. Addressing the details of those regulations falls beyond the purposes of the current analysis. However, it shall suffice to say that given that such rules were not designed with robots in mind, anticipating the way they may be applied in a trial is not easy, and the lack of significant data about previous occurrences does not allow to easily inferring workable conclusions.

Particularly complex is the disentangling of liability in cases of articulated human-machine interaction, for instance in the case of driverless vehicles before they become completely autonomous. In such cases, the responsibility of the human “driver”, still partially in control, interferes with that of the producer and programmer, who designed the vehicle. Precisely assessing how a court could distribute the negative consequences deriving from an accident in such cases is again quite difficult, and strongly rests on the very possibility of tracking down the material cause of the event.

Finally, if the possibility of machine learning is considered, further uncertainty is added to the determination of which party ought to be held liable for the harmful consequences arising from the use of the robot. On the one hand, producers may be called on to compensate pursuant to the claim that they failed to provide the device with adequate safety measures to prevent the negative outcome that arose as consequence of its learning capacity. Such reasoning is ultimately not any different from that of all product liability claims grounded on the defectiveness of design.

As for the case of the driverless vehicle, disentangling the two different kinds of liability may prove complex and certainly rests on the ascertainment of the facts that were performed ex post. However, unlike the case of driverless vehicles, the learning capacity that could prove problematic in this respect is much more remote, representing a problem only in a theoretical way, emerging only in the medium-to-long run [56].