The European Parliament proposed in their Resolution that “An obligatory insurance scheme, which could be based on the obligation of the producer to take out insurance for the autonomous robots it produces, should be established.”
42 Currently, in the European Union, the duty to take out insurance exists in a number of instruments. To name but few, these are the Motor Insurance Directive,
43 the Regulation on Insurance Requirements for Air Carriers and Aircraft Operators,
44 and the Directive on the Insurance of Shipowners for Maritime Claims.
45 The proposals in the Resolution and the Report drew an analogy between the compulsory insurance system in place in respect of motor third party liability under the Motor Insurance Directive, and the one that is sought to be implemented in respect of smart robots.
46 The below sub sections seek to address whether this analogy is well-founded by reference to the functions of compulsory insurance and the nature of motor and product liabilities. They also provide an overview of circumstances which may impinge on the moral hazard of smart robot producers and how measures taken to control the moral hazard may affect third parties.
3.1 Functions of Compulsory Insurance and the Analogy Between Compulsory Motor Liability and Compulsory Product Liability Insurance
Compulsory insurance is one of the effective mechanisms in dealing with the compensation of third party losses effectively. Albeit it may be difficult to enumerate all the circumstances which would justify the adoption of a compulsory insurance scheme, certain common parameters can be noticed in analysing the areas where the duty to take out insurance was imposed. One of these parameters would be the protection of the potential wrongdoer who may not necessarily be in a position to effectively assess the likely advantages of having insurance
47 for whom the insolvency risk may increase. This may be one of the reasons why motorists, who may often underestimate the potential consequences of driving both for themselves and others, are required to take out insurance. Similarly, insurance was also imposed in the European Union in respect of the operations of air carriers and aircraft operators, the operations of which have notably been regarded by the EU as carrying a great potential of insolvency risk.
48 The analysis of such potential was studied following the impact assessments conducted before the adoption of the relevant regulations,
49 and the operation of the insurance scheme adopted was assessed through minutely prepared reports.
50
A second, and perhaps more obvious justification for compulsory insurance is the efficient protection of third parties affected by the actions of the wrongdoer. Compulsory liability insurance would in this sense serve the tort liability norm of compensatory justice. Third parties would particularly benefit from compulsory insurance where the tort liability judgment exceeds the wealth of the wrongdoer/insured: Instead of being under-compensated by the insured, they would recourse to insurance, provided they have a right of direct action against the liability insurers. The risk of the victim in failing to be fully compensated is accordingly sought to be avoided by the introduction of compulsory insurance, as is the case in the European Union under the Motor Insurance Directive.
The scheme proposed by the European Parliament begs the question of whether insurance for product liability shall indeed be made compulsory for the manufacturers of smart robots. It is noteworthy that this suggestion was also formerly raised by the UK Department for Transport (DfT) in their Consultation on Automated Vehicles. It was initially proposed by the DfT that compulsory motor insurance should be extended to cover product liability in circumstances where the motorists were not in charge of the vehicle (i.e. where the vehicle was on autonomous mode).
51 This had required the owner of the vehicle to take out insurance that covered both the manufacturer’s and other entities’ product liability which would have responded to the claims by the ‘not-at-fault vehicle driver’ while the vehicle was on autonomous mode, as well as the ones by the passengers and third parties.
52 This approach was later on abandoned in favour of another policy requiring less radical changes based on the response received from the automotive and insurance industries. The relevant policy advocated a single insurer model (covering both the driver’s use of the vehicle and the AV technology) where the third party victim would have a right of direct action against the motor insurer, who would in turn have a right of recourse against the responsible party, where for instance the loss is caused by product failure.
53 This solution was advanced in the anticipation that product liability and motor insurers would in the future develop instruments so as to deal with the recourse stage as efficiently as possible, and that the government should have left the market dynamics play an active role without adopting an over-regulatory approach.
The obvious concern with respect to the analogy drawn in the Resolution and Report would lie in that in the case of product liability insurance, the insured will be a commercial entity (producer) and in the case of motor liability insurance, mostly a consumer with a rather limited wealth. Imposing a duty to take out insurance would therefore arguably be more justified in the latter case than in the former as there would be a greater risk that the damages would exceed the wealth of the insured and the third parties may accordingly be protected against this risk through compulsory insurance. One other remark could perhaps be expressed as regards robot producers having financial assets that are greater compared to the sources that the insurance companies can offer. In these types of cases self-insurance may arguably appear as a more convenient option as regards the level of protection guaranteed.
54
A further reason why the abovementioned analogy may be regarded as rather unfit rests upon the distinction between the markets for motor vehicles and robotics: As much as the former is predictable in terms of insurable risks, the same is yet to be achieved with respect to the latter. Furthermore, there needs to be a sufficiently large number of insureds bearing risk exposure profiles that are alike for the insurer to refer to past risk profiling experience to accurately predict and accordingly quantify the risk.
55 Probabilities of accidents by robotics may not always be easily estimated given the scientific uncertainty surrounding them, which will in turn cause difficulties for insurers in setting fairly charged premiums. Particularly with respect to ‘emergent behaviours’ of robotics, i.e. “modes of behaviour which were not predicted by the designer but which arise as a result of unexpected interactions among the components of the system or with the operating environment”,
56 the fundamental query for the insurers is how they will be placed to charge fair premiums where robots act in ways not even predictable for their programmers and trainers. This type of concern about smart robots which does not arise in the motor liability context would beg the question of whether other tools such as risk-sharing between operators
57 that is adopted particularly in respect of risks where knowledge of probabilities is limited, would be a more suitable option.
58 In the light of the foregoing, even if a preference is expressed in favour of a mandatory protection scheme, this should perhaps not be confined to compulsory liability insurance,
59 given that ‘emergent behaviours’ that may gradually become an area of concern in respect of predictability of robot actions would substantially make it difficult for insurance markets to offer affordable premiums.
Compulsory insurance is usually implemented as a solution to risks that would pose great danger to third parties, and it can therefore be argued that as the level of risks posed by different types of robots will not be identical, compulsory insurance may only be required and necessary for those robots that present a high level of risk of damage.
60 For instance, care robots would be more likely to pose a greater risk of bodily injury or death to third parties compared to robot toys, and whether the latter category should be subject to a compulsory insurance regime needs to be carefully thought through. In carrying out an impact assessment as to whether imposing compulsory insurance is necessary and justified in the context of robotics, circumstances such as the level of autonomy and predictability of the robot’s behaviour, human presence in the environment where the robot operates, robot’s physical capabilities and its connection with the environment may be taken into consideration.
61 However reason would dictate that due regard should also be had to whether insurance markets will easily accommodate a policy decision in favour of compulsory insurance. Oftentimes, developed and sufficiently large markets that are well equipped are required to cope with the demands of policyholders and third parties, and it would appear that the uncertainties surrounding robotics as well as different market characteristics in the EU are far from being reassuring in this regard.
3.2 Connection Between Strict Liability and Compulsory Insurance
The Product Liability Directive which would be likely to apply in establishing the smart robot producers’ liability for third party damages
62 establishes a strict liability regime.
63 There may be strong correlations between strict liability and the requirement of compulsory insurance which has also been showcased in several jurisdictions through the introduction of compulsory insurance for liabilities occurring without fault.
64 From the ‘insolvency’ perspective, the injurer under a fault-based liability scheme would face such a risk once the
costs of care would exceed its wealth, whereas a problem of underdeterrence would arise under strict liability as soon as the
damage exceeds the injurer’s wealth.
65 This latter situation would constitute one of the grounds for requiring compulsory insurance against the risk of underdeterrence by the injurer and accordingly that of the externalisation of costs.
66 A further justification for introducing compulsory liability insurance where strict liability applies may lie in that this could enhance incentives to reduce risk
67 which, in the context of producers of smart robots, could translate into incentivising the increase of safety levels of the products. Therefore, an analysis of whether the strict liability of producers under the Product Liability Directive should be complemented by a compulsory insurance scheme would need to be carried out to identify the advantages and drawbacks of such a policy decision.
68
Product liability is an area where although strict liability is established, no general duty to take out insurance is imposed—the Product Liability Directive does not compel manufacturers putting their products into circulation within the European Union to take out insurance cover against potential third party claims. There is a large number of producers operating in the EU which are covered against strict liability arising from the Product Liability Directive under general insurance contracts (product liability insurance can be provided as a sub-section or endorsement of a combined public liability policy) with a considerable number of producers not even being insured against this risk.
69 The foregoing being the case, sectoral legislation applicable to the producers of certain products may impose a duty to provide financial security, a recent example of which appears in the Medical Devices Regulation.
70
There may be several policy reasons for establishing strict liability, such as encouraging necessary incentives for investing in product safety; however to what extent this suggestion would prove right is controversial on the ground that it might rather persuade producers in purchasing insurance for matters outside their control.
71 Moreover, even where compulsory insurance is introduced, liability insurers will provide compensation to third party victims only where the victims are successful in proving that the product causing their loss was defective, and that there was a causal connection between the defect and the loss. It has already been acknowledged that these two instances constitute 53% of the cases where a third party claim was rejected due to their failure in discharging the burden of proof.
72 It would not be a fallacy to anticipate that this burdensome process would likely to be worsened in disputes involving robotics. Achieving the aim of protecting the victims may therefore lie in addressing this problem first, before a policy decision favouring the introduction of compulsory insurance can be made.
73 One should also not lose sight of the potential effect of introducing compulsory insurance on producers that may accordingly be incentivised to pass the cost of the compulsory insurance premiums onto the consumers in the form of an increase in product prices.
Product liability is mentioned as merely part of the network of liability that the Resolution proposed where the owners and users of robotics as well as programmers were mentioned as potentially liable parties. The list provided is possibly only of illustrative nature and could also cover whomever is involved in the chain having either given instructions to the robots or trained them. As much as the standard of liability of producers is relatively clear given the Product Liability Directive, whether the liability of the foregoing parties will be strict or fault-based is yet to be identified. In either case, the standard of liability established will be required to be adequate in addressing also robots’ ‘emergent behaviours’.
74 Determining the standard of liability is likely to require an analysis of the rules governing liability under the respondeat superior principle,
75 liability for the acts of children
76 and of animals,
77 as liability for the acts of smart robots is regarded to be analogous to the foregoing. One substantial challenge of this initiative would however lie in the lack of harmonisation of the tort law rules applicable in the EU jurisdictions.
78
3.3 The ‘Development Risk’ Defence
Under the Product Liability Directive, manufacturers are not liable if they prove “that the state of scientific and technical knowledge at the time when the product was put into circulation was not such as to enable the existence of the defect to be discovered”.
79 It has been argued in several instances whether unintended behaviour of smart robots resulting in a damage to third parties may constitute a ‘defect’ within the meaning of the Directive, and whether the development risk defence could relieve manufacturers of smart robots in a great number of cases on the ground that robotics technology is constantly evolving.
80 Before assessing the potentials for this suggestion, a general overview of the defence will be provided with a focus on the judicial approach to the defence, the branches of industry that most rely thereon, as well as the frequency of such reliance thus far.
The defence aims at striking a fair balance between fostering innovation within Europe and the protection of consumers,
81 yet the Member States were given the option not to adopt the defence in their national instruments implementing the Directive.
82 The practice on derogation accordingly differed among the Members States: Luxembourg and Finland transposed the Directive by adopting the derogation without limitations whereby the derogation was made applicable to all categories of products and producers.
83 Hungary, in turn, adopted the Directive together with the development risk defence, which however does not apply in respect of medical products.
84 In Spain, manufacturers cannot invoke this exemption in respect of medical products and food products where the latter are produced for human consumption.
85 Moreover in France, the defence may not be relied upon where the damage is caused by an element of the human body or by products derived therefrom.
86 These national restrictions are accompanied by a strict interpretation of the defence by the Court of Justice of the European Union (CJEU) which confirmed that the defence would apply where the producer could prove that the objective state of knowledge that is at ‘the most advanced level and not restricted to the relevant industrial sector’ at the time the product was put into circulation was not such as to allow the discovery of defect in the product.
87 The defence is also not based on the unavoidability of the defect, but on the accessibility of knowledge by the producer.
88 It may be relatively clear that a confidential study that has not yet been published may not satisfy the accessibility test, however a more elaborate question may be whether a study published in a single country only in the local language could do so.
89 The defence is notably the most recurring liability exemption which has been triggered in 4% of the cases;
90 the foregoing hurdles in invoking the defence nevertheless ended up in a minimal number of cases where the producers were successful.
On the one hand, the proponents of this defence could argue that removing it would endanger innovation,
91 however the successful reliance thereon could risk that consumers are left without compensation. This would accordingly result in a protection gap for consumers and a potential consequent recourse to social security systems established in Member States; hence a plausible risk-sharing scheme as regards the scientifically unknown risks is necessary and unavoidable.
92 It is noteworthy that the problem of no compensation in the event where the development risk clause is successfully invoked will arise particularly in schemes where insurance is taken out for products other than automated vehicles, where a system is implemented whereby motor insurers will be the ultimate payers of claims where the manufacturers or their insurers rely upon the development risk clause so as to exonerate from liability.
93 Where no such scheme is in place and the clause is successfully triggered, third party victims will not be able to be compensated by insurers. Consistent successful reliance on the development risk defence by producers could also disincentivize a risk-averse producer to take out liability insurance, as its purchase decision would be made based on its assets in relation to potential liabilities, the likelihood of these liabilities and the degree of risk aversion.
94
Whereas an option available is to remove the application of the clause in respect of AI and robotics amid concerns that third parties may be left uncompensated, chances are that this could raise product liability insurance premiums which may accordingly be passed onto consumers through price increases. This could also have a domino effect on R&D expenses whereby companies could possibly economise thereon and ultimately increase safety risks. Where, contrary to the proposal of the Resolution, product liability insurance is not compulsory and in theory no reliance is permitted for producers of robotics on the clause, it is submitted that it would be fairly difficult for producers to find a market to insure their development risks, given that they are rare and often result in severe damages.
95 On the other hand, a system where the producers can rely on the development risk defence and are required to take out product liability insurance would be likely to increase the moral hazard of producers who would be less incentivised to observe safety standards. In such a system, there would be a risk that third party victims may not be compensated unless the specific provisions making the insurance compulsory prohibit insurers to rely on the producers’ defences.
3.4 Deductibles
A tool adopted for controlling the behaviour of insureds is, among others, to agree deductibles in insurance policies. In first-party insurance, deductibles serve the function of eliminating some claims altogether where they do not reach the figure stated in the deductible clause; however in liability insurance they prevent third party victims from claiming losses not reaching the deductible limits from insurers,
96 who are then left with the option of seeking them from the liable parties themselves. In theory, insurers may impose a high deductible in a product liability insurance policy to evade claims not reaching the stated limits and incentivise producers to adopt safety measures given that the risk of those claims would have been allocated to them. This could however create an unnecessary hurdle for third parties particularly where the type of loss suffered is death or personal injury that exceeds the deductible. In such a case, third parties would have to claim both against the producer and the insurer (should they have a right of direct action against insurers) for full compensation.
Further complications in addition to the above may also arise due to the differences in the wording used in deductible clauses for aggregating losses. In product liability policies, deductibles are often written either on per-occurrence or per-annum basis. Where the latter may be relatively straightforward in providing for the maximum amount to be borne by the claimant within a single policy year, the former would give rise to considerable controversy because of the multifarious meanings that can be attached to ‘occurrence’.
97 For the purposes of damages arising from the acts of defective robots, the fundamental query would lie in whether (a) the defect that results in several harmful acts causing separate damages; or (b) each harmful act of the robot arising from the same defect causing separate damages; or (c) each separate damage, would qualify as ‘occurrence’. The interpretation of the term ‘occurrence’ would accordingly dictate whether the deductible would apply to the entirety of damages arising from the same defect, or whether a different deductible would apply for each act of the robot that results in damages. Due to the risk that varying meanings can be allocated to this wording in different European Union jurisdictions, unintended consequences in the treatment of third party victims may arise.
98
However, where insurance is mandatory, deductibles may not be relied upon by insurers. The Motor Insurance Directive, for instance, dealt with this particular issue by providing that insurers are not allowed to require an injured party
99 to bear an excess.
100 A similar provision may also be adopted in the context of product liability policies for personal injury damages arising from a defect in the robotics manufactured to the effect that the insurers would not have the right to rely on contractual provisions—such as deductibles—to deny third party claims.
Property damages suffered by third parties would however be subject to a different system than personal injuries’. The Product Liability Directive states that producers are not liable for property losses suffered by product users which do not exceed €500, provided that the item causing third party loss is ordinarily intended for private use and was mainly used by the third party as such.
101 This figure was either interpreted as a threshold whereby losses not exceeding the figure would not be claimed, or as an excess that would have to be deducted from the indemnity.
102 In both cases, unless the property loss exceeds the figure of €500, no liability of the producer—and accordingly of the insurer—will arise. Accordingly, no compensation will be available for the third party. This situation will further be accentuated where the limit is increased in respect of property losses arising from the use of new technologies
103 which will potentially leave out a great number of small claims arising from the acts of robotics that will have to be borne by the victims. It is also noteworthy that where the property damage exceeds both €500 and the policy deductible, third parties would have to claim both against the producer (for the difference between €500 and the policy deductible) and against the insurer (for the excess of the policy deductible).
3.5 Precautionary Measures
Another option available to insurers for controlling the moral hazard of producers is to monitor their behaviour through policy clauses such as precautionary measures.
104 The rationale behind monitoring such behaviour rests upon the fact that the liability of the producer would trigger the insurers’ own liability and any action taken towards decreasing the likelihood of this trigger would alleviate the insurers’ risk. In the general context of product liability, however, it may be difficult for insurers to achieve this aim due to several reasons. Firstly, given that for the insurer’s liability to arise the product would need to be defective within the meaning of the Product Liability Directive, the insurers’ monitoring would have to aim to reduce the occurrence of defects. How this can be ensured is, though, far from being an easy task: defects are often developed during the production stage, however product liability insurance would often be purchased before the product is put into circulation, i.e. after the product has been developed. Accordingly, any steps towards monitoring the behaviour of the producer would merely have
ex-post effect. Secondly, a clause seeking to monitor the behaviour of the producer by reference to compliance with the General Product Safety Directive
105 (GPSD—which operates
ex-ante) may ensure a certain level of control; yet would arguably not grant sufficient protection for insurers: Non-compliance with safety requirements enshrined in the Directive would not necessarily result in the defectiveness of the product, or, from the insurers’ perspective, compliance therewith would not in all circumstances prevent defect. Moreover, it is available to producers to allocate their risk of liability as well as expenses arising from recalling their defective products from the market onto liability insurers under ‘product recall insurance’.
106 This could further disincentivise a producer having this type of cover from adopting a higher level of care in complying with the GPSD.
The above suggests that insurers would frequently stipulate precautionary measures to have an
ex-post control, yet this arguably would not prove entirely useful for increasing product safety incentives that would have mostly occurred at the product development stage. An exception to this may however occur where a potential liability can be avoided if the producer is made aware that the safety of the product is called into question and acts to remedy the product deficiencies by for instance issuing additional user instructions. This latter possibility exists in the Medical Devices Regulation
107 which imposes a duty to provide financial security for manufacturers of devices within the scope of the Regulation.
108
Another problem that precautionary measures may pose is that the rules applicable thereto depends on the law governing the insurance contract. Accordingly, the control of moral hazard by insurers will depend on what consequences are attached to the breach of the precautionary measures as per the wording of the relevant policy, and any legislative rules that would be applicable to the clauses. The lack of harmonisation of the rules applicable to precautionary measures therefore stands as a hurdle which could obstruct the very aim of precautionary measures, i.e. to achieve deterrence: in jurisdictions where such clauses are strictly regulated and can be invalidated relatively easily, insurers would have to carefully draft their clauses so as not to lose the protection sought by their inclusion in the policies. Otherwise this would lead to the provisions not being applicable and lifting off the pressure on producers for observing safety standards. The diverse regulation of rules applicable to precautionary measures may further endanger the proportionate distribution of demands for product liability insurance among the insurance markets. This would notably beg the question of whether initiatives towards the harmonisation of insurance contract law principles such as the Principles of European Insurance Contract Law (PEICL)
109 could be an appropriate solution to this problem.
110
Another issue is to what extent precautionary measures would disturb victims’ rights against insurers where a right of direct action is granted to them for losses suffered from defective products. The obvious legal problem would lie in whether or not the outcome of any breach of precautionary measure by the producer (e.g. termination of the contract by the insurer, non-payment of any subsequent loss etc.) could be raised as a defence against the third party victim. As mentioned in the previous paragraph, the answer to this query would also depend on national law rules unless it is regulated at the EU level to avoid the risk of no compensation of third parties.
111 Such regulation would naturally be in the favour of victim protection, yet it could also be the subject of criticism by economists who would stress that the main aim of insurance is to cure the risk of underdeterrence and to remove risk from the injurer,
112 as opposed to principally protecting the victims.