key: cord-0961377-jz47whwa
authors: Sætra, Henrik Skaug
title: The foundations of a policy for the use of social robots in care
date: 2020-09-06
journal: Technol Soc
DOI: 10.1016/j.techsoc.2020.101383
sha: e53877fabd5421699cbe371066df2450a7d51eaa
doc_id: 961377
cord_uid: jz47whwa

Should we deploy social robots in care settings? This question, asked from a policy standpoint, requires that we understand the potential benefits and downsides of deploying social robots in care situations. Potential benefits could include increased efficiency, increased welfare, physiological and psychological benefits, and experienced satisfaction. There are, however, important objections to the use of social robots in care. These include the possibility that relations with robots can potentially displace human contact, that these relations could be harmful, that robot care is undignified and disrespectful, and that social robots are deceptive. I propose a framework for evaluating all these arguments in terms of three aspects of care: structure, process, and outcome. I then highlight the main ethical considerations that have to be made in order to untangle the web of pros and cons of social robots in care as these pros and cons are related the trade-offs regarding quantity and quality of care, process and outcome, and objective and subjective outcomes.

Should we deploy social robots in care settings? In this article I search for the policy desiderata for social robots in care. This endeavour is clearly related to, but also distinct from, the question of how social robots in care should behave. It is also related to fundamental questions of human-robot interaction, and such questions as what happens when we relate socially to robots. By drawing on the literature on social robots, political theory, and ethics, I provide the grounds for policy formation on the use of social robots in care.

On the one hand, there are many potential benefits to be had from a social robot. Economists and politicians see important demographic changes looming, which require a new supply of carers for the increasing population of elderly [1] . Robotic engineers see the benefits of providing assistance, entertainment, and companionship for people by equipping robots with social artificial intelligence, and researchers in the field of care have also found evidence of benefits, both physiological and psychological, of using social robots in care situations [2] .

On the other hand, there are potential negative consequences of deploying social robots in care. First of all, they have the potential to displace human beings, and human beings may in fact be an important factor in providing good care. Social robots that relate to human beings may also be harmful if the relations they form with human beings leave the humans worse off. Finally, it can be argued that any benefit that is observed in the use of social robots is achieved through deception and undignified treatment of the cared-for.

In order to evaluate these pros and cons, we need a better understanding of how social robots impact the quality and quantity of care. I use and modify a framework for assessing the quality of care though the examination of the structure, process, and outcome of care [3] . I show how social robots have both positive and negative consequences for all three aspects of the quality of care and how a basic ethical framework can be used to untangle the various effects of social robots on the quality of care.

I show that the answer to the first question in this introduction depends on whether or not we emphasise the outcome or process of care, and whether or not we decide to focus on individual recipients of care or the general level of care in society. We also have to consider whether those being cared for are the best judges of what a good outcome of care entails.

In sum, I show that the first question is deeply political and can best be answered by the tools of political philosophy and ethics. I share Coeckelbergh's [4] concern about implicit assumptions at play in the analysis of technology as well as his suggestions that political theory has much to offer in debates about foundational political principles. The answer to the question of whether or not or how to use social robots in care situations is dependent on the definition of a good society and the role of technology in such a society [5] . This is an attempt to highlight the main ethical and political issues involved for a broader audience, in particular those involved in politics and the formation of healthcare policy.

Robots in care has garnered attention and some commercial success over time, partly because of certain demographic developments observed in modern societies [6, 7] . For example, people grow older, and many of the elderly face challenges such as dementia and a general need for care [1] . Upholding the ratio of carers to elderly is simultaneously perceived as a great challenge, and it is hoped that robots can meet some of the needs created by a shortage of carers [8, 9] . According to Fong, et al. [10] , social robots will, for example, 'assist in health care, rehabilitation, and therapy'.

The social robots I discuss in this article are created to interact with people and fulfil certain functions that people normally seek in human or animal social companions. Robots, it is argued may require social functionality in order to operate properly when integrated in human social settings -particularly when human-robot interaction and cooperation is the goal [11] . When the term social robot is used in this article, it refers to robots that are socially receptive but not sociable robots, as the latter are considered to have 'their own internal goals and motivations' according to Breazeal [12] .

While social robots have a physical presence and can be endowed with the ability to provide physical assistance such as lifting, holding, feeding, etc, I will only consider the social aspect of such robots. Socially assistive robots are defined by Feil-Seifer and Mataric [13] as the intersection between socially interactive robots and assistive robots. The aspect of robots here analysed can include cognitive, but not physical, assistance, which means that I do not discuss socially assistive robots in their entirety. The type of social robot in question cannot manipulate objects or provide any form of physical assistance. It is not a lifter, washer, feeder, mover or butler. Neither will it be equipped with sensors and other capabilities primarily intended to assist patients with physical monitoring and assistance [14] . While the robot cannot assist physically, it can provide emotional and cognitive assistance [15] . These three forms of assistance are what Johansson-Pajala, et al. [16] consider when they speak of care robots, and the physical aspect is excluded from the current analysis. According to their definition of a personal care robot, which specifies 'robotics that improve the quality of life of humans, on a non-medical basis', the social robots included in the analysis could still be considered such robots [16] . Examples of robots that together fulfil the same functions as the social robots here analysed are Buddy, Paro, Leka, and Cutii [17] [18] [19] [20] .

I consider the type of robots discussed by Sharkey and Sharkey [21] and Bradwell, et al. [7] , which provides companionship in addition to basic cognitive assistance. Included in the analysis are current or near-future social robots, and their intelligence is as a consequence limited. They can interact rather convincingly with other beings, primarily through spoken and body language. There is, however, no strong AI or general intelligence present, and these robots are not assumed to understand anything in the human sense of the word [22] .

Coeckelbergh [23] and Sparrow [24] have both described thought experiments concerning robot eldercare. They describe counter-factual scenarios where the elderly are cared for in facilities where robots perform all necessary duties. This article is an attempt to identify the core ethical issues invoked by the use of social robots in care, and to develop the policy desiderata that follow from the consideration of these issues.

The question, for those developing policies regarding the use of social robots in care, is whether or not social robots should be deployed. An examination of the pros and cons is in order -which arguments can be made for and against the deployment of social robots?

A premise for considering the deployment of social robots is that it must be based on the provision of a tangible benefit. There are two main categories of such benefits, namely the increase in the quality or efficiency of care [25] . I will consider both categories, each of which contains several different arguments. I must also note that in the following I examine findings from multiple types of robots (animal-like and humanoid) as well as various types of patient groups (elderly, with or without dementia, the mentally ill, etc.). This is not to suggest that findings are fully transferable between these categories, and I simply use this examination to build a strong case for potential benefits of deploying social robots in care situations.

Quality of care is an issue often discussed but not quite as often defined. For example, Johansson-Pajala, et al. [16] speak of the need to 'facilitate the acceptance and normalisation of care robots in elder care for exploiting the benefits such as increased quality in elder care and improved independence in everyday life'. The fact that robots can improve quality in eldercare is taken as a given, but we must delve a bit deeper into both the concept of care and quality.

Care involves situations in which someone is dependent on assistance from others for various reasons, particularly in old age. The provision of care is often assumed to be based on a moral claim, either involving individuals close to a person or society at large [26] .

Many definitions of care have been attempted, some very broad and some very specific [26] . One definition by Bubeck [27] states that '[c]aring for is the meeting of the needs of one person by another person, where face-to-face interaction between carer and cared-for is a crucial element of the overall activity and where the need is of such a nature that it cannot possibly be met by the person in need herself'. Care, then, is based on the presence of certain needs-'those which make us dependent on others' [27] . Bowden [28] further states that caring 'expresses ethically significant ways in which we matter to each other, transforming interpersonal relatedness into something beyond ontological necessity or brute survival'.

Such definitions illuminate the problem of the very concept of a care robot. The latter definition would, I argue, completely rule out the possibility of care robots, but the first may be amended to allow for the provision of robot care. If we simply state that care involves the meeting of the needs of a person but skip the demand that the carer be a person, robots can care. Furthermore, Bubeck [27] states that face-to-face interaction between the carer and the J o u r n a l P r e -p r o o f cared-for is crucial, and I will assume that robots are able to interact face-to-face, while recognising that this is a controversial assumption. 1 Meacham and Studley [31] similarly ask if a robot can care. They believe that it can, as they see care as based on the external rather than the internal aspects of the carer and the environment.

In developing policy desiderata for the use of social robots, I begin with Donabedian [3] and his three main categories for evaluating quality of care: structure, process, and outcome, as shown in Figure 1 . Figure 1 : Aspects of quality of care [3] Structure involves human and physical resources, such as facilities and staff. Process involves an examination of how care is both attained and provided. Outcome encompasses measures of the health status after care is provided, which involves both objective measures of health and patients' knowledge, behaviour, and satisfaction. This framework lets us evaluate the achievement of good outcomes through bad processes, good process that do not have the desired outcome, and both objective and subjective evaluations of the achieved outcomes. We will now turn to some of the potential benefits of social robots in care, as discussed in the literature on social robots.

In a systematic study of the effects of effects of 'socially assistive robots', Bemelmans, et al. [25] found positive effects on physiological factors, such as stress levels, in the cared-for. This is presumably achieved through the same kind of mechanism as those involved in animal-assisted therapy (AAT). Such therapy has been shown to calm patients and have various other positive effects [32] .

Furthermore, the cognitive assistance provided by a social robot could, for example, lead to a healthier lifestyle by way of assisting in healthy eating, suggesting activities, managing medication, etc. Such factors could lead to objective improvements in care outcome. Patients could become healthier, which could be considered an objective improvement in care outcome.

A direct consequence of improved physiological health could be improved psychological health. This would be an indirect benefit of social robots. There are, however, indications that social robots could also improve psychological health directly.

Bemelmans, et al. [25] found positive effects from socially assistive robots on several psychological factors, including mood, loneliness, social connections, and communication. These benefits are also associated with AAT, and one of the selling points of social robots, such as Paro the robotic seal, is that they do provide the therapeutic effects of animal-assisted therapy [18] . The heralded benefits of such robots are great, and includes stress reduction, stimulation of interaction between carers and the cared-for, improved relaxation and motivation, and improved socialisation [18] . Furthermore, Moyle, et al. [33] show that companion robots increased the pleasure scores of the cared-for. This means that a potential benefit of social robots is that they make the caredfor happier in the sense that they report higher experienced pleasure. If the care we provide causes people to report high satisfaction, that is one of the subjective outcome factors involved in evaluating the quality of care.

I will also note that we need not assume that human (or animal) therapy is the measure to which care by robots must aspire. In an examination of potential benefits of social robots, we must accept the possibility that robots with advanced artificial intelligence may provide better care. This could be the case if the cared-for preferred not to expose their vulnerabilities to other human beings. Some might, for example, prefer to have a robot monitor them, for example in the shower and in the bathroom, as opposed to having a human caretaker [34] . Another possibility is that robots, armed with advanced artificial intelligence, could provide more advanced and effective care than human caretakers could.

While robot help with physical tasks is not considered, robots can provide cognitive assistance. This means that they can provide guidance regarding the proper time for and amount of medication, they can assist in nudging the cared-for towards a somewhat more active lifestyle, they can remind the cared-for who their family members are, where they left important items, etc.

This kind of assistance means that a social robot can empower, and 'facilitate and enhance older people's autonomy' [24] . It would let the cared-for be more independent (of other people) than they would otherwise be, and it could also allow the cared-for to 'age-in-place', which could be beneficial both for the cared-for and those who pay for the alternativeinstitutionalised care [35] . In addition, they can perform safety functions such as making sure that the oven is turned off, and even monitor and report if the cared-for should fall, have a heart attack, etc. The latter functions, however, are not part of the social aspect of social robots.

Care is costly, no matter how we look at it. On a social level, it requires vast amounts of resources, both human and material. It is also necessary, but most tend to agree that more efficient care, ceteris paribus, is preferable to the opposite. If not for cynical reasons and the quest for profit, increased efficiency would allow us to provide more care.

One claimed benefit of care robots is that they could provide care at a lower cost. Sparrow and Sparrow [36] are sceptical of such claims, and point out that robots with the required sophistication to perform the duties promised are quite expensive. However, let us assume that social robots without physical assistance capabilities can be produced relatively cheaply J o u r n a l P r e -p r o o f now or in the near future. If we were to achieve a slightly reduced demand for human care by deploying such robots, we will assume that this leads to monetary efficiency.

All sorts of things may be wrong with the all-robotic facility imagined by Sparrow [24] , but we cannot automatically assume that the care provided in such facilities would be worse than the alternative. One of the main reasons for pursuing economic efficiency is that we will be able to provide more care for less money. This means that poor societies could provide care for those that would otherwise not be cared for, and even in rich societies care could be provided for more people for longer periods of time and in new situations [37] .

We cannot assume that the sole goal of making care more efficient is either to increase profits from care or use the money saved from care in other sectors of society. Increased monetary efficiency is thus considered to be a benefit.

Another potential benefit of social robots is that we might change the way care is provided. Social robots need not replace human caretakers, but they could assist them [38] . While social robots provide certain tasks, human caretakers could provide other tasks for which they previously did not have time. Care could also be provided faster if humans and machines cooperated on the performance of tasks.

Another benefit is seen in situations in which robots have clear and obvious advantages over human beings. One such settings would be in hazardous situations, for example in poisonous or radioactive settings. In a hypothetical setting where care would either have to be provided in such a setting or not at all, a social robot would obviously provide benefits. This would be akin to the robots that disarm bombs-taking advantage of their immunity, durability, and expendability in order to perform tasks that humans either cannot or will not perform.

In the COVID-19 situation in 2020, the elderly were particularly at risk and were isolated in care facilities in large parts of the world. Only their caretakers could enter the facilities, and while necessary, this also involves the risk of infected caretakers bringing the virus into the facilities. A social robot alone would not fix this situation, as it could not provide any physical assistance. However, it could provide companionship at a time of particular need, in a situation where the cared-for were not allowed to meet their friends and relatives.

We have seen that there are several potential benefits to be had from social robots in care. As to the structural aspects of quality of care, increased efficiency, and particularly new opportunities derived from such robots' unique nature could lead to increased structural quality of care. This is because structural conditions are improved by freeing up resources, opening up for new use of existing resources, and creating new arenas of care.

The main advantages, however, are related to the outcome aspect of care. Both objective and psychological factors could be improved by deploying social robots. Objective and subjective health status could be improved. These could, but need not, be associated.

Lastly, we could also see improvements in the process aspect of care. One possibility is that robots replace human caretakers in situations where the cared-for feel particularly vulnerable, but also through more efficient care in terms of the access to care and the speed of the provision of specific tasks. All aspects of quality of care could potentially be positively impacted by the deployment of social robots.

Before we decide to deploy social robots, we must also examine the possible negative consequences of using such robots in care settings. As with the benefits, I here consider potential negative consequences drawn from the literature on social robot and robots in general, and I do not limit the discussion to the use of robots in care. This approach provides us with a broad arsenal of possible objections considered relevant for the use of social robots in care settings.

Two main concerns related to human-robot interaction are deception and authenticity [38] . In addition to these, there are questions of dignity, respect, and recognition [21, 24] . Stahl and Coeckelbergh [39] consider the effects of healthcare robotics in general, which includes robots providing physical assistance. While they consider factors such as de-humanisation and deception, they focus more generally on issues related to the use of information technology and robotics, such as job loss and data ethics (including issues of privacy and data protection). My task is aimed at a specific subset of healthcare robotics; the more general issues related to information technology and robots are not considered here.

I therefore do not consider issues of machine ethics, understood as the examination of how robots should act as subjects [38, 40] . Machine ethics is of great importance for guiding and evaluating the development of social robots, and I support the call for more ethical involvement in innovation processes [39] . In this article, however, I search for the policy desiderata for social robot in care, and this endeavour is clearly related to, but also distinct from, the question of how social robots should behave. 2 

A robot is not a biotic being, and there is some concern about relations with social robots not being authentic or meaningful [36, 41] . The objection is based on two related, but different, concerns. The first is that relationships with robots may change human beings. The second is that human beings have a need for human contact and relationships, and that the introduction of social robots will crowd out human contact.

While the relationships between humans and robots have been studied for a long time, there is still a lack of certainty with regard to the consequences of forming relations with robots [21, 29] . Turkle [41] examines how easily we form social bonds with machines and how much we seem to enjoy them, but asks, 'What if a robot companion makes us feel good but leaves us somehow diminished?' This strikes at the core of the benefit of differentiating between objective and subjective outcomes.

This objection is also related to a purported need for human contact. One way of approaching this problem is to consider Maslow's hierarchy of needs and what he calls social or belonging needs [42] . Some also speak of a need to relate to others and that a lack of such relations leads to mental disintegration [43] . Such objections assume that robots cannot fulfil core companion skills, but there is still an ongoing debate about what authenticity really refers to and not least to the potential of social AI to develop into what we might perceive as true social companions [44, 45] .

A related factor is what Sparrow [24] calls the human need for recognition and respect, 'which robots cannot provide'. Recognition 'consists in the enjoyment of social relations that acknowledge us in our particularity and as valued members of a community', and respect 'consists in social and political relationships wherein our ends are granted equal weight to those of others in the community' [24] . The question of equal weight to each person's ends and goals is further discussed in section 6, and the concept of recognition is clearly also related to the question of authenticity.

Another perspective on the value of respect is the concept of dignity. Sharkey and Sharkey [21] write about the prospect of an erosion of the dignity of the cared-for, and relate this to the idea that it is undignified for people to be cared for and kept happy through a process that a person 'might have abhorred' had they had full or former mental capacity.

Some see those with dementia as passing through a second childhood and thus see this as reason to treat them like children [46] . Once more, we see that there is a fundamental concern about the difference between objective and subjective evaluation of outcomes. A person with dementia might be perfectly happy playing with a robotic seal-even believing that it is real-but some could argue that such an outcome would be bad, because this person is not their true self, and that such satisfaction should not be the measure used to evaluate the outcome. This can be related to Berlin's distinction between an empirical and an authentic self. The former is the person as they appear today, while the latter is some potential version of this person [47, 48] . The problem, then, is who should be considered when evaluating outcomes, the actual cared-for or some potential cared-for as interpreted by someone else.

Carers and relatives are also potential stakeholders worthy of consideration, and research shows that they tend to see therapy with dolls, for example, as 'demeaning, patronizing and inappropriate' [49, 50] . However, studies show that other relevant stakeholders (children and young adults considering the use of robots in care, and specifically in the context of elderly family members), have relatively few qualms about using companion robots in care [7, 51] . In a scoping review of research related to the use of PARO in care, Hung, et al. [2] argue that the literature contains a gap with regard to user (clinicians, policy-makers, families) and patient perspectives on the use of social robots in care.

Another concern about the use of social robots in care is deception. One possibility is quite simply that a person with dementia may not be able to distinguish a relationship with a robot from a relationship with a living person [21, 36] . Another possibility is that the cared-for knows that a robot is not a living being, but still cannot help reacting to it as if it were [52] . 3 Deception is a term defined and used in various ways, and here I follow Williams [54] in seeing it as a) a breach of trust by the deceiver and b) manipulation by the deceiver of the deceived. However, we also need to identify the actors involved when discussing robot deception. I assume that a robot itself does not have agency and that it cannot be morally responsible for its actions. Someone else owns the actions of a robot, and it is this someone who deceives if a robot can actually be used for deception. I thus consider the perspective of the cared-for, and whether or not they are able to perceive the world around them accurately. The deceived and not the deceiver is thus in focus. The intention and even the identity of the deceiver is now of less importance.

Robot deception is interesting both because it can be considered intentional on behalf of the creators of social robots and because it is more effective than (but not necessarily qualitatively different from) the anthropomorphism of other non-animate objects.

A crucial point is that deception cannot be considered universally condemnable unless we simultaneously agree to abandon all sorts of games, theatre, movies, etc. Our societies are based in large part on the idea of benevolent or prosocial deception [55, 56] . Still, we might decide that robot deception is not desirable or not desirable in the domain of care. However, studies show that robot deception is beneficial in terms of achieving higher levels of cooperation between humans and bots in games [57] . Traeger, et al. [58] also show that robot deception can be used to improve human relationships and interaction by strategically making robots act 'vulnerable' and insecure.

While deception is mentioned here as a concern, several authors point out that social robot deception may in fact be beneficial and even a moral imperative if it leads to good consequences [59, 60] . Meacham and Studley [31] note that human carers will also often be somewhat deceptive, and Isaac and Bridewell [61] note that robots need to deceive in order to be beneficial. They emphasise the fact that much deception is both benign and pro-social.

A final argument against the use of social robots is based on the idea that this will lead to a situation in which robots displace human caretakers and eventually land us in a dystopia in which the elderly are taken care of by machines alone [24, 37] . This is a slippery-slope argument that will not be assumed to be a logical necessity. I will consider it possible for social robots to be introduced into care only where they are beneficial, and agree with the idea that automation is as much about enabling humans to work more effectively as replacing human beings directly [38] . Concerns related to the displacement of humans by robots and the policy implications of this possibility will not be considered in detail here [62] . However, viewing social robots as additions to the care settings, and not replacements for humans, require us to provide both the resources and education necessary for the care sector to properly deploy and integrate social robots in the care setting [6, 62] .

Summing up, the negative aspects of deploying social robots in care involve all three aspects of quality of care. First, it may lead to structural changes with fewer human beings and more machines. This change, however, is only negative if we simultaneously consider and evaluate the nature of social robot care. Second, the process of being cared for by a robot is highly objectionable to those who view human-robot relations as either a) insufficient or b) harmful. Similarly, the process of social robot care may involve undesirable forms of deception. Some consider the process both demeaning and inhumane. Third, regarding outcome, the effects of human-robot relations could be that we are diminished [41] . If human contact is in fact a basic human need, the increased use of robots in care may lead to the deprivation of basic needs, and mental disintegration [43] . I will note that the outcome-related negative effects of social robots are largely hypothesised, based on anecdotal evidence, and somewhat vague [29] .

From the preceding discussion, it follows that all aspects of quality of care, as derived from Donabedian [3] , can be argued to be both positively and negatively affected by the use of J o u r n a l P r e -p r o o f social robots. Untangling this web of pros and cons requires the introduction of an ethical framework for assessing and weighing the various arguments-a task to which I shortly turn. Before doing so, however, I propose a modification of the model for evaluating quality of care. First, I propose that the quality factors must be evaluated per person, and that we separate structure from the other two, as this aspect is closely connected to the quantity of care and should be considered as such, also on a societal scale. This separation makes it easier to illustrate the balancing act that is care policy. The new model is shown in Figure 2 . It is time to examine the core ethical concepts required to evaluate the various issues raised thus far. This examination will by necessity be brief, and the main idea is to highlight the various concerns that need to be accounted for in a policy for social robots in care and in particular to elucidate the trade-offs involved.

Beginning at the end of care, I examine how we might approach the evaluation of the outcomes of care. The outcomes, as we have seen, consist of both objective and subjective factors. It thus becomes clear that it is not simply a matter of weighing process against outcomes. We must also decide how to weigh the various forms of outcomes.

Sparrow [24] proposes that we focus on objective factors of care, as people's own evaluations of their situations may lead us astray. In particular, people have desires and preferences that may be objectively bad for them, which leads him to argue in favour of a sort of paternalism. This assumes that someone else could know better than I do what is good for me and thus have the right to override my preferences. This might be 'hard' paternalism, but it is also the basis of 'soft' paternalism, like nudging [63] . Both Sparrow [24] and Coeckelbergh [60] argue in favour of evaluating outcomes by considering the capabilities of the cared-for-an objective approach [64] .

An alternative approach is to consider people's own judgements and preferences. This is the basic idea behind utilitarianism, which involves the quest for maximising the amount of happiness, when everyone's pleasure and pain is factored in [65] . Batayeh, et al. [66] discuss socially responsible innovation in relation to healthcare, and state that this involves moving 'beyond tradition ideas of patient wellness that are only physiological'. While they propose that we include a range of physical outcomes, such as financial wellness and overall community wellness, I here propose that we could consider the patient's subjective perception of wellness.

One particular kind of utilitarianism is hedonism. It comes in two forms: psychological and ethical. The first is a theory of human motivation, and explains human behaviour as our desire for pleasure (and aversion to pain) [67] . The other form, ethical hedonism, is a theory of the evaluation of what is good, and it is this form of hedonism we might apply if we want to emphasise the satisfaction of the cared-for, as measured by their happiness [68] .

If we assume that social robots do provide pleasure, as experienced by the cared-for, this is considered good in a hedonistic ethic regardless of how this pleasure came about. However, when hedonism is coupled with utilitarianism, the opportunity to deny a person increased happiness arises if this increase comes at the cost of the general level of happiness.

We must, however, note that utilitarianism and hedonism are related but not identical to consequentialism. We could regard criteria other than individual happiness as the yardstick with which to evaluate various outcomes, such as the capabilities discussed by Nussbaum and Sen [64] or some combination of subjective and objective factors.

One problem, as pointed out by all the sceptics of deception and inauthentic relations with social robots, is that such an ethic would also accept some outcomes that many of us would consider to be undesirable. Sparrow and Sparrow [36] , Sparrow [24] , and Saetra [37] invoke the example of Robert Nozick's experience machine. This is a machine that could provide us with as many life-like experiences we would like while we are on life support in some tank [69] . While not actually performing any actions in a physical sense, we would experience life as if we did, with the added benefit, of course, that everyone could receive experiences tailored to maximise pleasure and joy, as scarce goods, wars, difficult human relationships, etc. would not come in the way of happiness [37] .

The relevance of this example hinges on the assumption that deceiving the elderly with a robot is similar to the deception created by the experience machine. The subjective outcome may surpass anything achievable without such deception. The objective outcome, in terms of physiological health, for example, will be assumed to be at least equally good. Let us assume the elderly will be less stressed and more active due to their newfound happiness. The makers of an experience machine will certainly also be capable of providing physical stimulation in order to keep our bodies from withering while hooked up to the machine.

If satisfaction is increased and the objective outcome kept stable or even improved, what reason could we have to object to such a scenario? One answer is process. We could care about how the outcomes comes about regardless of the actual outcomes, and this takes us to the ethics of process.

How can the process lead us to say that an outcome that a person perceives as good is in fact bad? This is where we turn from consequentialist ethics focused on outcomes to deontological ethics of actions and process. For a consequentialist, the outcome can justify the means, but most deontologists will object to this.

Let us return to the example of the experience machine, which provides a good subjective outcome. Nozick [69] nevertheless states that being hooked up to such a machine is highly undesirable, akin to suicide.

The first reason for this is that we want to actually do things and not simply have the experience of doing them. This is similar to the objections based on the value of authenticity and deception. Outcomes and experiences are themselves merely by-products of that which is valuable, which is an authentic process where we actually live our lives and interact with human beings. However, if the experiences and memories that were produced seem real, what possible reason could we have to care about how they came about?

Nozick's second reason is that we have a desire to 'be a certain way, to be a certain sort of person' [69] . If an elderly person were to be deceived by a robot seal into believing they were having a social relationship with a living being, for example, or that they were the captain of a pirate ship, and derived great joy from this, they would still not actually be the kind of person that they desired. But, again, why should we care, if the experience is real to the person in question? Sparrow [24] would say that respect would be lacking and that a person deceived in such a way had been deprived of recognition. Such a life would, according to Sharkey and Sharkey [21] , not be dignified. People would be happy but they would not be treated as adult and competent beings-their happiness would be akin to the happiness produced in children when parents tell them that Santa is real.

Finally, Nozick [69] pre-empts the objections made by many opponents of the uncritical use of social robots, namely that the human-made experiences they produce are fundamentally different from deeper experiences. In order for this objection to be effective, however, we must be able to point out exactly how man-made (robot) experiences are inferior to experiences produced by natural, or living, entities. We do not object to all experience created or assisted with human-made tools. If we are to stop social robots in care merely because they are human-made, we need more knowledge of just how, if at all, relating to them is harmful. Turkle [41] and Darling [70] provide compelling reasons for caution, but the harms of relating to robots must be better understood and studied if they are to inform policy for social robots in care.

If we are to disallow outcomes that are experienced as good, we could rely on what Nozick [69] calls side constraints. These are comparable to what is usually called individual rights, which is the most popular line of defence against unbridled utilitarianism. By appealing to such constraints or rights, I could stop you from optimising social happiness if that happiness were achieved by breaking some inviolable right. Such rights could, for example, be the right not to be killed, coerced, or defrauded. We could also add the right not to be deceived and the right to be treated with respect. Such rights could be inviolable and not amenable to the calculus of utilitarianism (Nagel, 2013) . This is akin to the call for Kantian ethics in order to prevent the hedonism previously described [36] . This would imply that people should never be treated as a means to the happiness of others, but always as ends.

Unbridled hedonism and consequentialism might very well lead us to accept us all being hooked up to the experience machine, but inviolable rights not to be deceived or given due respect would also lead us into troubled water, for example by preventing hugely beneficial applications of new technologies because they infringe on some set rights [37] .

If we have a desire to create a society in which technology promotes a good society, we must determine how such a society treats individuals [5] . More than that, however, we must also allow for the fact that every society consists of multiple people, and any policy aimed solely at the individual runs the risk of having dire consequences for certain groups of individuals or even all individuals.

In general, it seems that social robots in care may have negative overall effects on the process of care. This is based on the objections which emphasise the beneficial nature of human contact for recognition, the need for respect, the value of what is natural (not man-made), and the general idea that deception is bad.

We have also seen that social robots can have beneficial effects on both objective and subjective health statuses. They may provide social interaction that is experienced as pleasant, which also stimulates sociability and stress reduction in those being cared for. We can imagine that use of social robots can lead to physiological and psychological benefits. However, we must not assume that such benefits would necessarily be maintained if robots completely replaced human beings in care. Let us assume that such a situation, as in Sparrow [24] , would also lead to negative outcomes, caused for example by social needs not being met.

Finally, the question of how social robots affect the structural dimension of care is debatable. Sparrow [24] argues that care robots are not a panacea that will save the economy of care. However, we have assumed that a pure social robot without physical assistance capabilities can be quite cheap and that its use will lead to overall economic benefits if it allows for a lower ratio of carers to the cared-for.

Is care of a slightly lesser quality, provided for more people, a good thing? Our definition of care describes the meeting of needs, and I will assume that the needs in question are nontrivial and that not meeting said needs will have substantial negative consequences. A policy must define the point at which people should receive care. Providing care for more people than those who are in need is not desirable, which means that we have a target value for the quantity of care. When this target value is reached, we should do whatever we can in order to provide the best possible quality of care while maintaining the same quantity.

In the following I will make three assumptions. First, that the financing of the health sector is static. Second, that those who are defined as being in need of care have a real need for care, and we cannot redefine these needs in order to change the target quantity of care. Third, that the society in question does not have the financial possibility to provide human care for all in need of care. These assumptions are of course highly debateable, but they will allow us to see how the use of social robots can be evaluated.

First of all, social robots are potentially cost-effective, which means that they can be employed in order to change the structural conditions of care, allowing us to provide care for more individuals. Let us assume that this means that we increase the amount of material care resources available and slightly reduce the amount of human care workers.

If this is the only way to provide care for all those in need, the only real alternative in this scenario is to state that no care is better than social robot care. Since we see potential outcome benefits from the use of social robots, this statement will not be accepted.

After we have used social robots to provide care for all those in need, we must decide how best to deploy them and whether or not we should deploy more of them. If we assume that a social robot is not better than human caretakers, the main reason to introduce more social robots in care would be to rationalise the care sector and reduce the amount of money spent on care in general, but this is ruled out by the assumption of static financing. At this point, we are left to debate the relative benefits of human versus robot care.

We are, however, also at a point where the ethical principles involved in the use of social robots become clear. Employing social robots in care means that some who would otherwise receive human care, will receive care that is of slightly lesser quality. However, those who would otherwise not receive the care they need will receive much better care, even if that care is not optimal.

From a policy standpoint it makes little sense to refuse the use of social robots on the grounds that some individuals would be worse off than before. This would not be a case of using people as means for achieving the ends for others. If we consider the other alternative, not using social robots could mean that we used all those in need of care, who would now not receive care, as a means to an end for a lucky few who now received human care.

If we view policy formation as the choice among various scenarios for the future rather than the change of actual conditions today, we are in a much better place than if we have to consider the negative consequences for the privileged actual people of today. This is also where Rawls's original position and veil of ignorance become useful [71] . Without debating Rawls's premises and what is hidden by his veil, we can imagine the original position quite simply as a situation in which policy makers meet, without knowing who they themselves are, in order to make policy for the use of social robots in care based on the desiderata established thus far, as shown in simplified form in Table 1 . The outcome of such a process depends on an evaluation of the various benefit and drawbacks listed, and this evaluation is necessarily performed in light of some theory of ethics. It will often be an implicit system of ethics, but reaching an agreement on which policy is best will be very difficult unless we make the ethical assumptions we employ explicit.

The advantage of the thought experiment that is Rawls's original position is that alternative scenarios can be considered without considering the loss of actual persons. In such a situation it is obvious that we will not be able to make any reasonable decision without having some concern for the overall utility and consequences of our decision. Choosing a scenario where only the very best level of care is provided would necessarily lead to a lack of care for a vast number of persons, and there are no ethical systems that I know of that would suggest such a choice.

Finally, it is important to note that my focus on what Held [26] calls the 'dominant moral theories' and abstract reasoning is not universally accepted as a beneficial approach to the ethics of care or the ethics of healthcare robotics. Held [26] herself emphasises the claims of particular others, and believes that abstract and universalistic rules will lead us astray. She further states that for 'most advocates of the ethics of care, the compelling moral claim of the particular other may be valid even when it conflicts with the requirement usually made by moral theories that moral judgments be universalizable, and this is of fundamental moral importance' [26] . Preston and Wickson [72] also emphasise how focusing on, for example, the feminist perspective and care ethics will lead to a different understanding of the impacts of technology than traditional approaches. This, they argue, is also compatible with a range of other ethical frameworks. Stahl and Coeckelbergh [39] similarly suggest that abstract philosophical approaches to the ethics of healthcare robotics is problematic, as these are too far from the practical aspects of innovation of design. I accept these objections to my approach, but emphasise that my focus on policy formation makes an abstract and philosophical approach to the ethical aspects of social robots in care necessary.

Social robots in care have the potential to result in both positive and negative consequences for care. These consequences have important implications for the ethical evaluation of the use of social robots and for the formation of a policy of social robots in care. I have presented Donabedian's [3] system of the three aspects of quality of care, which was modified to become a system for evaluating the quality and quantity of care. Structural aspects of care are seen as mainly related to the quantity of care while process and outcome are seen as core components of the quality of care. It is, of course, possible to state that much care is quality care, and the reader can easily make such a move without changing the gist of my argument.

The main potential benefits of social robots in care are related to structure (efficiency) and outcome. Social robots have the potential to increase care efficiency, thus enabling the provision of more care. They also have the potential to improve both physiological and psychological factors and the satisfaction of the cared-for.

On the other hand, the process of care is potentially negatively affected by the deployment of social robots, as they may reduce the amount of human contact. Robot care may also be undignified and disrespectful, and it involves a form of deception. In addition, relations with robots may be harmful, which is a factor of the outcome of care.

In evaluating these outcomes, I have made certain assumptions in order to highlight the most important trade-offs that must be made by policymakers. The major assumption is that the financing of the care sector is static. This implies that we cannot simply sidestep difficult choices by simply allocating more money to the sector in order to advocate against the deployment of social robots. Furthermore, the social robots I consider are potentially cost effective.

First, this means that one potential trade-off is some loss of quality in order to provide a larger quantity of care. The use of social robots may decrease the quality of care, but if this means that more people will receive care, this can easily be imagined to be an acceptable trade-off.

Secondly, policymakers must decide how to evaluate process versus outcome. The evidence suggests that social robots may provide outcome benefits at the cost of an objectionable process. Should policymakers decide that a good process and a strong focus on individual rights to respect, dignity, etc. is most important, they may have to accept worse outcomes.

Finally, a crucial question is whether or not to accept people's own evaluations of their situations or if we have a desire to overrule these in order to improve objective outcome gains.

Social robots in care thus have the potential to influence the quantity and quality of care, and policymakers should thoroughly examine the various aspects presented here. However, the policy desiderata in themselves are not sufficient for the formation of policy, as a system of ethics is a necessity for making the described trade-offs. This could be an implicit and unacknowledged system of ethics. While it could, it seems clear that making this system of ethics explicit will make the process more transparent, and by consequence possibly more legitimate.

Robot technology for future welfare: meeting upcoming societal challenges-an outlook with offset in the development in Scandinavia

The benefits of and barriers to using a social robot PARO in care settings: a scoping review

The quality of care: how can it be assessed?

Technology and the good society: A polemical essay on social ontology, political principles, and responsibility for technology

Technology and the good society

Social care technologies for older people: evidence for instigating a broader and more inclusive dialogue

Ethical perceptions towards real-world use of companion robots with older people and people with dementia: survey opinions among younger adults

Overcoming carer shortages with care robots: Dynamic value trade-offs in run-time

An Overview of Machine Medical Ethics

A survey of socially interactive robots

Getting to know each other-artificial social intelligence for autonomous robots

Toward sociable robots

Defining socially assistive robotics

Can connected technologies improve sleep quality and safety of older adults and care-givers? An evaluation study of sleep monitors and communicative robots at a residential care home in Japan

The use of socially assistive robots in the design of intelligent cognitive therapies for people with dementia

Care Robot Orientation: What, Who and How? Potential Users' Perceptions

Meet Buddy

PARO Therapeutic Robot

Help exceptional children live exceptional lives

Cutii

The eldercare factory

Rebooting AI: building artificial intelligence we can trust

How I Learned to Love the Robot': Capabilities, Information Technologies, and Elderly Care

Robots in aged care: a dystopian future?

Socially assistive robots in elderly care: A systematic review into effects and effectiveness

The ethics of care: Personal, political, and global

Caring: Gender-sensitive ethics. London: Routledge

Robot rights

Ethics and infinity

Could a robot care? It's all in the movement

Animal-assisted interventions for elderly patients affected by dementia or psychiatric disorders: a review

Exploring the effect of companion robots on emotional expression in older adults with dementia: a pilot randomized controlled trial

Granny and the robots: ethical issues in robot care for the elderly

The ethics of socially assistive robots in aged care. A focus group study with older adults in Flanders

In the hands of machines? The future of aged care

First, They Came for the Old and Demented

Ethics of Artificial Intelligence and Robotics

Ethics of healthcare robotics: Towards responsible research and innovation

Towards a principle-based healthcare agent

Alone together: Why we expect more from technology and less from each other

Motivation and personality. Delhi: Pearson Education

Escape from freedom

Human-Computer Interaction Problem in Learning: Could the Key Be Hidden Somewhere Between Social Interaction and Development of Tools?

The Parasitic Nature of Social AI: Sharing Minds with the Mindless

Evidence and mechanisms of retrogenesis in Alzheimer's and other dementias: management and treatment import

A measure of freedom

Psychiatry: A journal of the psychiatry of late life and allied sciences

A pilot study on the use of dolls for people with dementia

Children's perceptions of social robots: a study of the robots Pepper, AV1 and Tessa at Norwegian research fairs

Designing sociable robots

Robot Betrayal: a guide to the ethics of robotic deception

Truth & truthfulness: An essay in genealogy

Lying and deception in human interaction

Prosocial lies: When deception breeds trust

Behavioural evidence for a transparency-efficiency tradeoff in human-machine cooperation

Vulnerable robots positively shape human conversational dynamics in a human-robot team

The Paro seal robot: demeaning or enabling

Care robots and the future of ICT-mediated elderly care: a response to doom scenarios

White lies on silver tongues: Why robots need to deceive (and how)

Robotics and conservation of human resources

The politics of libertarian paternalism

The quality of life

Utilitarianism and other essays

Socially responsible innovation in health care: Cycles of actualization

Happiness and goodness: philosophical reflections on living well

The Cambridge dictionary of philosophy

Anarchy, state, and utopia

Anthropomorphic Framing in Human-Robot Interaction, Integration, and Policy

A theory of justice

Broadening the lens for the governance of emerging technologies: Care ethics and agricultural biotechnology

J o u r n a l P r e -p r o o f Highlights • A system for evaluation the quality of care as structure, process and outcome is used for evaluating the consequences of using social robots in care • Untangling the effects of social robots in care requires us to distinguish between outcome and process, and between objective and subjective outcomes • A set of ethical and political theoretical frameworks are discussed, and the author argues that making philosophical assumptions explicit are essential for policy formation on the issue of social robots in care