According to a number of authors (e.g. Matthias 2015; Sparrow and Sparrow 2006; Sparrow 2002; Wallach and Allen 2009; Sharkey and Sharkey 2011), the development and creation of social robots often involves deception. By contrast, some have expressed doubts about the prevalence of deception in robotics (e.g. Collins 2017; Sorell and Draper 2017). It seems that there is disagreement in the field about what counts as deception, and whether or when it should be avoided.
A primary aim of this paper is to make the case that, despite some claims to the contrary, there is deception in social robotics. The idea that there is deception is sometimes rejected on the basis that the developers and programmers of social robots may not have intended to deceive. In an effort to counter this resistance, we examine definitions of deception, and argue that deception can still occur even in the absence of conscious intention. If a person believes that a social robot has emotions and cares about them, they are being deceived: even if no-one explicitly intended that belief. An important first step to preventing the harmful consequences of deception in social robotics is to recognise that it can and does occur. Once this is recognised, an important next step is to consider what harmful effects it could have. We identify potential examples of such effects and follow this with a discussion of who could be held responsible, and whether such consequences could be prevented.
download No Robots for Old Men
Download: https://urlcod.com/2vIGDo
These examples show that both non-human animals and humans can be subject to deception without any intentional attempt to deceive. They support our claim that a deception can be said to have occurred in robotics if the appearance and the way that a robot is programmed to behave, creates, for example, the illusion that a robot is sentient, emotional, and caring or that it understands you or loves you. These are all deceptions because, at present and for the foreseeable future, robots are machines that are not alive, sentient or capable of feelings. Yet none of these deceptions necessarily require a conscious intention on behalf of robot manufacturers or designers who may simply have meant to entertain.
At the other extreme, Sparrow (2002) argues that the self-deception involved in an imaginary relationship with a robot is inherently wrong and violates a duty to see the world as it is. Presumably his position would also apply to the treatment of people with dementia, even though some deceptions (such as offering them a robot seal pet to care for) might alleviate their distress and anxiety. He also argues that those who design and manufacture robots that encourage such beliefs are unethical.
It is possible to identify two kinds of risk that could result from the development and presentation of social robots that appear to have emotions and to be able to understand and care about humans. We consider in turn (i) those that stem from the deception involved in robots that appear to have emotions and to care for us; and (ii) those that originate in over-estimations of the ability of robots to understand human behaviour and social situations.
There is evidence that adults form attachments to robots even when they are remote controlled bomb disposal robots (Carpenter 2016) or robot vacuum cleaners (Sung et al. 2007). When the robots are furry robot pets, or cute looking humanoids, attachment formation is even more likely. Such emotional attachments could have negative consequences for vulnerable adults such as those with dementia or other cognitive limitations (Sharkey and Sharkey 2012; Sharkey 2014). They might choose to neglect their relationships with fellow humans to focus their emotions and attentions on the robot instead. They could become anxious and concerned about their robot companions. In addition, robot companions which give rise to the deceptive illusion that they care and understand, could result in a reduction of contact with other human beings for vulnerable individuals. Friends, family, and care providers in general, might come to believe that the social and attachment needs of an older person were being met by a robot companion or pet, and as a consequence might reduce the time they spent with them.
The creation of a robot, or computational artefact, that encourages the belief that it is able to make moral decisions is particularly concerning. There is a growing awareness of the risks associated with algorithmic decision making by machines that are trained on big data, but which have no understanding of the meaning or effects of their decisions (Sharkey 2018). An extreme example of algorithmic decision making can be found in the example of lethal autonomous weapons, where concerns are being raised about giving robots the power to make life or death decisions about who to kill on the battlefield (see the Campaign to stop Killer Robots website for arguments for a ban of such weapons). An example that is closer to home is that of the autonomous car, and discussions of its decisions about where to turn and who to hit in the event of an accident (Lin 2016). The appearance and behaviour of a social robot that creates an illusion of understanding could foster belief in its moral competence.
There are also risks that arise from being deceived into overestimating the abilities of robots, and inappropriately placing them in positions of responsibility, e.g. in charge of classrooms, or of the vulnerable in care homes. People can overestimate the strengths and competences of a computational algorithm, but the development of robots with humanoid appearances that appear to understand and respond to human emotions and needs, has the potential to be that much more compelling.
There is a particular need to recognise the risks for the more vulnerable members of society, who are often dependent on others to make decisions for them. The youngest and oldest members have a greater need for protections to be in place to limit their exposure to robots, to ensure their access to human companionship and care and to prevent their exploitation by social robots.
It may be impossible, even undesirable, to prevent all deception in robotics. As we have seen, some deceptions can lead to health benefits. There are also humanoid, and animal-like, robots that are entertaining: anthropomorphic entertainment devices date back to antiquity.
If robots cannot be held responsible for deception, could the users of robots be implicated? It is well known that humans are anthropomorphic about robots and other machines, or any inanimate objects with certain features. It is less clear that humans have much choice in this matter. An animated and apparently needy robot can be hard to resist. As pointed out by Turkle (2011), a robot that seems to require care and nurturing is particularly compelling. Vulnerable old people and young children may be especially susceptible to such anthropomorphic cues (Epley 2007).
Scheutz (2011) suggested two ways to minimise the negative effects of deception, but we do not think that either are likely to be effective. One suggestion is to legally require a robot to continuously remind the user that it is only a machine and has no emotions. But it is not clear that this would prevent people from forming an attachment to it, and it might interfere with the comfort and relief from anxiety that interactions with a robot pet may create for some people requiring it for therapeutic reasons. The second suggestion is to equip the robot with an emotional system or to develop robots with a sense of morality. This seems unlikely to happen in the near future. There have been attempts to program robots with ethical rules (Anderson et al. 2006; Winfield et al. 2014), but these are rules that lack universality. They apply only in very limited and highly specific contexts. For example, Anderson et al. 2006) programmed a robot to use a set of ethical rules to decide whether or not to remind a patient to take their medicine, and whether or not to report the patient for not taking their medicine. As argued by Sharkey (2017), there is as yet little indication that it will be possible to code a set of ethical rules that will work in all circumstances and in many contexts.
We suggest that a better approach would be to put the onus of proof on robot manufacturers and sellers. Just as in the pharmaceutical industry, they could be required to provide evidence that a given robot application would not cause psychological harm or derogate any human rights. Thus, certain sensitive applications such as the use of robots in care situations could have a default of prohibition unless convincing evidence was provided that demonstrated benefits to wellbeing. Sensitive applications would include those involving babies, young children, and vulnerable older people. Some form of quality mark could be established to indicate that the robot had passed a set of ethical checks. The Foundation for Responsible Robotics has recently carried out a pilot project on the development of an assessment framework for a quality mark for AI based robotics products (FRR pilot report, under embargo) The planned quality mark would assess robot products for the extent to which they adhered to a set of 8 principles for robots; namely (1) security, (2) safety, (3) privacy, (4) fairness, (5) sustainability, (6) accountability and (7) transparency and (8) well-being. The approach is promising and could be adapted to include consideration of the risks of deception in social robots as explored here.
Responsibility for wrongful deception can be attributable to the combined contributions of users, developers and marketers of robot applications. Preventing harmful deception is difficult, but our suggestion is for an evidenced quality or kite mark indicating that tangible efforts have been made to foresee, recognise and avoid likely negative consequences and demonstrating any claimed benefits. It is important to find ways to ensure that deception in social robotics does not lead to robots replacing meaningful human care, or to misplaced trust in decisions made by machines. 2ff7e9595c
Comentários