Non-consensual personified sexbots: an intrinsic wrong

All humans mentioned in this paper are adults of normal mental faculty; they are all persons (by whatever criteria we use to define personhood), but the sexbots are not. In future, it may be possible for roboticists to create artificially intelligent robots which could meet some philosophers’ criteria for personhood; love robots or ‘erobots’ are being developed (see Anctil & Dubé, 2019) and programmers are making huge leaps forwards in the creation of interactive and socially intelligent robots (Dautenhahn, 2007). Nevertheless, robotic persons do not exist at the time of writing; sexbots which are currently on sale commercially (4woods, 2018; Realdoll, 2019; TrueCompanion, 2019) have some rudimentary robotic functions, such as smiling, moving their heads, eyes and arms, and having simple conversations, but they fall short of philosophical definitions of a person. I write as if the same is true of the sexbots under discussion in this paper. When I say that a sexbot is lifelike, I mean it is lifelike in its looks, rather than in its cognitive capacities. Although one might be morally concerned about the ethical implications of creating a non-robotic sex doll which represents someone against their wishes, the intrinsic wrongness of making a sex robot is all the more pertinent because the creations have human-like robotic functions: robotics and artificial intelligence are improving rapidly, and so sexbots are becoming more lifelike in their behaviour and appearance, and less like mere inert sex dolls.³ The movements and conversational abilities of a sexbot contribute to its appeal, making it seem more lifelike and more similar to a human than a non-moving sex doll would be (Danaher, 2017a). Despite their humanlike features, the sexbots available for sale and under discussion in this paper are objects which cannot be morally wronged.

Because philosophical exploration of sexbots is still in its infancy, there is presently no terminology to differentiate between types of sexbot; it will be useful for me to separate sexbots into two broad types, which I call generic sexbots and personified sexbots. The two categories are exclusive but not exhaustive, and although there may be some sexbots which do not fit neatly into either category, differentiating between the two types of sexbot will be useful to my argument herein, and to future arguments regarding sexbots. I define the first type of sexbot (a ‘generic sexbot’) as one which has not been intentionally made to represent or resemble a specific person. The creator would simply choose a set of features such as blonde hair, an hourglass figure and so on, and synthesise these to form a sexbot. I define the other type of sexbot (a ‘personified sexbot’) as one which is intentionally a representation of a particular person (the ‘human subject’). In personified sexbot creation, the creator uses the human subject as a template, and ‘copies’ their features to form a sexbot which is intended to, and does in fact, look just like the human subject. I use the term ‘sexbotification’ to refer to the process of using someone’s looks as a template for creating a personified sexbot.

Consider a case where someone creates a sexbot of his imaginary ideal woman, but by coincidence the sexbot closely resembles someone he has never seen. It might be thought that this is an example of a personified sexbot, however it is not, as the sexbot is not a representation of that particular human subject. Hilary Putnam gives an example which helps to clarify why the creation of a sexbot which accidentally resembles someone does not count as a sexbot representing that person. Putnam’s example involves an ant which crawls on some sand and by chance, carves out something which resembles a caricature of Winston Churchill (Putnam, 2012, p. 1). Putnam rightly argues that the lines in the sand are not a representation of Churchill because there was not an appropriate accompanying disposition in the mind of the creator—in other words, the ant was not trying to draw a picture of Churchill (ibid 1–5). If an artist intentionally produces a caricature of Churchill, then even if the ant’s line in the sand is qualitatively indistinguishable from the artist’s work, we should still say that the artist’s picture is a representation of Churchill, but the ant’s is not (ibid). Analogously, suppose two men create sexbots, where Alan creates a generic sexbot of his imaginary ideal woman, and Bob creates a sexbot which represents a woman he knows—Cara. Even if by pure chance both sexbots are qualitatively indistinguishable from one another (and from Cara), we should still say Alan’s sexbot is a generic sexbot which does not represent anyone, while Bob’s is a personified sexbot which represents Cara. Although Alan’s sexbot serendipitously resembles Cara, this is not sufficient for us to say that it is a representation of her. This is because (analogous to the ant example) there was not an appropriate accompanying disposition in the mind of Alan: he was not trying to produce a sexbot which represents Cara. Thus, Cara has not been sexbotified by Alan, but she has by Bob.

I will now briefly distinguish between two ways in which an action can be considered to be wrong; it is important to delineate these ways because my argument focuses on only one way in which an act can be wrong, viz. intrinsic wrongness. According to moral philosophers, the wrongness of actions can be characterised in different ways: instrumental wrongness and intrinsic wrongness. Broadly speaking,⁴ consequential ethicists such as utilitarians claim that an act is instrumentally wrong if it brings about negative consequences (see Bentham, 2000, pp. 14–18). A second way of understanding what makes an act wrong is more deontological in nature; a prominent example of such a moral system is that outlined by Kant (1997). For Kantian ethicists, the rightness or wrongness of an act is determined by the nature of the act itself, meaning that an act can be intrinsically wrong even if it does not cause any harmful consequences. For example, stealing is said to be intrinsically wrong, even if no one ever finds out or is harmed by it. Instrumental wrongness and intrinsic wrongness are not mutually exclusive: someone might consider an act to be both instrumentally wrong and intrinsically wrong, such as murdering a child.

Several writers debate whether it is wrong to create or use sexbots (Danaher, 2017b, c, 2019; Di Nucci, 2017; Harvey, 2015; Levy, 2008; McArthur, 2017; Migotti & Wyatt, 2017; Richardson, 2016). These writers only discuss the potential consequences of creating what I am calling generic sexbots, but no writers that I am aware of address the creation of personified sexbots.⁵ This is what I attempt to do below.

I consider two scenarios to highlight the intrinsic wrong in creating a personified sexbot without consent, as compared to creating one with the consent of the human subject. The scenarios are:

In this section I outline why creating a personified sexbot is a form of sexual objectification; I will later establish that doing so without consent is intrinsically wrong. Objectification involves using a person as if they are an object rather than a person. Kant famously instructed us never to treat a person as “merely as a means” (Kant, 1997, p. 38). Kant’s claim is that we should respect the humanity of other people, and not use them solely for our own gain.¹⁰ More recently, this idea has been taken up by feminist philosophers who write about sexual objectification (Dworkin 1974; MacKinnon, 1987, 1989; Nussbaum, 1995, 2007). Drawing on the work of Kant, Nussbaum maintains that objectification is not always wrong, but it becomes wrong when it involves treating someone as a mere means to an end; this is the conception of objectification that I follow. Nussbaum outlines seven features which form a cluster concept of objectification. These involve treating someone as if:

Langton largely supports Nussbaum’s very useful list of seven features, but adds another three which she suggests are also salient features of objectification:

Langton’s first two features (although very similar to one another) help to distinguish sexual objectification from, for example, a burglar, who meets most of Nussbaum’s features, but whom we would not think of as objectifying his victim (Langton, 2009, p. 230). Langton’s ‘silencing’ bears some similarities to Nussbaum’s ‘denial of subjectivity’, but they are distinct: for example, I may treat a toddler as if he cannot speak, whilst still acknowledging his feelings and taking these into account.

For now, let us leave aside the issue of whether objectification is wrong, and consider solely whether sexbotification is a form of objectification, using Nussbaum’s and Langton’s ten features.¹¹ Firstly, sexbotification meets two of Langton’s features of objectification (Langton’s features no. 8 and 9) because it reduces the human subject to her body and her appearance. Indeed, because the types of sexbots that can currently be created are very lifelike in their looks but not in their cognitive capacities, the primary appeal of a sexbot lies in its human-like bodily appearance. A skilled sexbot creator can reproduce a person’s bodily appearance with astonishing similarity, but there will be little other similarity between the sexbot and the human subject. The human subject’s body and appearance are what is important when creating a personified sexbot: their personality, intelligence, interests, sense of humour, and other features are seemingly unimportant: a personified sexbot is a copy of a person’s bodily appearance, so it seems apt to say that sexbotification involves reduction to body and appearance to the senses. When a sexbot creator uses the human subject’s appearance as a ‘blueprint’ for the sexbot’s looks, this also seems to be a form of instrumentalisation (Nussbaum’s feature no. 1). Of course, we may say that a portrait artist also instrumentalises people in this way, and that such an act is prima facie morally permissible. But at this stage I am not commenting on the permissibility of sexbotification, merely establishing that it is a form of sexual objectification. We might also say that creating a personified sexbot involves treating the human subject as someone who lacks boundaries (Nussbaum’s feature no. 5) because the sexbot will be ever-consenting. The sexbot can also be bought or owned (Nussbaum’s feature no. 6). However, we might think that a sexbot creator can believe that a sexbot can be bought and has no boundaries, whilst recognising that the same is not true of the human subject whom it represents. Nonetheless, since the sexbot is a physical representation of the human subject, creating it evidently involves treating the human subject in a sexual way, just as a pornographic photographer or film director treats the actors in a sexual way by creating the pornography.

Given that Nussbaum’s and Langton’s features form a cluster concept of objectification, it is not clear how many of the features need to be met in order for something to count as objectification. I have claimed above that sexbotification meets roughly five of the ten features identified by Nussbaum and Langton, and thus it is probably apt to be called a form of objectification; others may disagree. But I will show below that non-consensual sexbotification meets more of Nussbaum’s and Langton’s features of objectification, and so even if someone maintains that (consensual) sexbotification is not a type of objectification because it does not meet enough of the features, then they can still accept my main argument that non-consensual sexbotification is objectification, and an intrinsically wrong form of objectification at that.

Even if one rejects Nussbaum’s and Langton’s list of features, it is clear that sexbotification is a form of sexual objectification because of the similarities it bears to other types of pornographic creations. Pornography treats women as sex objects (Jütten, 2016, p. 41, Langton, 2009; MacKinnon, 1987), displaying their bodily orifices as ripe for male penetration. Taking intimate photos of someone is sexually objectifying them, even if no direct physical contact occurs between the human subject and the photographer. Indeed, deepfake pornography¹² can be produced by an agent who has never even met the person in the videos they create. Non-consensual sexbotification would appear on McGlynn et al’s (2017) ‘continuum of image-based sexual abuse’, along with upskirting,¹³ revenge porn, and deepfake porn. But sexbotification is even more clearly a sexual act than is making pornographic photos or videos of someone, because the result (i.e. a sexbot) is fully physically interactive and capable of sexual intercourse, whereas photos and videos are merely for looking at.

It may be the case that it is intrinsically wrong to non-consensually create any kind of sexual product (such as sex dolls, pornography, and realistic sexual artwork), though I do not establish that here. However, the highly lifelike appearance and interactive nature of sexbots means that their creation is all the more troubling: a sexbot could potentially look and feel just like the human subject. If a user wants to have a sexual experience with S, then a personified sexbot of S will provide a far more realistic approximation than artwork, photographs, videos, or even an inert sex doll would provide. The intrinsic wrongness of making a fully interactive sexbot of someone without their consent therefore seems more pressing than other sexual wrongs such as taking photos without consent and looking at them whilst masturbating. Sexbotification is a shortcut; a way of fulfilling one’s fantasies without making direct physical contact with the human subject whom the sexbot represents. But as pornographic photography shows us, one need not touch another person in order to sexually objectify them. Although sexbotification falls short of using and abusing the human subject herself, there are few acts closer to engaging in real sexual intercourse with her; in sexbotifying a woman, the creator produces an ever-consenting¹⁴ automaton with her face and appearance. Sexbotification of a person replicates and represents the desired human subject, but with just a few modifications: the most depraved and violent sex acts can be performed on a sexbot without recourse. We might even refer to sexbotification as a form of “remote sexual violence”, as it has been suggested that upskirting and revenge porn are (Wittes et al., 2016).

Creating a personified sexbot is different from creating a generic sexbot, probably because we see our images as an extension of ourselves; sexbotification utilises the looks of the human subject, and an individual’s looks are an essential feature of their being. If someone sees me, then intentionally makes photorealistic drawings of me and stabs the drawings, this is an act which is directed towards me, as compared to someone stabbing images of an imaginary character he has created which does not represent anyone in particular. A sexbot is a sexual artefact by definition, so sexbotification is a sexual act which is clearly directed towards a particular individual: the human subject. This seems as intuitive as the suggestion that taking pornographic photos or videos of someone is an act of sexual objectification directed towards that individual.

Sexbotification shares much in common with the creation of deepfake porn. Both involve creating a visual product which depicts the victim in a sexualised way, and the creator can do this without ever coming into contact with the victim. Indeed, one need not even be in the same country as their victim, as the Scarlett Johansson case cited at the outset demonstrates. Deepfake pornography and so-called ‘sextortion’¹⁵ can also cross international borders (Wittes et al., 2016, p. 18); both are morally reprehensible¹⁶ forms of objectification which have been called “remote sexual violence” (McGlynn et al., 2017; Wittes et al., 2016). If we accept that deepfake porn and ‘sextortion’ are forms of objectification, it is reasonable to maintain that sexbotification is also a form of sexual objectification, and as such it should require consent of the human subject.

Some writers might argue that when a man creates any female-like sexbot—whether generic or personified—he intrinsically wrongs women qua women as a social class. This is not a position I defend here, but it can be consistent with my argument. Consider: even if all pornography is wrongful objectification, it is still reasonable to maintain that revenge porn, ‘sextortion’ and upskirting have an additional layer of wrongness on top of their objectifying women qua women by being pornographic; we can recognise that there is a particular person who is a victim in an additional way that others are not. Analogously, even if one maintains that all sexbot creation wrongs women qua women, it can still be the case that sexbotification without consent is an additional intrinsic wrong towards the human subject. Hereafter, I write as if sexbot creation is generally permissible; even if other writers deem generic sexbot creation and consensual sexbotification to be impermissible, they can still be sympathetic to my argument that they are more permissible than non-consensual sexbotification.

Recall that I am following the Nussbaumian thesis that objectification is not always wrong, but it becomes wrong when it involves disregarding the subject’s humanity by treating them as a mere means to an end (Nussbaum, 2007, p. 51). Nussbaum builds on the Kantian notion that humans are intrinsically valuable and should not be used merely for one’s own gain—to do the latter is intrinsically wrong, regardless of consequences. Although I have suggested in the previous section that sexbotification is a form of objectification, this would not necessarily mean it is intrinsically wrong; its intrinsic wrongness is established below.

Consent is pivotal to my argument regarding personified sexbot creation. Goldman presents a view of sexual consent which seems intuitively plausible, and can be utilised to support my argument. Goldman argues that sexual relations must be consensual if they are to be morally correct (Goldman, 1977); I take this as a baseline standpoint in general sexual practices, but I argue for it below in relation to sexbot creation. Sexual activity is the most intimate of human activity and we are inclined to think that all sexual contact or activity requires consent from all parties involved. Goldman writes that sexual activity is immoral when “the transactions are not freely and rationally endorsed by all parties” (Goldman, 1977, p. 282). When A obtains consent from B, this transforms the sexual interaction from an intrinsically impermissible one (where A was using B merely as a means to A’s end), into an intrinsically permissible one (where A is valuing B as an end in themselves) (Goldman, 1977, p. 283). Like Nussbaum, Goldman does not characterise objectification as something which is always impermissible, but it becomes impermissible when particular conditions (such as mutual consent) fail to be met. So both Nussbaum and Goldman argue along Kantian lines that sexual objectification without consent is intrinsically wrong because it uses a person without regard for their humanity (Goldman, 1977, pp. 282–283; Nussbaum, 1995, 2007). This is the account of objectification which I adopt below in relation to sexbotification.

Recall the two scenarios mentioned earlier:

Nussbaum accepts that objectification can be permissible and sometimes even ‘wonderful’ (Nussbaum, 1995, p. 251); it is only when someone uses a person merely as a means to an end (without valuing their humanity) that the act is intrinsically wrong. We should note that in Nussbaum’s account, like Kant’s, all the work is being done by the word ‘merely’; the claim is that it is permissible to treat someone as a means to an end so long as you value their humanity as well, but it is impermissible to treat them as a mere means without valuing their humanity. Valuing someone’s humanity would often involve taking their wishes into consideration where possible. Failing to obtain consent from the person being sexbotified is not valuing their humanity.

In Scenario A, Roy values Katie’s humanity, because he asks for her consent to create the sexbot, and only when Katie gives her consent does Roy proceed. So although Roy uses Katie as a means to an end (of sexual gratification), he also values her as ‘an end in herself’ by asking her permission; this means he does not treat her merely as a means to an end. Thus, Roy’s sexbotification of Katie is intrinsically permissible according to the Nussbaumian characterisation of objectification (Nussbaum, 2007, p. 51). Although Roy objectifies Katie, because this is done consensually, he has not committed an intrinsic wrong towards her, because he has valued her humanity. Sexual relations are immoral when they are not endorsed by both parties, and even if sexual use is one-sided (as in Scenario A where Roy sexbotifies Katie but Katie does not sexbotify Roy) this is permissible when done with sensitivity to the desires and demands of the other (Goldman, 1977, p. 282). I stipulated earlier that Katie has freely and rationally endorsed the creation of the personified sexbot which represents her, and Roy was sensitive to her desires and demands. We can say that Roy’s treatment of Katie has respected her wishes, and the creation of the Katie-sexbot has been endorsed by both parties involved. This is what makes scenario A morally permissible, and differentiates it from Fred’s treatment of Jane in scenario B.

Someone might assert that it is intrinsically wrong to create any personified sexbot. However, making such an assertion would be flawed because it would capture too much; it would capture instances of consensual sexbotification which are not prima facie instances of intrinsic wrongs. Sharkey et al. discuss our sexual future with robots, and suggest that sexbots of partners in long-distance relationships could be created where the couple can speak directly to one another through the robots’ mouths, in order to create “a mutual sexual experience” (Sharkey et al., 2017, p. 6). Although this sort of setup involves sexbotification (and therefore objectification), it does not resonate as something which is intrinsically wrong because the experience is consensual. On Nussbaum’s and Goldman’s account of objectification, this sort of arrangement is intrinsically morally permissible, as Roy’s consensual creation of the Katie-sexbot is.

In Scenario B, it does not take much scrutiny to realise that Fred is using Jane merely as a means to an end. Jane’s looks are an essential part of herself—they are unique and fundamental features of Jane. In creating the Jane-sexbot Fred capitalises on Jane’s very existence, performing a sexual act directed at her, in the same way that ‘sextortion’ and the creation of deepfake porn are sexual acts relating to particular victims—namely, those who are depicted in the images (McGlynn et al., 2017).

In “Objectification and sexbotification” section I showed that all sexbotification meets some but not all of Nussbaum’s and Langton’s features of objectification. Let me briefly explain how non-consensual objectification in particular meets more of the features. Firstly, Fred’s treatment of Jane denies her autonomy, agency and subjectivity (Nussbaum’s features no. 2, 3 and 7) because when he ignores her wishes he acts as if her feelings do not exist or need not be taken into account, and that she has no agency. Indeed, an agent respects a person’s autonomy, agency and subjectivity when he obtains her consent. Fred, of course does not obtain consent, and so he meets these three features of objectification. We can also say that Fred acts as though Jane has no boundaries or that those boundaries can be violated (feature no. 5). Jane had essentially drawn a personal boundary inasmuch as she had decided against having a sexbot made in her image; this is a line she is not willing to cross—a boundary. Fred violates that boundary and acts like it does not exist when he sexbotifies Jane. Furthermore, when Fred ignores Jane’s protests and creates the sexbot representing her against her wishes, this is effectively silencing her (feature no. 10). Even if someone is unconvinced that consensual sexbotification qualifies as sexual objectification (it met only half the features of objectification), we can be satisfied that non-consensual sexbotification is objectification (it meets 9 of the 10 features).¹⁷ There is insufficient space for me to argue for each of these in detail, and some opponents may feel that non-consensual sexbotification does not meet some of the features I have claimed above. However, as a cluster concept, it need only meet some of the features in order to be considered an example objectification, and it seems evident that making a sexbot to represent someone against their wishes would qualify as an act of sexual objectification, just as creating pornography is. Simply being a form of objectification does not make it wrong though—the (intrinsically) wrong-making feature of non-consensual sexbotification is its lack of consent.

Why should we think that there is a significant moral difference between consensual sexbotification and non-consensual sexbotification? We should think this because consent plays a pivotal role when making a moral evaluation of sexual activity between two people. Consider:

Lawrence and Michelle have sexual intercourse (both of them are over 18, conscious, of normal mental faculty, they are not under the influence of drugs or alcohol, and neither is in a position of power over the other).

Is their sexual intercourse morally permissible or impermissible? The answer, of course, is that we don’t know. Unless we know whether both parties consented to the intercourse, we are unable to make a moral determination regarding the permissibility of the act. If it were the case that both parties consented, then we would likely acknowledge that the sex was permissible. If, instead it were the case that one party—Michelle, for example—did not consent to the intercourse, then this fact is morally transformative; consent is the ‘moral magic’ of sexual matters (Alexander, 1996; Hurd, 1996). The non-consent of Michelle transforms the moral status of the act from permissible to impermissible in the laws of many countries, and it is (justifiably) seen as intrinsically wrong by any reasonable person (Archard, 2007; Cowling, 2001; Goldman, 1977; Wertheimer, 1996). In fact, we place such a high value on consent that even in cases where no harm ensues from a particular case of non-consensual sexual activity, we still believe that there is a victim who has been intrinsically wronged. Consider an example:

Natalie was drugged and unconscious when she was raped. The rapist used a condom, and Natalie was not injured nor impregnated, and she has no recollection of the incident. (Adapted from Gardner and Shute in Stewart 2010: 26–27).

In this example, Natalie has not suffered any negative consequences, and is unaware that anything untoward has taken place. Despite the fact that no negative consequences have occurred, we can still legitimately claim that the assailant’s actions were intrinsically wrong, because Natalie did not consent. Stewart remarks that a case such as Natalie’s is “a violation of her right to sexual autonomy” (Stewart, 2010, p. 27). In Nussbaum’s and Kant’s terms, Natalie has been used merely as a means to an end, rather than being valued as an end in herself (Nussbaum, 2007, p. 51). The wrong-making feature of Natalie’s treatment is her lack of consent, rather than anything else the rapist did. In the case of Natalie, we accept that what was done to her was wrong, even though no harm occurs; in other words, we see it as intrinsically wrong even though it is not instrumentally wrong. The salient fact is that sexual activity morally requires consent, and Natalie had a sexual activity performed on her without her consent, which is intrinsically wrong. I maintain that the same is true in a case where someone is non-consensually sexbotified. We see this in other areas of sexual activity: consent is morally transformative in the taking and sharing of explicit images of a person. It is generally seen as permissible to take or share images of an adult who consents, but it is a different matter entirely when the person depicted does not consent.

Some philosophers may argue that women cannot freely or fully consent to sexual treatment or activity in the same way and to the same extent that men can, particularly in a society where there is some degree of sexual inequality (Langton, 1993, pp. 323–326), however this is not a position I defend here. I take the position that (generally speaking) women are capable of consenting (or withholding consent) to sexual activity. Even if it is the case that women’s consent is less free than men’s consent is, we still tend to think that there is a difference between woman X who engages in consensual sexual activity, and woman Y who is forced to engage in sexual acts against her wishes. If a woman can consent to sexual activity such as sexbotification, and consent is morally transformative in sexual matters—as was shown above—then we should conclude that sexbotifying someone with their consent is intrinsically permissible, whereas doing so without their consent is intrinsically wrong.