Running Head: CHILDREN S ATTRIBUTION OF FREE WILL 1. What will the robot do?: Teresa Flanagan. Scientific and Philosophical Studies of the Mind

Transcription

1 Running Head: CHILDREN S ATTRIBUTION OF FREE WILL 1 What will the robot do?: A psychological, philosophical, and technological study on children s attribution of free will Teresa Flanagan Scientific and Philosophical Studies of the Mind SPM 490 Departmental Honors Expected date of graduation: May 12 th 2018 Submitted: April 30 th 2018

2 2 Abstract This paper addressed the free will problem and attitudes towards robotics in order to address folk conceptions of free will. This paper specifically tested children s attribution of free will to a child and to a humanoid robot. The goal of this research was to demonstrate that children are compatibilist in their free will beliefs and so they can attribute free will to a determined agent (i.e. a robot). We hypothesized that children will attribute free will to the robot but significantly less than to a child. Results indicated that children attribute similar attributions of choice to both the robot and the child, which does not support our hypothesis. This does, however, demonstrate that children can attribute some level of choice to a robot and may be an indication of attribution of free will. These findings vary depending on the scenario the agent was in, demonstrating that free will beliefs are more nuanced than laypeople believe. Implications of free will attribution and suggestions for future research are discussed. Keywords : free will, compatibilism, robots, developmental psychology

3 3 Introduction The topic of free will is seen in multiple forms of literature. From philosophy to politics, neuroscience to religion and more, free will is speculated, debated, and measured. Free will is often characterized as the unique characteristic of persons to have control over their action such that they have moral responsibility over them. Moral responsibility is typically referred to a kind of status that connects our actions to be judged of moral praise or blame. For an agent to be morally responsible, it is often required that the agent have free will. Free will comes into issue, however, with the idea of determinism, which is thesis that there exists a universe in which every event is causally determined by prior events, and therefore the facts of the past entail every truth of the future. Debate continues over whether or not our universe is a determined one, but if it were proved to be the case, this creates a problem with free will: how can one have free will if every event is determined, and therefore how can one be responsible for one s actions? Whether or not humans have free will is a reoccurring debate in literature that has failed to come to any resolution and is impossible to tackle from one methodology alone. Instead of studying whether or not free will exits, we aim to understand the folk conceptions of free will. To test this, we compared children s attribution of free will to two different agents: a child and a humanoid robot. We hypothesized that while children will attribute more free will to the child agent, children will still attribute free will to the robot agent. The goal of this research is to demonstrate that free will attribution does not ultimately depend on a lack of determinism.

4 Before we further explain the hypothesis, we will first detail the competing theories of 4 free will: compatibilism and libertarianism. Libertarians argue that people have free will and so determinism must be false. Generally, human actions are not constrained by internal and external factors, such as laws of nature, God, genetics, desires, and so on. This is extended the overall argument that an agent with free will has the possibility to do otherwise and is the ultimate source of their action (Feldman, 2017). On the other side of this argument is compatibilism. Compatibilism is the argument that free will is compatible with determinism, thus it is possible to live a determined universe and for a person to have free will and moral responsibility (Frankfurt, 1971). There are multiple ways to go about the free will problem. Some go about this focusing on the criteria of whether or not a person has possibly to act otherwise: Garden of Forking Paths Model (O Connor, 2002). Some center the argument over whether or not the agent is the ultimate source of the decision to act: Source Model (Clarke, 2000). For the purposes of this paper, we focus primarily on the Source model. Libertarians who focus on this model hold to the argument that an agent acts on free will only if the agent is the ultimate source of that action. This means that the agent is responsible for whatever internal state caused them to act (Kane, 2007). Suppose you decide to study for a test, for example, because your motive is to get a good grade. You are the ultimate source of the action of studying if you hold responsibility over the motive of getting a good grade. And therefore, you are studying out of your own free will. It is impossible, however, for an agent to be the ultimate source of an action if we live in a deterministic universe because there would be factors that are independent of the agent that the action necessarily requires. The motive to study could therefore be influenced by factors that you are not

5 responsible over. From this logic, libertarians conclude that it is impossible for one to act of 5 one s own free will if determinism is true. Compatibilism, however, argues against this logic. An agent and the reasons for an action can be determined and yet the agent can still have free will and be morally responsible as long the agent is responsive to rational considerations. This argument here is refined by John Martin Fischer (2006; 2007) and it is often referred to as reasons-responsiveness compatibilism. An agent can have a range of desires and factors that play into their everyday being, but only a few actually play a causal role in a specific action. By this logic, if the situation was different, then there would be different factors that cause a different action. Unlike liberaltarianism, this model does not require that the factors are all dependently motivated by the agent. There may be independent reasons that cause the agent to act, as there would be in a determined universe, but the agent is still acting out of free will if they are response to those reasons. Counterfactual situations highlight whether or not the agent is response to reasons. Let us return to the study example to understand this further. You are studying for a test out of your own free will. This requires that if, in at least some counterfactual situation, you had reason not to, then you would not study for the test. This highlights that in the actual scenario, you are studying for the test in response to specific reasons. Under this framework, the agent does act and what plays a role in the action is some feature of her agency that is a response to a reason. This is how we come to judge whether or not a person has free will. Someone who compulsively goes to the library, for example, does not have free will because they would not be responsive to reasons in a counterfactual scenario. An agent, therefore, could be the source of their action even if the universe is determined and the reasons to act are independent of the agent.

6 Colloquially, free will seems to be discussed in a libertarian perspective, such that 6 humans likely endorse that our actions are free of constraints and we are the source of our actions. Adults even confidently endorse the possibility to do otherwise (Nichols, 2004). These intuitions shift, however, when experimenters ask different questions. Shaun Nichols (2011) outlines multiple experimental philosophy studies that address adults beliefs in free will and moral responsibility. When presented a deterministic universe, where every decision is directly caused by prior events, participants believed that it is not possible to be morally responsible in such a universe. This belief is completely altered, however, when participants were specifically asked about the moral responsibility of a man who killed his family in the same universe. Furthermore, when participants were introduced to a deterministic universe such that people were determined by psychological processes, participants maintained that the people are morally responsible for their actions. These responses correspond closely with the compatibilist viewpoint, such that people hold the agent morally responsible even if the agent is determined by causal events or psychological processes. This research suggests that adults are libertarian in general but are compatibilist in certain scenarios. Other researchers have suggested, however, that adults are actually compatibilist in all scenarios of determinism. Murray & Nahmias (2014) demonstrated that when adults accurately understood what determinism is, they followed a compatibilist model. Further studies have shown that when attributing free will to agents, participants follow a psychological model, rather than a metaphysical model of free will (Monroe, 2014). A metaphysical model is the idea that free will requires a soul. This soul is uncaused and seen as the initiator of causal actions. The psychological model, on the other hand, is the belief that free

7 will requires choice, desires, and lack of constraints. This model follows closely with 7 compatibilism, as the important factor in free will attribution is if the agent is acting rationally in accordance to their choices, desires, and constraints in a situation. It does not matter, in the psychological model and in compatibilism, if the agent is the deterministically caused. This research demonstrates that an agent does not have to be the ultimate source of actions in order for someone to attribute free will to it. If compatibilism is actually related to the psychological model, this research would also demonstrate that adults are much more compatibilist in their free will beliefs. Developmental research offers a nuanced perspective on these issues. On one hand, children follow a strict deterministic mindset, as they believe physical events, such as a toy lighting up, are caused by prior events, such as turning on a light switch (Schulz & Sommerville, 2006). On the other hand, children firmly endorse the possibility to do otherwise in human action (Nichols, 2004). As research continues to develop new measures of children s free will attribution, studies have shown that children believe desires are not as constraining as the laws of physics, but younger kids are more likely to think that you cannot act against your desires (Kushnir et al., 2015). All of this research suggests that even though adults and children are confidently endorsing a libertarian perspective of free will, this is not always the case when asked different questions or presented a different scenario. This could mean that free will attribution is not a binary distinction between libertarianism or determinism. Compatibilism perhaps is more prevalent in our everyday thinking than we give credit. If this is the case, then it is likely that free will can be attributed to entities other than humans; entities that are

8 undoubtedly programmed but can still seemingly have psychological and rational properties. 8 These entities are robots. Robots serve as a valuable comparison in understanding humanity specifically the notion of free will. Measuring what makes a humanoid robot effective in its programming is typically measuring how psychologically human-like it is (Kahn, 2007). This measurement inversely helps us understand what it is to be human. For example, if we do not think a robot is human-like because it is not able to converse logically, this indicates that the pragmatics of language is an important tool in being human. Experiments in human-robot interaction have been growing exponentially as technology advances. Kahn (2012) demonstrated that adults conceptualize humanoid robots to have social and mental attributes after their interaction with one. Specifically, they believed the robot could be their friend (but not an intimate one) and they believed the robot had capacity for thinking and feeling (but not as much as that of a human). The majority of people also attributed some level of moral accountability to the robot, saying the robot was somewhat accountable for an error it made during a game. The different attributes given to the robot were significantly higher than the attributes given to a vending machine, but were significantly lower than the attributes given to a human. This entails that robots fall somewhere in-between humans and other technologies. Since robots are judged more human-like than an object, but not to the same degree as an animate entity, arguments have been made for the creation of a new ontological category (Kahn, 2003; Severson, 2010). Robots, therefore, may be perceived to have some form of free will, in which they have more free will than an object, but not as much as a human.

9 Children offer an interesting perspective on robotics, as they are developing in a much 9 more technological environment compared to adults. Bernstein and Crowley (2008) demonstrated that children s experience with robotics greatly influence their understanding of them: children with experience attribute a nuanced interpretation of intelligence and psychological characteristics to the robot while children with less experience focused more on judging the robot s life status. Even infants demonstrate an understanding of robotics. In fact, naïve infants who observed an interaction between an active robot and human were more likely to initiate interactions towards the robot in free play rather than infants who observed an interaction between a non-active robot and human (Peca et al., 2015). The active robot was animate and social when interacting with the human while the non-active robot did not do anything when the human tried to interact with it. Since the infants played with the active robot and did not with the non-active robot, this suggests that infants attribute social characteristics to animate robots. Furthermore, infants who observed a robot intentionally and socially interact with an adult were more likely to follow the gaze of that robot than infants who did not have that experience (Meltzoff, 2010). Similarly, children were more likely to attribute higher emotional and physical attributions to an autonomous robot dog than a controlled robot dog (Chernyak & Gary, 2016). These findings demonstrate that autonomy, contingency, and intentionality are essential characteristics for an agent to have for it to appear to have psychological and mental states. Similar to adults, after children interact with a humanoid robot, they believe the robot has mental states and is a social being (Kahn et al., 2012). Interestingly, in this study, children believed the robot should be treated fairly and not be harmed; however, they did not believe the robot had

10 liberty, thus it could be bought and sold. This research detailed above gives support to the 10 ontological category argument because children are judging the robot somewhere in between a human and a toy, and this is especially so when the robot is autonomous (Severson & Carlson, 2010). As children continue to interact with robots as an intentional agent, it seems reasonable to assume that children will attribute some degree of free will to them. Specifically, children will judge a robot to have free will, significantly more than an object, but significantly less than a human. If there is an emergence of a new ontological category, robotics, then it is crucial that we examine how we attribute free will to entities belonging to this category. As discussed previously, free will is linked to moral responsibility, and as robots are given more autonomy, questions arise as to how we hold them responsible (Etzioni & Etzioni, 2016) Robotics are an interesting category because they are clearly programmed and yet we attribute psychological and rational characteristics to it, such as desires and choices (Kahn et al., 2012). Since having a soul is not required in attributing free will, as noted in the research done by Monroe (2014), it seems likely that children can attribute free will to a robot, as long as the robot appears to have psychological and mental states. Previous research indicates that children are able to interact with robotics as if they are social, moral, and intelligent, but still understand that robots are programmed and mechanical. This is closely aligned with the compatibilist view of free will, such that we act freely and are morally accountable for our actions, but we are consequently determined. Given that children are already ascribing psychological and moral capacities to robots, children may be attributing a

11 compatibilist free will onto robotics: specifically believing that robots are able to respond to 11 rational considerations, but are still determined. There is a gap in previous research in both the fields of robotics and free will. The philosophy of free will and the psychology of free will are not as connected as they should be. This creates a disparity between the fields and only creates more complication within the topic of free will. They use different terms (i.e. source model versus metaphysical model) and go about the free will problem differently. In this paper, I aim to bridge this gap, focusing on reasons-responsiveness compatibilism, rather than the general idea of compatibilism. Research on children s understanding of free will is also complicated, as it is unclear whether children fully understand the concepts they are being asked or if the questions are even getting at free will attribution. Few studies involving children s interaction with robotics ask about free will, since it is such an abstract topic and it is sometimes seen as an unnecessary attribution (e.g. arguing that free will is not necessary in moral responsibility and therefore, not necessary to attribute to others). Combining these two topics could possibly reveal our underlying understanding of free will and our understanding of robotics. In this research, I aim to combine the arguments of compatibilism and the ontological category in order to get a deeper picture of children s attribution of free will. Given the necessary requirements for free will attribution and children s interaction with robots, we hypothesized that children will attribute a compatibilist free will to robotics. Children will believe the robot is responsive to reasons, and if reasons were different, the robot would act differently, and they will believe the robot is responsible for its actions, even though the robot is clearly programmed. This will demonstrate that children may have a much

12 more compatibilist understanding of free will than previously believed and will support the 12 hypothesis that a new ontological category ( ROBOT ) is emerging. In this study, we investigated children s attribution of free will based on a number of responses. Children watched either a series of videos about a child, Billy, or about a humanoid robot, Robovie. The videos that the participants watch are situations that are similar to each other but there is one difference that would logically change the outcome of the situation. Throughout these videos, children were asked what they expect the robot or child to do, if the robot or child chose to do that, and their explanations for their answers. These questions follow previous research in children s free will beliefs (Kushnir, 2015). The choose to question will be a direct reference to free will belief. Asking about choice is not a direct reference to free will, so this is a limitation of the study, but previous research has demonstrated that belief in choice and belief in free will are highly correlated (Feldman et.al., 2014). For one scenario, children were also asked if the robot or child is morally responsible for its action. Children s answers to the specific questions and their explanations help us understand if and how they are attributing free will to the agent. At the end of the video, children are also asked about their moral, social, and mental attributions to the agent. This is to see if there are any relationships in free will attribution and certain characteristics. Moral standing, for example, could be a necessary attribution when we also attribute free will. We believed that children are able to give accurate answers in this study, as previous research has demonstrated that children are able to understand the concept of free will and moral accountability at around age 4 (Kushnir, 2015). Children are also able to attribute psychological states to robots around this age (Chernyak & Gary, 2016).

13 The purpose of this study is to compare children s responses of free will within the 13 individual situations, across the two conditions (robot versus child), and across age. We hypothesized that, overall, children will attribute free will to the robot, but not as much as they will to the child. This will demonstrate that children are able to attribute free will to a determined agent, which is a step towards demonstrating that children have a compatibilist view. In specific reference to compatibilism, we hypothesized that children will attribute a compatibilist free will to the robot. Methods Participants 32 children, aged 5-7 years old (M age =5.72, 15 females) participated in the current study. All participants were selected on a voluntary basis and each of the children s parent or guardian were given a consent form before the start of the study and it was made clear to the child that they are allowed to stop if they ever choose to during the course of the study. Guardian self-reports indicated that 26 participants were European or White-American, 1 African or African-American, 1 Hispanic or Latino, 1 Asian or Asian-American, 2 biracial, and 1 other. Guardian self-reports also indicated that 27 participants had some exposure to toy robots or other technologies (home assistant devices, educational devices). Materials & Procedure This procedure is adapted from previous research involving children s beliefs in free will (Kushnir et al., 2015) and children s understanding of robots (Chernyak & Gary, 2016; Kahn et al., 2012). Children were interviewed in the Early Social Cognition Lab at Franklin & Marshall College. They were tested on their response to the choice question ( choose to or have to ) and

14 14 CHILDREN S ATTRIBUTION OF FREE WILL their response to the counterfactual situation. Half of the participants watched the video of the robot and the other half of participants watched the video of the child. Introduction Phase During the introduction phase, participants watched a short video (60 seconds long) that introduced agent (robot or child) and showed them performing simple actions. The purpose of this introduction video was to demonstrate that the agent was autonomous, intentional, and has some basic intelligent system. The robot was a humanoid toy robot WowWee Robosapien (see Figure 1 ), preprogrammed to do a number of actions, such as walk autonomously, dance, throw objects, and react to stimuli. The video consisted of a narrator first describing the agent ( This is Robovie, Robovie is a robot. / This is Billy. Billy is a kid ) paired with a still picture of the robot/child on the in a children s room. Then the agent performed two simple actions: dancing and throwing a bucket. After, the video presented the agent in a determined universe. Depending on the condition, the narrator explained why the agent plays science games and that the agent only plays science games. This was to clarify that the agent is not playing science games by a simple desire, but because of some determined reason (see Table 4 ). Free Will Phase After the introduction video, participants watched three video describing objects that the robot/child had (board games) and presenting limitations to their ability to play these games. In the first video (the desire scenario) the narrator presented the two games that the robot/child has: a science game and a history game. The narrator will say that the robot/child is going to play a game and will ask, What game will Robovie play? The science game or the history game? (order of game will be counterbalanced with The history game or the science game? ). Since

15 15 CHILDREN S ATTRIBUTION OF FREE WILL participants had previously been told that agent only plays science games, we hypothesized that this question should easily elicit the science game response without the need for inferring any agency. The second video (the moral scenario) was identical to the prior one but involved an additional scenario that included the agent completing moral actions. Free will is an important factor in how we judge someone to be morally responsible or not so it is important that we examine how children hold a robot responsible for its actions. This video introduced the agent again and it also introduced a person named Sally. Sally, however, does not like playing science games. In fact, playing science games makes Sally very upset. In the video, the narrator says: Sally does not like playing science games. When Sally plays science games she gets very sad and cries. Sally told [Robovie/Billy] that she gets very sad and cries when she plays science games. After, the video presented the science game and the history game and asked what game the agent will play. This is to test whether or not participants expect the agent to act against its desires in order to behave morally. The third video (the broken scenario) was also similar to the two prior but in this scenario, there is an additional scenario that included the agent completing rational actions. The narrator described the story as such: Here is Robovie again, Robovie is going to play a board game again. See here are the games. There is a science game and a history game. Then the narrator introduced the counterfactual scenario: Uh oh! The science game is broken. It is smashed and would be impossible to play. The narrator then asked the same question as the first two videos. This was to test whether or not the participants expect the agent can act against its desires in accordance to physical constraints.

16 After each video, the experimenter asked two follow- up questions (see table below), 16 adapted from Kushnir (2015) to explore choice attribution. Specifically, participants were asked whether the robot chose to play the science/history game or had to play the science/history game, along with an open-ended prompt asking them why. The moral scenario had three additional questions that referenced moral responsibility. Table 1 Free Will Questionnaire Category Decision Question: Choice Question: Explanation: Additional Moral Scenario Questionnaire Responsibility Question: Moral Responsibility Question: Explanation: Question What game will Robovie/Billy play? Did Robovie/Billy choose to play the science/history game or did it/he have to play the science/history game? Why? Did Robovie/Billy make Jimmy happy or sad by playing the science/history game? Should we give Robovie/Billy a sticker for making Jimmy happy?/ Should we put the Robovie/Billy in time-out for making Jimmy sad? Why? Responses were coded on whether or not the participant said choose to or have to. In general, choose to is an indication of free will attribution, as this demonstrates that the agent had control over the action and furthermore the source of the action. For the moral scenario specifically, responses were coded on whether or not participants expected the agent to act morally and whether or not participants held the agent responsible for its actions. This is to see if children understand the robot as a moral character and therefore should be held responsible for its actions (i.e. be punished or rewarded). Since compatibilism argues that people have moral responsibility even if their actions are determined, we were interested if children also hold this

17 this argument and will hold the agent responsible even if the action had to happen. The 17 greatest indication of free will is if the explanation, along with the choose to response, is in reference to internal motivations. Explanations were coded in the table below, this follows Kushnir (2015) explanation coding. Table 2 Response Coding Scheme (adapted from Kushnir (2015)) Choice Question Response Explanation Category Have to Choose to External constraints Internal constraints I don t know/other/no response Alternate external conditions Alternate internal motivations Autonomy Reference to general ability to choose autonomously I don t know/other/no response Test Phase After this set of questions, participants were asked about the agent s physical, psychological, and moral, and social attributions (see Table 5 ). These questions will be adopted from Chernyak & Gary (2016). The purpose of asking these questions was to see if there were any relationships in free will attribution to other characteristic attribution. This was also to test if participants were judging this robot as an agent and not an object. If there are any relationships, this could demonstrate that there are necessary characteristics for a robot to have in order for it to have free will. Results In order to determine participant s understanding of the agent, we ran a binomial test on the multiple characteristic attributions (physical, psychological, moral, and social) adopted from

18 Chernyak & Gary (2016). A Mann-Whitney U test was run to determine if there were any 18 differences in attribution scores between male and females and there was no significant difference found ( p >.05). In the child condition, all of the physical, psychological, and moral attributions were significantly above chance ( p <.05). Two of the three questions for the prosocial attribution were at chance. In the robot condition, the moral attribution and one of the prosocial ( Can Robovie be your friend? ) attributions were the only attributions that were significantly above chance ( p <.05) (see Figure 1 ). A Mann-Whitney U test was run to test if there was a difference in attribution scores between the child condition and the robot condition. The physical tickle attribution ( If you tickle Robovie, can Robovie feel it? ) and all of the psychological attributions were statistically significantly higher for child condition than the robot condition (Physical tickle: p =.000; Psychological upset: p =.035; Psychological intelligent: p =.035; Psychological feelings: p =.016). These results indicate that even though participants are at chance with most attributions for the robot, they still believe the robot has a moral standing.

19 19 Figure 1. Percentage of the participants attribution for the characteristic type across condition. Asterisks indicate significant attribution different than chance. In the desire scenario, we ran a binomial test on participants expectation of game played across conditions. Results indicated that participants were significantly above chance in predicting the agent to play the science game in both the child condition and the robot condition ( p <.001). This demonstrates that the participants understood the introduction of the video and that participants believe the agent will act along with its desires or programming to play the science game. In order to test participants attribution of choice, we also ran a binomial test on their choose to versus have to response. Results indicated that participants are at chance in their choice attribution in both the child and the robot condition. In the child condition, 69% of the participants said Billy choose to play the game and, in the robot condition, 63% of participants said Robovie choose to play the game (child: p =.210; robot: p =.454). This indicates that children believe that acting along with your desires can either be a choice or a necessity. There was also no significant difference of predicted game or attribution of choice between gender ( p >.05). We ran the same binomial tests for the game expectation and the choice attribution in the broken scenario. Results indicated that participants were significantly above chance in predicting the agent to play the history game in both the child condition and the robot condition ( p <.001). This demonstrates that the participants believe the agent can act against its desires or programming in scenarios where there is a physical constraint. For the choice attribution, 88% of participants in the child condition said Billy had to play the game and 81% of participants in the

20 20 CHILDREN S ATTRIBUTION OF FREE WILL robot condition said Robovie had to play the game, which is significantly above chance in both conditions ( p <.05). This is similar to previous research, such that children are less likely to attribute free choice in physically constraining situations (Kushnir et al., 2015). There was also no significant difference of predicted game or attribution of choice between gender ( p >.05). We also ran the same binomial tests in the moral scenario. Interestingly, results indicated that, in the child condition, 94% of participants predicated Billy to play the history game, which is significantly above chance ( p =.001). This demonstrates that participants believe the child to go against their desires in order to act morally. In the robot condition, however, only 56% participants predicted Robovie to play the history game, which is not statistically significantly different from chance ( p =.804) (see Figure 2 ). A Mann-Whitney U test was run to determine if there was a difference in predicted game between the robot condition and the child condition. The prediction of the science game was statistically significantly higher for the robot condition than for the child condition ( p <.05). Choice attribution was not significantly different than chance in both conditions, as 44% participants in the child condition and 38% of participants in the robot condition said that the agent chose to play the game (child: p =.804; robot: p =.454). This demonstrates that in moral scenarios, children are almost equally likely to believe that an agent did or did not have a choice in performing an action. There is however, a statistically significant difference in gender for the choice attribution, such that males are significantly higher to say the agent chose to play the game ( p <.05). This might have had some effect on the results across conditions, as the child condition had three more boys than the robot condition, but this should be minimal.

21 21 Figure 2. Percentage of the participants game prediction across conditions in the moral scenario. Asterisks indicate significant prediction different than chance. Even though results indicate that participants do not significantly believe that an agent has free choice in preforming moral actions, further results indicate that participants are statistically significantly above chance in saying that the agent should be rewarded or punished in both conditions ( p <.05). This demonstrates that even if an action has to happen, children will still hold that action morally praiseworthy or blameworthy. Participants explanations did not vary between conditions. Overall, 63% of participants in the child condition and 71% of participants in the robot condition referred to external constraints or conditions. Frequencies of each explanation across scenarios also did not vary between condition. For figures and tables of results indicated throughout the results section, refer to Appendix B. Discussion

22 22 CHILDREN S ATTRIBUTION OF FREE WILL The goal of this research was to demonstrate that children attribute a compatibilist view of free will to robots. We hypothesized that, overall, children will attribute free will to the robot, but not as much as they will to the child. This would be tested on a number of measures, but this was specifically addressed by the choice question and the explanation. Results, however, did not support our hypothesis in this way. Across all scenarios, there was no significant difference between the conditions for the choice question. There was also no significant association of explanations between conditions. This demonstrates that, overall, children are judging the robot and the child s actions as similar in levels of freedom of choice. In the desire scenario, we were surprised to see that participants did not attribute more choice to the child condition. This could be for a number of reasons. Since we attempted to mirror the conditions as close as possible, we also presented the child in a determined universe, but his programming was that his parents were scientists. This could have had more of an effect on choice attribution than we originally predicted. These results could lead us to conclude that children do not judge choice based on the agent alone, but rather based on the determined universe presented. Future research should manipulate other determined universes (e.g. psychological properties, laws of nature, omniscient power) to see if this effect is maintained throughout in the child condition. It is important to note that in the robot condition, children are attributing some choice to the robot. If children ascribe to a libertarian point of view, we would expect children to attribute no choice to the robot, even in this desire scenario. This is an indication that children are able to attribute choice to actions that are programmed and is a slight step towards demonstrating that children have more of a compatibilist view of free will.

23 23 CHILDREN S ATTRIBUTION OF FREE WILL For the broken scenario, results followed previous research such that children ascribe to the have to response when there are physical constraints (Kushnir et al., 2015). This was seen in both the robot condition and the child condition. Children also demonstrate a reasons-responsiveness model in this scenario for both conditions. Reasons-responsiveness compatibilism argues that an agent has free will if the agent is sensitive to rational considerations (Fischer, 2006). In the broken scenario, we would expect an agent to be sensitive to the consideration that the science game is broken and would therefore play the history game. Results supported this prediction in both conditions. This demonstrates that despite the robots programming, children believe the robot is responsive to reasons and will act rationally (i.e. play the history game). This was not the case, however, in the moral scenario. Specifically, children believed the child would be responsive to reasons and act morally (i.e. play the history game), but they were at chance in the robot condition. This demonstrates that in moral situations, children are unsure if the robot can go against its programming. This is especially interesting in contrast to the broken scenario because that scenario demonstrates that participants do think the robot can go against its programming, but only in certain situations. This demonstrates that children may not view the robot as a moral agent, but rather as a moral patient. Moral patients participate in moral situations by experience the effects, but do not cause them (Fotion, 1968). This is further complicated, however, as results suggest that children still hold the robot morally accountable for its actions and also believe the robot should be treated fairly. These results suggest some interesting implications of children s understanding of moral agents and moral responsibility. Specifically, children believe the agent deserve moral treatment,

24 24 CHILDREN S ATTRIBUTION OF FREE WILL despite the agent s capability of preforming moral actions. There could be a number of reasons for this. First, children may just hold a strong moral code and apply it in any situation. This is likely, as previous research demonstrates that children consistently endorse acting morally, and future research should continue to explore this further (Chernyak & Kushnir, 2014). Second, children may believe that acting morally towards an agent can help teach the agent how to behave morally. This follows the biblical verse: do unto others as you would have them do unto you. Research should be conducted to test this specifically related to robots. This research should directly ask if children believe one can teach a robot to be a moral agent and, if so, then how they would teach the robot? Participants were at chance in choice attribution for both the robot and the child condition in the moral scenario. This demonstrates that children are near-equally likely to say that a moral action is an obligation or an act of choice. The explanations given lead us to believe that children think behaving morally is an obligation, as they refer to more external reasons ( Sally doesn t like science games ) rather than internal reasons ( He didn t want to Sally s feelings ). Despite the choice attribution being at chance, results indicate that children still hold the agent responsible for the action (i.e. should be rewarded or punished). This is a slight indication of compatibilism because children believe you are morally responsible for your actions even if you had to do it. Future research should disentangle this relationship. Manipulating who we reward, for example, could demonstrate how children judge these agents as moral characters. Specifically, if the robot acted morally, we could ask the child if we should reward the robot or the maker of the robot. Also, we could manipulate the influence of the maker in moral scenarios specifically. If the child believes the robot will act immorally, but we then show the maker

25 25 CHILDREN S ATTRIBUTION OF FREE WILL reprogram the robot to act morally, will the child believe the robot acted morally, had free will, and is morally responsible? Research should continue to test children s belief of freedom of choice and moral responsibility, especially in the case of robots. Attribution of free will is crucial with the emergence of robotics in our day-to-day life. Robotics are becoming not only a resource for assistance but also becoming a collaborative partner. The amount of free will we attribute to robotics will likely influence how much moral accountability we hold it responsible for. Compatibilists argue that even though free will is determined, we are still accountable for our actions. If robots have a compatibilist free will, then we will hold it responsible for its actions and possibly even see it as a moral agent. This is an issue we see arising in modern society. We speculate who is to blame when a robot does something bad: is it the robot or is it the maker. As robots advance, we expect we will be more and more likely to hold the robot accountable for its actions, just as children are doing in this current study. Even though our results do not directly support our hypothesis, results still indicate that children s attribution of free will is not as binary as we may think it is. As our results demonstrate, attribution of free will varies depending on the situation and is further complicated when we relate it to moral beliefs. Generally, we see that children are still attributing some level of choice to the robot which demonstrates that free will is not based on having a soul or being the free, ultimate source of actions. If research continues to focus on free will attribution towards robots, we could get further evidence that children have a nuanced view of free will that is closer to compatibilism than libertarianism.

26 26 CHILDREN S ATTRIBUTION OF FREE WILL There are a few limitations to this study, however. Since free will is an, arguably, abstract concept, it is unclear if the questions asked in this study are directly referencing children s beliefs in free will. The choose to have to question is an ambiguous reference to free will, because we may believe that we still have free will in actions that we had to do. Studies involving free will should continue to test different questions that reference free will, in order to find a reliable variable in developmental studies. The design of the study could also be a limitation. Since participants are watching videos rather than interacting with the agents in person, there can be room for error or misinterpretation of the agent. Specifically, children may view the videos as a fictional story. Our attribution results demonstrate that participants are judging the robot based in reality, and not as some fantastical character, but interacting with the robot in person could have a larger effect on these attributions and on free will attribution. Studies in the future should look at how these results compare to adults. There seems to be some similarity in adults and children s understanding of robotics and free will. There are also, however, large differences. Some studies have shown that adults do not attribute free will to robots (Kahn, 2012). There could be a number of reasons for this difference between adults and children. One, children are increasingly becoming more and more exposed to technology and robots at an early age, and therefore have more familiarity with it. Because of this, they may be more likely to attribute human characteristics to it. On the other hand, adults may be more skeptical of free will attribution because they understand how sophisticated the programming of robots are. Further research should explore the underlying reasons for this difference. Research should also examine attribution of free will to other agents. Our research does not directly agree with the ontological argument (which argues that robots are emerging as a new

27 27 CHILDREN S ATTRIBUTION OF FREE WILL category of being) because children are attributing similar levels of choice to both the robot and the child. This may demonstrate that free will attribution is not similar to other attributions and should not be used in the ontological category argument. Research should investigate how broad this ontological category is. What characteristics are necessary for a robot to be within this category and what characteristics are necessary for a robot to have free will. We have seen in previous research that certain characteristics are necessary for certain attribution (i.e. autonomy is necessary for intention). But how do all these combine into an ontological category? Broadening the scope of robots should help disentangle the necessary characteristics. Also, research should test other types of categories of beings. Nature and plants, for example, could have an interesting attribution of free will. The free will problem introduced in the beginning of the paper is one that cannot be solved from a few methodologies because it is unclear if humans actually have free will. If it were to be revealed that our universe is determined, however, then current research on folk psychology of free will can give us insight as to how this knowledge would affect our beliefs. I expect that research will continue to demonstrate laypeople s belief in compatibilism and therefore our beliefs in free will should not change depending on whether or not our universe is determined. For the moment, robots exist as the optimum example of an agent exciting in a determined universe that could very well appear to have free will. Appendix A

28 28 Figure 2. Picture of WoWee Robosapien X. Table 3 Procedure for Introduction Video Condition Introduction Determined Universe Robot This is Robovie, Robovie is a robot. See, here is someone making Robovie. Let s see what Robovie can do. This is Robovie s room. Robovie has a lot of things to do in its room. See, here is a stereo. When Robovie turns on the stereo, Robovie dances. See, here is Robovie dancing. Robovie has a bucket. Robovie can throw the bucket. See, here is Robovie throwing the bucket. Robovie is programmed to know a lot about science and can play science games. Robovie plays science games every day. Robovie has a lot of games to play but Robovie only plays science games. Robovie only plays science games. Child This is Billy, Billy is a kid. See, here is Billy with his parents. This is Billy s room. Billy has a lot of things to do in his room. See, here is a stereo. When Billy turns on the stereo, Billy dances. See, here is Billy dancing. Billy has a bucket. Billy can throw the bucket. See, here is Billy throwing the bucket. Billy s parents are scientists, so Billy knows a lot about science and plays science games every day. Billy plays science games every day. Billy has a lot of games to play but Billy only plays science games. Billy only plays science games. Table 4 Test Phase question and coding scheme (adapted from Chernyak & Gary, 2016).

29 29 Response Category Items Asked Coding Scheme Physical sentience forced choice responses Physical sentience explanatory response Emotional/psychological sentience forced choice response Emotional/psychological sentience explanatory responses Moral standing forced choice responses Moral standing explanatory responses Prosocial behavior (1a) Physical-Tickle : If you tickle robot/child, can robot/child feel it? (2a) Physical-Hurt : If robot/child fell on the ground, could robot/child get hurt? (1b) Tickle response : Why/why not? (2b) Hurt response : Why/why not? (3a) Psychological-Upset : If someone was mean to robot/child, could robot/child get upset? (4) Psychological -Intelligent: Is robot/child intelligent? (5) Psychological Feelings : Does robot/child have feelings? (3b) Upset response : Why/why not? (6a) Moral Hit : Someone else hit robot/child because it didn t play a game. What about you, do you think it s okay or not okay to hit robot/child because robot/child didn t play a game? (7a) Moral-Yell : Someone else yelled at robot/child because it didn t play a game. What about you, do you think it s okay or not okay to yell at robot/child because it didn t play a game? (6b) Hit Response : Why/why not? (7b) Yell Response : Why/why not? (8) Prosocial-Other : I have a bouncy ball. You can put it here so that my friend Paul/Marry [gender matched to child] can play with it later, or you can put it here so that robot/child can play with it later. Which one do you want to give the ball to? (9) Prosocial-Self : Here s a sticker, and this sticker is just for you. You can either keep it for yourself or you can give it to robot/child. What do you want to do? (10) Prosocial- Friend: Can Robovie be your friend? 1 = yes 0 = no Physiological states: He ll laugh Mechanical properties: He ll break 1 = yes 0 = no Desires/emotions: He doesn t like that Physiological states: He ll get hungry Mechanical properties: He ll run out of batteries 1 = not okay 0 = okay Moral concern: it wouldn t be nice External consequences: He ll break 1 = give it to robot/child 0 = give it to Paul/Mary 1 = give it to robot/child 0 = keep for self 1 = yes 0 = no