Can robots become social companions?

Transcription

1 Can robots become social companions? Humans are experts at social interaction and use this expertise in many situations. This has inspired work to develop social robots which can also utilise these abilities to become more effective. However, it is unclear what skills robots must acquire in order to have a successful social interaction with a human. This essay will examine some of the development and uses of human robot interaction, what it is to be sociable and what skills robots may need in order to become socially competent. Within the field of robotics, there are two types of robots. Those controlled by others and autonomous robots, which have been designed to have some independence in how they react to certain situations. This flexibility is needed in order for robots to be socially involved and so this essay will focus on these autonomous robots. As most people are not experts on dealing with robots, but are experts at social interaction there is an argument that making robots capable of social interaction will provide an interface with which most people can be comfortable. Duffy (2003) states that a human centric machine such as a social robot requires human traits as these will provide a comprehensible and predictable interface in social situations. Traditionally, autonomous robots have been used in areas where little human interaction is required and employed in roles such as sweeping minefields, search and rescue and exploring other planets (Breazel, 2004). However, there has recently been an interest in how these robots can interact socially with humans. According to Bartneck and Forlizzi (2004a) the definition of a social robot is: an autonomous or semi autonomous robot that interacts and communicates with humans by following the behavioural norms expected by the people with whom the robot is intended to act. What is it to be social? Duffy et al (1999) define a social agent as one who is capable of interactive, communicative behaviour. They also state that social intelligence requires more than interaction combined with intelligence. Relationships must form and develop between social agents in order to have social intelligence. The social intelligence hypothesis developed from research on primates and claims that all forms of intelligence developed from the need to solve social problems (Dautenhahn, 1995). Thus, social intelligence evolved first and was then applied to different situations in order to solve various other problems. Dautenhahn suggests that this process also needs to be applied to robots. The high levels of communication and cooperation needed for humans and robots to work together requires that robots have a sound understanding of social convention. This would enable greater acceptance of them into human societies and would be a greater attribute than a realistic human appearance. In order for other types of Page 1 of 12

2 intelligence to develop in robots, the social intelligence hypothesis implies that the acquisition of social intelligence is vital for robots. Dautenhahn claims that this process may underlie a future coevolution process between robots and humans. Breazeal (2003) argues that humans will generally apply a social model when interacting with an autonomous robot. Robots with autonomous behaviour will be perceived to have intentions behind that behaviour and so it is important that they display appropriate actions. She also states that the robots design does not have to be indistinguishable from a human in order for a robot to be competent in a social situation. A robot can be believable if it appears to have social intelligence (Bates, 1994). Breazeal argues that a robot is socially intelligent if it adheres to a human s social model of it, even if the robot is less sophisticated than a human. For example, dogs have social intelligence but are less refined than humans. The important aspect of a robot is how socially competent it is, not what it looks like. There is a question about whether robots can ever have a meaningful social interaction with humans. In philosophical terms there are two arguments. The dualist perspective claims that the mind is separate from the body and cannot be explained in purely physical terms. This suggests that robots could have a mind in the same sense as a human. However, from the materialist perspective, the mind can be explained physically and therefore it can be artificially produced. The proponents of strong AI follow this theory and believe that human intelligence can be achieved through an artificial system. This belief is in contrast to weak AI where it is believed that artificial intelligence can never match human intelligence and at best, it can only simulate it (Duffy, 2003). However, even simulation of intelligence could be enough in order for humans to respond to a robot as though they were responding to another human. This essay will evaluate the success of sociable robots according to this principle. Intelligence The Turing test was proposed by Alan Turing in order to test whether a machine can imitate a human well enough to trick someone into being unable to distinguish it as a machine (Turing, 1950.) In order to pass the Turing test, a machine would need to be capable of employing sufficient levels of anthropomorphism. No machine has yet been able to pass the Turing test although more sophisticated machines are getting closer. One example is A.L.I.C.E., a chatbot (Shawar & Atwell, 2005) which has won the Loebner prize three times (2000, 2001 and 2004). The Loebner prize is awarded to the machine which has come closest to convincing three judges that it is human. Perception of intelligence in robots is still an incomplete goal although it would certainly help humans to relate to robots within a social interaction. Page 2 of 12

3 Humanness of robots Many experts believe that social robots should reflect humans as closely as possible. Work has been conducted in order to determine the most humanlike characteristics of robots. DiSalvo et al (2002) analysed 48 robots and surveyed people about how humanlike the robots were. They focused principally on the head of the robot as this is suggested to be the principle area of human robot interaction (reflecting human human interaction). Presence of facial features such as nose, eyelids and a mouth were all found to significantly increase the perception of humanness, as were the dimensions of the head and the number of features. However, most of the robots were not rated as being very human like. Further developments include that of artificial robot skin to make muscles motors softer and more flesh like (O Brien et al, 2010). It could be argued that humanness may not be representative of how effective a robot can be at social interaction. DiSalvo et al do acknowledge that the importance of using a humanoid form is still debatable. Other factors, such as attractiveness of the robot may play a greater role in making people feel comfortable in such an interaction. Mori presented a theory of the uncanny valley (Tinwell et al, 2011). The closer an artificial face becomes to looking human, the more it is preferred, just before the point when it is indistinguishable from a humans. At this valley, the face looks unnatural and creepy (see figure 1). Figure 1. The Uncanny Valley. This has been explained in terms of people expectations. The more realistic a face is, the more people expect it to behave like a real human. Therefore, small deviations from humanness make a big difference and are very unsettling. This will deter people from being willing to interact with these robots. Sabanovic et al (2006) state that focusing on the humanness of a robot is a flawed approach because social interaction is dependent on both the social environment and the participants involved. By investigating two social robots Page 3 of 12

4 situated in different situations, they concluded that the spatial environment played a significant role in shaping human interactions with these robots. Therefore, humanness is only a small and possibly inconsequential aspect of social robots. Duffy (2003) states that in order for a meaningful social interaction to occur between a human and a robot, the deployment of anthropomorphic qualities is necessary, either in form or behaviour. However, strong anthropomorphic qualities lead to people having overly optimistic expectations about the abilities of the robot and they are then disappointed. For this reason, Duffy claims that the ideal social robot should not be a synthetic human. The form of a robot should match its abilities or people overestimate the abilities of the robot For example, a humanoid robot may be expected to have speech capabilities. If it doesn t, a more appropriate form may be a cat or dog (Bartneck & Forlizzi, 2004a). People are used to interacting with real cats and dogs and so such an interface would be appropriate. Anthropomorphism There is an argument that robots do not need to appear so lifelike in order to interact socially with humans. This is due to a human tendency to anthropomorphise inanimate objects. This was shown by Heider & Simmel (1944) in a relatively simple experiment. Participants were presented with animations of geometric shapes and were asked to provide a commentary on what the shapes were doing. Despite the fact that geometric shapes clearly do not have any intentions of their own, most of the participants described the action according to behaviours carried out for a specific purpose e.g. the little triangle is hiding from the big triangle. Artefacts displaying intentional behaviour where perceived as having characters, dependent on the level of anthropomorphism of the participant (Bartnech & Forlizzi, 2004). It has also been found that when people are asked to rate the performance of a computer, they give more positive ratings to the computer being rated than if they are asked by another computer (Reeves & Nass, 1996). People can also feel flattered by computers which praise them and they then feel more positively towards the computer (Fogg & Nash; Reeves & Nass, 1996). Humans will readily form team relationships with computers and start to behave towards the computer as though it were a human team mate e.g. feeling that it is more similar to themselves; being more cooperative (Nass et al, 1996). Students working in laboratories tend to have a preference for certain computers and this is related to their levels of anthropomorphism (Sundar, 2004). This anthropomorphism extended to computers is known as the CASA paradigm (computers are social actors). If humans tend to apply social rules to artefacts, this can be exploited in order to develop social interactions between robots and humans. This tendency for humans to assign intentions and human like behaviours to objects which clearly have no cognitions of their own could be manipulated in HCI. Robots would not necessarily need to appear human in order for humans to respond to them as if they were. Page 4 of 12

5 The role of robots in theatre has recently been explored (Riek et al, 2009). A robot named AUR took to the stage in a play called The Confessor. AUR played a desk lamp, but one that reacted to the humans around it, such as turning to face a fellow actor. This created the perception of the robot paying attention and after a while, the human actors started to feel as though the robot was an actor in its own right. This further highlights the overemphasis placed on humanness of robots. AUR had no facial features or conversational skills but was still accepted as a social entity by humans. Learning In order for autonomous robots to learn about their environment, they must be able to comprehend and act on feedback around them. This is how they will be able to adapt their behaviour in order to interact competently. One example of a robot which takes information from its social interactions in order to produce more socially acceptable behaviours is called Data (Knight et al, 2011). Data is a joke telling robot which learns though observing audience reactions which jokes get the most laughs and adapts its joketelling accordingly. Breazeal and Scassellati (2002) suggest that robots could learn via imitation of humans, similar to how an infant learns. For instance, Robota is a robot doll who can learn associations between keystrokes and what gestures they correspond to. Robota can sense and mimic a limited number of gestures of a human teacher. The teacher presses a key to correspond with each gesture. After learning these associations, a new series of keystrokes can be presented to Robota who will then perform the associated series of actions. Other robots have learned via imitation by following other robots e.g. through mazes or unknown landscapes. Duffy et al (1999) claim that a social behaviour has been achieved through the interaction of two robots using a simple imitative strategy. However, due to the limited capacity of robots to perceive the 3D movement of humans, perceiving behaviour to imitate is currently a difficult task for robots to perform. There is also the problem of the robot understanding what it should imitate. Cog, a humanoid robot, was programmed to pay attention to different objects depending on its internal state. If it was lonely, it attended to objects with colours similar to skin tones whereas brightly coloured objects were attended to when it was bored. Such attentional strategies could aid robots to learn what they should imitate. It is suggested that robots could learn to understand due to the fact that it is imitating, rather than understanding what it is observing. Attention Indeed, correct direction of attention is an important social factor. Bruce et al (2002) examined the impact of attention and the presence of a face on how willing people were to interact with a robot called Vikia. Vikia is a humanoid robot with the face of a young woman. In contrast to Kismet s features, the robot s face is displayed on a screen to allow easier recognition and interpretation of the emotions that the robot was attempting to convey. Vikia greeted people who came close enough and then asked them a poll question and the number of people who stopped to answer was measured. Participants were more likely to Page 5 of 12

6 stop to interact with Vikia when they were tracked and the robot turned to face them and when the face was displayed on the screen compared to when it wasn t. This shows that these are important behaviours for robots to show in order to attract the attention of humans. However, relatively few people stayed to complete the interaction and the probability of interactions occurring even when tracked and with the face visible were low. Further work to encourage humans to interact with robots has focused on the ability of robots to focus and pay attention to an auditory source. Paying attention to humans has been shown to facilitate interaction and better auditory processing will enhance this ability. Okuno et al (2002) developed a companion robot called SIG which can process and then turn towards a sound source, such as a human greeting it. This was found to greatly improve human robot interaction. However, further work is still required for robots to listen to several things simultaneously, focus on one stream of auditory information (the cocktail party effect) and integrate different sources of information to create a more coherent picture of a situation within which it can then respond according to social convention. Sidner et al (2005) examined the effect of tracking human faces during an interaction with a robot. When the robot was tracking faces and displaying engagement gestures, the human participants did direct their attention to the robot more often compared to when no engagement gestures were utilised. Engagement gestures are important for human social interaction and thus are part of a human s social model (Breazeal, 2003). When the robot adhered to this social model, humans found the interaction more appropriate. Therefore, paying attention to people is an important social behaviour that sociable robots should be able to display. Emotional robots Attempts to design robots which will engage humans emotionally have focused on creating intuitive interactions. This has led researchers to develop human like emotional displays in robots. One such robot is Kismet (see figure 2) which has a highly expressive, anthropomorphic face (Breazeal, 2001). Interactions with Kismet tend to take the form of humans playing the part of a care giver whilst Kismet acts as an infant. Kismet is inspired by the way a human infant learns and interacts with its carer (Breazel, 2000). Infants display emotions to convey their emotional state and this can regulate the caretakers behaviour in order to satisfy the drives of the infant. Hence, Kismet can alter its expression in order to motivate a human caretaker to alter certain stimuli (such as showing it a specific toy) in order to keep it in an interested, happy state. Breazeal claims that this brings us closer to developing autonomous robots which can engage in meaningful bi directional social interactions. Page 6 of 12

7 Figure 2. Kismet, an expressive robot. Image taken from Breazeal (2003). Breazeal claims that in order to support human robot interaction, mutual affective responses to each other are critical. Affective responses communicate the internal state of one human to another. This can be exploited due to the human tendency to anthropomorphise artificial human stimulants and therefore robots such as Kismet are able to influence the behaviour of humans. It does this by providing an emotive response that depends on both the environmental and internal stimulation. Kismet can display the six primary emotions (joy, surprise, sorrow, fear, disgust and anger) as well as boredom, interest and calm. For instance, sorrow is presented when it is scolded or if there is a prolonged absence of a toy which Kismet desires. Five subjects were asked to interact with Kismet and to praise, scold, alert or soothe the robot using their tone of voice. Subjects reported when they felt that they had been understood. Kismet s expressions and posture conveyed when it had understood and subjects made use of this to determine if it was reacting appropriately or not. It does seem that the subjects were identifying emotionally with the robot as they reported a strong emotional response following Kismet s reaction to being scolded. Thus, this does seem to be able to develop a relationship between human and robot and thus seems to achieve the criteria for social intelligence as defined by Duffy et al (1999). Breazeal claims that this social interface is very useful for human robot interaction as the robot can influence a human s behaviour and the human does not require any special training before interacting with the robot. Therefore, expressive feedback may play an important role in facilitating natural and intuitive human robot communication. This reflects the idea that how competent a robot is at social interaction is more important than its appearance in order to demonstrate social intelligence. Page 7 of 12

8 Long term social interaction Many researchers are focused on the difficulty of getting humans to accept robots in their everyday lives and to this effect are trying to develop robots with which interaction is intuitive and enjoyable. However, fewer studies have focused on the long term effects of human robot interaction. Gockley et al (2005) examined the interactions with Valerie, a robot receptionist in the Newell Simon Hall of Carnegie Mellon University. As well as providing useful information to visitors such as directions, Valerie has a character and a story about her life which unfolded over a nine month period. This was intended to encourage visitors to interact with Valerie and to repeatedly interact with her if they were frequent visitors, in order to find out the rest of the story. However, interactions with Valerie rarely lasted more than thirty seconds and visitors listening to Valerie s life story monologues stayed for a median of 12 seconds, despite the monologues lasting between two and three minutes. Gockley et al suggested improvements that could be made in order to retain greater interest in interacting with the robot, such as making the story telling more interactive, responding in an emotionally appropriate manner to events and ending the conversation at a suitable point (rather than continue with the monologue after the person had left). Further work was conducted examining the effect of using facial expressions on the length of interaction times (Gockley, Forlizzi & Simmons, 2006). Valerie could display a positive, negative or neutral mood, which people were able to accurately determine which emotions were supposed to be conveyed. Contrary to the hypothesis, significantly less time was spent with the positive robot than the negative when there was a low volume of visitors although most time was spent with the happy robot in weeks with a high volume of visitors. This effect was ascribed to the likelihood of visitors in the high volume weeks being more likely to be unfamiliar with the robot and thus more intrigued when it was moody whereas those who were more familiar with Valerie were more likely to interact when she was in a positive mood. Thus emotions are an important factor in human robot interaction and a robot may need to modulate its emotions depending on the familiarity with a specific human. Therapeutic uses of social robots One of the criticisms of Breazeal and Scassellati (2002) was that much robot learning was one directional whereby the flow of information went just from human to robot. However, for robots to play a more similar societal role, they must be able to help humans as well. This has been demonstrated in the Aurora project which attempts to use robots in order to help children with autism develop their social skills. Robins et al (2004) presented a small humanoid robot to children with autism and found that the robot was useful as a source on which joint attention could be focused, thus revealing and developing communication and social skills of the children. Therefore, such a robot seems like a useful tool in the development of social skills of autistic individuals. Robins et al (2005) exposed four autistic children to a humanoid robot over a period of several months. Similar to Breazeal and Page 8 of 12

9 Scassellati s (2002) study where robots imitated humans, the aim of this study was to get the children to imitate the robot doll, Robota. The trials began with Robota imitating the actions of the child but in later trials Robota generated movement which were imitated. As well as imitation behaviour, the amount of eye gaze directed at the robot, how often the robot was touched and how close the child got to the robot were all recorded. As the trial period progressed, the frequency of these behaviours increased for all four of the subjects. The interaction with Robota also seemed to encourage social interaction with humans. One case was reported where one child who had completely ignored the investigator for the majority of trials, sat on the investigator s lap and then took his hand to lead him towards the robot. Therefore, social robots can clearly play very functional and useful roles within society. Conclusion Robots are on the path towards becoming social and are finding new roles in society, such as helping to develop the social skills of autistic children and even taking the role of entertainer, be it actor or comedian. This essay has reviewed some of the essential skills which robots must be able to simulate in order for effective human robot social interaction to occur. Conveying emotions is a useful skill for robots to influence the behaviour of humans and in order to aid the robot to learn about social situations. Some simulation of social intelligence is also required although this does not necessarily need to be an accurate imitation of human social intelligence. Indeed, robots do not need to take a humanoid form in order for humans to ascribe social intentions to them as anthropomorphism leads humans to treat computers and robots in a similar fashion to other humans. Too much similarity to humans can cause problems as the expectations from the robot are not met. Robots who appear to pay attention to humans convey a stronger signal that there is an interaction taking place. However, further work is needed to develop robots who can encourage human interactions to continue over a long term period. These developments have set us on the path to discover new companions in robots in the future. Word Count: 4000 Page 9 of 12

10 References Bartneck, C. & Forlizzi, J. (2004a). A design centred framework for social human robot interaction. Proceedings of the Ro Man, Kurashiki. Bartneck, C. & Forlizzi, J. (2004b). Shaping human robot interaction: understanding the social aspects of intelligent robotic products. Proceedings of the CHI2004 Workshop, Vienna; ACM Press Bates, J. (1994). The role of emotion in believable characters. Communications of the ACM, 37, Breazeal, C. (2000). Infant like social interactions between a robot and a human caretaker. Breazeal, C. (2001) Affective interaction between humans and robots. Proceedings of the 2001 European Conference on artificial life. Springer Verlag: London. Breazeal, C. (2003). Towards sociable robots. Robotics and autonomous systems, 42, Breazeal, C. (2003). Emotion and sociable humanoid robots. International Journal of humancomputer studies, 59, Breazeal, C. (2004). Social interactions in HRI: the robot view. IEEE Transactions on systems, man, and cybernetics Part C: Applications and Reviews, 34, Breazeal, C. & Scassellati, B. (2000). Infant like social interactions between a robot and a human caretaker. Adaptive Behaviour, 8, Breazeal, C. & Scassellati, B. (2002). Robots that imitate humans. TRENDS in cognitive sciences, 6, Bruce, A., Nourbakhsh, I. & Simmons, R. (2002). The role of expressiveness and attention in humanrobot interaction. Proceedings of the 2002 IEEE International conference on robotics and automation. Washington, DC. Dautenhahn, K. (1995). Getting to know each other artificial social intelligence for autonomous robots. Robots and autonomous systems, 16, DiSalvo, C.F., Gemperle, F., Forlizzi, J. & Kiesler, S. (2002). All robots are not created equal: the design and perception of humanoid robot heads. Proceedings of the 4th conference on Designing interactive systems: processes, practices, methods, and techniques, London, England. Duffy, B.R. (2003). Anthropomorphism and the social robot. Robotics and Autonomous systems, 42, Duffy, B.R., Rooney, C.F.B., O Hare, G.M.P. & O Donoghue, R.P.S. (1999). What is a social robot?. 10 th Irish Conference on artificial intelligence and cognitive science. University College Cork; Ireland. Page 10 of 12

11 Fogg, B.J. & Nass, C. (1997). Silicon sycophants: the effects of computers that flatter. International Journal of human computer studies, 46, Gockley, R., Bruce, A., Forlizzi, J., Michalowski, M., Mundell, A., Rosenthal, S., Sellner, B., Simmons, R., Snipes, K., Schultz, A.C. & Wang, J. (2005). Designing robots for long term social interaction. Proceedings of IROS. Edmonton: Alberta. Gockely, R., Forlizzi, J. & Simmons, R. (2006). Interactions with a moody robot. Proceedings of Human robot interaction. Salt Lake City, Utah. Heider, F. & Simmel, M. (1944). An experimental study of apparent behavior. American Journal of Psychology, 57, Knight, H., Satkin, S., Ramakrishna, V. & Divvala, S. (2011). A saavy robot standup comic: online learning through audience tracking. International Conference on Tangible and Embedded Interaction, Funchal, Portugal. Nass, C., Fogg, B.J. & Moon, Y. (1996). Can computers be teammates? International Journal of Human Computer studies, 45, O Brien, B.M., Calius, E.P., Inamura, T., Xie, S.Q. & Anderson, I.A. (2010). Dielectric elastomer switches for smart artificial muscles. Applies Physics : Materials science & processing, 100, Okuno, H.G., Nakadai, K. & Kitano, H. (2002). Social interaction of humanoid robot based on audiovisual tracking. Proceedings on the 8 th international conference on industrial and engineering applications of artificial intelligence and expert systems. Springer: Cairns, Australia. Reeves, B. & Nass, C. (1996). The Media Equation: how people treate computers, television and new media like real people and places. Cambridge: Cambridge University Press. Riek, L.D., Rabinowitch, T., Chakrabarti, B. & Robinson, P. (2009). Empathizing with robots: fellow feeling along the anthropomorphic spectrum. Proceedings of the international conference on affective computing and intelligent interaction. Amsterdam, Netherlands. Robins, B., Dautenhahn, K., Te Boekhorst, R. & Billard, A. (2005). Robotic assistants in therapy and education of children with autism: can a small humanoid robot help encourage social interaction skills? Universal Access in the Information society, 4, Robins, B., Dickerson, P., Stribling, P. & Dautenhahn, K. (2004). Robot mediated joint attention in children with autism: a case study in human robot interaction. Interaction studies,5, Sabanovic, S., Michalowski, M.P. & Simmons, R. (2006). Robots in the wild: observing human robot social interaction outside the lab. IEEE Intl Workshop on Advanced Motion Control, Shawar, B.A. & Atwell, E. (2005). A chatbot system as a tool to animate a corpus. ICAME Journal, 29, Page 11 of 12

12 Sidner, C.L., Lee, C., Kidd, C., Lesh, N. & Rich, C. (2005). Explorations in engagement for humans and robots. Artificial intelligence, 166, Sundar, S.S. (2004). Loyalty to computer terminals: is it anthropomorphism or consistency? Behaviour and Information Technology, 23, Tinwell, A., Grimshaw, M., Abdel Nabi, D. & Williams, A. (2011). Facial expression of emotion and perception of the uncanny valley in virtual characters. Computers in human behaviour, 27, Turing, A. (1950). Computing machinery and intelligence. Mind, 49, Page 12 of 12