Stefano Furlan, E. Medvet, Giorgia Nadizar, F. Pigozzi
{"title":"On the Mutual Influence of Human and Artificial Life: an Experimental Investigation","authors":"Stefano Furlan, E. Medvet, Giorgia Nadizar, F. Pigozzi","doi":"10.1162/isal_a_00492","DOIUrl":null,"url":null,"abstract":"Our modern world is teeming with non-biological agents, whose growing complexity brings them so close to living beings that they can be cataloged as artificial creatures, i.e., a form of Artificial Life (ALife). Ranging from disembodied intelligent agents to robots of conspicuous dimensions, all these artifacts are united by the fact that they are designed, built, and possibly trained by humans taking inspiration from natural elements. Hence, humans play a fundamental role in relation to ALife, both as creators and as final users, which calls attention to the need of studying the mutual influence of human and artificial life. Here we attempt an experimental investigation of the reciprocal effects of the human-ALife interaction. To this extent, we design an artificial world populated by life-like creatures, and resort to open-ended evolution to foster the creatures adaptation. We allow bidirectional communication between the system and humans, who can observe the artificial world and voluntarily choose to perform positive or negative actions towards the creatures populating it; those actions may have a shortor long-term impact on the artificial creatures. Our experimental results show that the creatures are capable of evolving under the influence of humans, even though the impact of the interaction remains uncertain. In addition, we find that ALife gives rise to disparate feelings in humans who interact with it, who are not always aware of the importance of their conduct. Introduction and related works In the 1990s, the commercial craze of “Tamagotchi” (Clyde, 1998), a game where players nourish and care for virtual pets, swept through the world. Albeit naive, that game is a noteworthy instance of an Artificial Life (ALife) (Langton, 1997), i.e., a simulation of a living system, which does not exist in isolation, but in deep entanglement with human life. It also reveals that ALife is not completely detached from humans, who might need to rethink their role and responsibilities toward ALife. We already train artificial agents by reinforcement or supervision: trained agents are notoriously as biased as the datasets we feed them (Kasperkevic, 2015), and examples abound1. For instance, chatbot Tay shifted from lovely to toxic communication after a few hours of interaction with users of a social network (Hunt, 2016). The https://github.com/daviddao/awful-ai field of robotics is no exception to the case, and while robots, a relevant example of ALife agents, are becoming pervasive in our society, we—the creators—define and influence them (Pigozzi, 2022). One day in the future, a robot could browse for videos of the very first robots that were built, eager to learn more about its ancestors. Suppose a video shows up, displaying engineers that ruthlessly beat up and thrust a robot in the attempt of testing its resilience (Vincent, 2019). How brutal and condemnable would that act look to its electric eyes? Would our robotic brainchildren disown us and label us “a virus” as Agent Smith (the villain, himself an artificial creature) does in the “Matrix” movie (Wachowski et al., 1999)? At the same time, how would such responsibility affect the creators themselves? Broadly speaking, when dealing with complex systems involving humans and artificial agents, whose actions are deeply intertwined, what results from the mutual interaction of humans and ALife? In particular, do artificial agents react to the actions of humans, displaying short-term adaptation in response to stimuli? Do these actions influence the inherited traits of artificial creatures, steering their evolutionary path and long-term adaptation? And, conversely, are humans aware of their influence on ALife? Do they shift their conduct accordingly? We consider a system that addresses these questions in a minimalist way. We design and implement an artificial world (Figure 1), populated by virtual creatures that actively search for food, and expose it to a pool of volunteer participants in a human experiment. We consider three design objectives: (a) interaction, that is bidirectional between human and ALife; (b) adaptation, of creatures to external stimuli, including human presence; (c) realism, of creatures to look “familiar” and engaging for participants. Participants interact with the creatures through actions that are either “good” (placing food) or “bad” (eliminating a creature): we then record the participants’ reactions. At the same time, creatures can sense human presence. We achieve long-term adaptation through artificial evolution, and, for the sake of realism, we design the creatures to be life-like. As a result, the goodness or badness of human actions can potentially Figure 1: Our artificial world: worm-like agents are creatures that search for food (the green dots). affect the evolutionary path of creatures, as well as their relationship with humans. Humans, on the other side, can feel emotions in the process. Participants thus play the role of a “superior being”, absolute from any conditioning authority (Milgram, 1963), with power of life and death upon the creatures. Whether their actions will be good or bad is up to them: a philosophical debate on human nature that goes back to Thomas Hobbes (1651) and Jean-Jacques Rousseau (1755), with their opposing views propagating through history. Other studies crafted artificial worlds, e.g., Tierra (Ray, 1992), PolyWorld (Yaeger et al., 1994), and Avida (Ofria and Wilke, 2004), with several different goals: they mostly investigate questions related to evolutionary biology (Lenski et al., 2003), ecology (Ventrella, 2005), open-ended evolution (Soros and Stanley, 2014), social learning (Bartoli et al., 2020), or are sources of entertainment and gaming (Dewdney, 1984; Grand and Cliff, 1998). Albeit fascinating, none of these addresses the main research question of this paper, i.e., the mutual influence of human life and ALife. Our work also differs from multi-agent platforms, whose focus is on optimizing multi-agent policies for a task (Suarez et al., 2019; Terry et al., 2021). The work that is the most similar to ours pivots around the “Twitch Plays Robotics” platform of Bongard et al. (2018). While paving the way for crowdsourcing robotics experiments, it is, rather than an artificial world, an instance of “interactive evolution” (with participants issuing reinforcements to morphologically-evolving creatures), and does not detail the influence of creatures on participants. We instead concentrate on the bidirectionality of interaction, and branch into two complementary studies: the first aimed at quantifying the effects of human interaction on artificial creatures, and the second focused on surveying how humans perceive and interface themselves with ALife. Concerning the former, we simulate human actions on the system and analyze the progress over time of some indexes, whereas for the latter we perform a user study involving a pool of volunteer participants interacting with the creatures. The experimental results confirm the importance of focusing on the bidirectionality of human-ALife interaction, and open a way towards more in depth analyses and studies in the field. Not surprisingly, we find that an artificial world subjected to human influence is capable of evolving, yet the real impact of human behavior on it, be it positive or negative, remains enigmatic. In addition, we discover two main currents of thought among people who interface themselves with ALife: those who feel involved and are aware of the consequences of their actions on an artificial world, and those who perceive ALife as a not attention-worthy farfetched artifact. The artificial world","PeriodicalId":309725,"journal":{"name":"The 2022 Conference on Artificial Life","volume":"1059 1","pages":"0"},"PeriodicalIF":0.0000,"publicationDate":"1900-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"The 2022 Conference on Artificial Life","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/10.1162/isal_a_00492","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"","JCRName":"","Score":null,"Total":0}
引用次数: 0
Abstract
Our modern world is teeming with non-biological agents, whose growing complexity brings them so close to living beings that they can be cataloged as artificial creatures, i.e., a form of Artificial Life (ALife). Ranging from disembodied intelligent agents to robots of conspicuous dimensions, all these artifacts are united by the fact that they are designed, built, and possibly trained by humans taking inspiration from natural elements. Hence, humans play a fundamental role in relation to ALife, both as creators and as final users, which calls attention to the need of studying the mutual influence of human and artificial life. Here we attempt an experimental investigation of the reciprocal effects of the human-ALife interaction. To this extent, we design an artificial world populated by life-like creatures, and resort to open-ended evolution to foster the creatures adaptation. We allow bidirectional communication between the system and humans, who can observe the artificial world and voluntarily choose to perform positive or negative actions towards the creatures populating it; those actions may have a shortor long-term impact on the artificial creatures. Our experimental results show that the creatures are capable of evolving under the influence of humans, even though the impact of the interaction remains uncertain. In addition, we find that ALife gives rise to disparate feelings in humans who interact with it, who are not always aware of the importance of their conduct. Introduction and related works In the 1990s, the commercial craze of “Tamagotchi” (Clyde, 1998), a game where players nourish and care for virtual pets, swept through the world. Albeit naive, that game is a noteworthy instance of an Artificial Life (ALife) (Langton, 1997), i.e., a simulation of a living system, which does not exist in isolation, but in deep entanglement with human life. It also reveals that ALife is not completely detached from humans, who might need to rethink their role and responsibilities toward ALife. We already train artificial agents by reinforcement or supervision: trained agents are notoriously as biased as the datasets we feed them (Kasperkevic, 2015), and examples abound1. For instance, chatbot Tay shifted from lovely to toxic communication after a few hours of interaction with users of a social network (Hunt, 2016). The https://github.com/daviddao/awful-ai field of robotics is no exception to the case, and while robots, a relevant example of ALife agents, are becoming pervasive in our society, we—the creators—define and influence them (Pigozzi, 2022). One day in the future, a robot could browse for videos of the very first robots that were built, eager to learn more about its ancestors. Suppose a video shows up, displaying engineers that ruthlessly beat up and thrust a robot in the attempt of testing its resilience (Vincent, 2019). How brutal and condemnable would that act look to its electric eyes? Would our robotic brainchildren disown us and label us “a virus” as Agent Smith (the villain, himself an artificial creature) does in the “Matrix” movie (Wachowski et al., 1999)? At the same time, how would such responsibility affect the creators themselves? Broadly speaking, when dealing with complex systems involving humans and artificial agents, whose actions are deeply intertwined, what results from the mutual interaction of humans and ALife? In particular, do artificial agents react to the actions of humans, displaying short-term adaptation in response to stimuli? Do these actions influence the inherited traits of artificial creatures, steering their evolutionary path and long-term adaptation? And, conversely, are humans aware of their influence on ALife? Do they shift their conduct accordingly? We consider a system that addresses these questions in a minimalist way. We design and implement an artificial world (Figure 1), populated by virtual creatures that actively search for food, and expose it to a pool of volunteer participants in a human experiment. We consider three design objectives: (a) interaction, that is bidirectional between human and ALife; (b) adaptation, of creatures to external stimuli, including human presence; (c) realism, of creatures to look “familiar” and engaging for participants. Participants interact with the creatures through actions that are either “good” (placing food) or “bad” (eliminating a creature): we then record the participants’ reactions. At the same time, creatures can sense human presence. We achieve long-term adaptation through artificial evolution, and, for the sake of realism, we design the creatures to be life-like. As a result, the goodness or badness of human actions can potentially Figure 1: Our artificial world: worm-like agents are creatures that search for food (the green dots). affect the evolutionary path of creatures, as well as their relationship with humans. Humans, on the other side, can feel emotions in the process. Participants thus play the role of a “superior being”, absolute from any conditioning authority (Milgram, 1963), with power of life and death upon the creatures. Whether their actions will be good or bad is up to them: a philosophical debate on human nature that goes back to Thomas Hobbes (1651) and Jean-Jacques Rousseau (1755), with their opposing views propagating through history. Other studies crafted artificial worlds, e.g., Tierra (Ray, 1992), PolyWorld (Yaeger et al., 1994), and Avida (Ofria and Wilke, 2004), with several different goals: they mostly investigate questions related to evolutionary biology (Lenski et al., 2003), ecology (Ventrella, 2005), open-ended evolution (Soros and Stanley, 2014), social learning (Bartoli et al., 2020), or are sources of entertainment and gaming (Dewdney, 1984; Grand and Cliff, 1998). Albeit fascinating, none of these addresses the main research question of this paper, i.e., the mutual influence of human life and ALife. Our work also differs from multi-agent platforms, whose focus is on optimizing multi-agent policies for a task (Suarez et al., 2019; Terry et al., 2021). The work that is the most similar to ours pivots around the “Twitch Plays Robotics” platform of Bongard et al. (2018). While paving the way for crowdsourcing robotics experiments, it is, rather than an artificial world, an instance of “interactive evolution” (with participants issuing reinforcements to morphologically-evolving creatures), and does not detail the influence of creatures on participants. We instead concentrate on the bidirectionality of interaction, and branch into two complementary studies: the first aimed at quantifying the effects of human interaction on artificial creatures, and the second focused on surveying how humans perceive and interface themselves with ALife. Concerning the former, we simulate human actions on the system and analyze the progress over time of some indexes, whereas for the latter we perform a user study involving a pool of volunteer participants interacting with the creatures. The experimental results confirm the importance of focusing on the bidirectionality of human-ALife interaction, and open a way towards more in depth analyses and studies in the field. Not surprisingly, we find that an artificial world subjected to human influence is capable of evolving, yet the real impact of human behavior on it, be it positive or negative, remains enigmatic. In addition, we discover two main currents of thought among people who interface themselves with ALife: those who feel involved and are aware of the consequences of their actions on an artificial world, and those who perceive ALife as a not attention-worthy farfetched artifact. The artificial world