Emotions are important cognitive faculties that enable animals to behave intelligently in real time. The author argues that many important current and future applications of autonomous robots will require them to have a rich emotional repertoire, but this raises the question of whether it is possible for robots to experience their emotions consciously, as people do. Under what conditions would phenomenal experience of emotions be possible for robots? This is, in effect, the “hard problem” of robot emotions. This paper outlines a scientific approach to the question grounded in experimental neurophenomenology.
TopIntroduction
In this paper, I focus on the hard problem of robot emotions: the possibility and preconditions for a robot experiencing its emotions, and on its implications for the ethical treatment of robots. Everyday notions of ethical treatment depend in part on the recipient’s capacity to suffer, which includes pain, but goes beyond it, to include feelings of distress, agony, sorrow, anguish, and loss. From the opposite perspective, ethical treatment also involves the capacity to experience joy or well-being, but I will not address positive emotions in this paper. Future autonomous robots’ capacity to feel will affect not only our treatment of them, but also their treatment of us. For we expect robots to treat us well, but that will be more likely if they can empathise with our feelings of suffering and joy. This capacity is more compelling if it goes beyond intellectualised empathy and includes empathetic feeling (such as we have via mirror neurons). But why should we equip robots with emotions at all?
An emotion may be defined as a state that is evoked by a reward or a punishment, which might be either present or remembered, and that serves as positive or negative reinforcement (Rolls, 2007). This reinforcement leads to changes in behaviour that are adaptive in the sense of inclusive fitness (Plutchik, 2003). Certainly, many robots will not need emotions, but animals depend on emotions for efficient, real-time behaviour, and for analogous reasons we can expect them to be valuable in some autonomous robots. More specifically, Rolls (2005, 2007) enumerates a number of functions fulfilled by emotions, which have analogues in robotics. First, emotion is motivating and directed toward action, and likewise robots need a means for selecting goals and organising subservient activities. Also, natural emotions provide for response flexibility through a “bow tie” organisation. That is, many different stimuli may lead to a single behavioural goal, represented by the emotion, which can be achieved by a variety of means (Rolls, 2006). Likewise in robots, it is useful to identify general motivational states that can be triggered by a variety of stimuli and fulfilled in a variety of ways. Further, in animals an emotion establishes a persisting state (e.g., a mood) that biases cognitive processing to be more appropriate to the situation, and emotions can serve the same purpose in robots. In particular, emotion is crucial in memory encoding and rapid retrieval of behaviourally relevant information, which is valuable in robots as well. Further, natural emotions trigger autonomic and endocrine responses, which affect adrenaline release, heart rate, and other functions. Analogously, robot emotions might affect power management, reallocation of computational resources, adjustment of clock rates, preparatory deployment and priming of sensors and actuators, and so forth. Finally, emotions are critical in regulating interactions among animals, promoting cooperation and other forms of social organisation, and in facilitating communication of mental states, attitudes, intentions, etc. through emotional expression. These functions are also important in robots that cooperate with each other or with humans (Breazeal, 2003; Breazeal, Brooks, Gray, Hoffman, Kidd, Lee, Lieberman, Lockerd & Chilongo, 2004).