Abstract

The core principles of the evolutionary theories of emotions declare that affective states represent crucial drives for action selection in the environment and regulated the behavior and adaptation of natural agents in ancestrally recurrent situations. While many different studies used autonomous artificial agents to simulate emotional responses and the way these patterns can affect decision-making, few are the approaches that tried to analyze the evolutionary emergence of affective behaviors directly from the specific adaptive problems posed by the ancestral environment. A model of the evolution of affective behaviors is presented using simulated artificial agents equipped with neural networks and physically inspired on the architecture of the iCub humanoid robot. We use genetic algorithms to train populations of virtual robots across generations, and investigate the spontaneous emergence of basic emotional behaviors in different experimental conditions. In particular, we focus on studying the emotion of fear, therefore the environment explored by the artificial agents can contain stimuli that are safe or dangerous to pick. The simulated task is based on classical conditioning and the agents are asked to learn a strategy to recognize whether the environment is safe or represents a threat to their lives and select the correct action to perform in absence of any visual cues. The simulated agents have special input units in their neural structure whose activation keep track of their actual “sensations” based on the outcome of past behavior. We train five different neural network architectures and then test the best ranked individuals comparing their performances and analyzing the unit activations in each individual’s life cycle. We show that the agents, regardless of the presence of recurrent connections, spontaneously evolved the ability to cope with potentially dangerous environment by collecting information about the environment and then switching their behavior to a genetically selected pattern in order to maximize the possible reward. We also prove the determinant presence of an internal time perception unit for the robots to achieve the highest performance and survivability across all conditions.

Highlights

  • The neuroscientific interest in understanding the basis of emotional behavior has been rising in recent years, in particular since the fall of the classic “limbic system” theory, which asserted that limbic areas were uniquely involved in the mediation of emotions and separated from the neural structures dedicated to cognition [1]

  • Apart from the reward trend, the main investigation needs to be conducted on the behavioral strategies and the differences of performances emerged among the various architectures

  • Since the agents could use no visual cue for deciding whether the condition was dangerous or safe, but had to rely exclusively on their sensation units, it becomes important to evaluate how these activation values shaped into behaviors which led to scoring such a high reward and guaranteed an adaptation for both the conditions

Read more

Summary

Introduction

The neuroscientific interest in understanding the basis of emotional behavior has been rising in recent years, in particular since the fall of the classic “limbic system” theory, which asserted that limbic areas were uniquely involved in the mediation of emotions and separated from the neural structures dedicated to cognition [1]. The lack of a recognized methodology brings up the debate about how much the emotional features should or should not be included directly in the design of these systems, or, in other words, what is the level of engineering and that of spontaneous emergence when we measure their emotional response [5] Another issue which arises in the field of simulated agents is the lack of an agreed and reliable experimental paradigm for both a computational and a strictly neuroscientific investigation of the emergence of emotional behavior under an evolutionary perspective. While the former measures the emergence of an improvement in the result of actions over generations, the latter aims at simulating different decision making mechanisms in the organism’s life span

Methods
Results
Discussion
Conclusion
Full Text
Paper version not known

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call

Disclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.