Abstract
dc:description.abstractThe holy grail of both AI and cognitive science is human-level intelligence. Whereas AI relies on computer algorithms to simulate human abilities, cognitive scientists investigate the brain to understand the underlying mechanisms. For most of human history, emotions were thought to be nothing more than disturbances for cognition. Therefore, they were usually ostracized from research on intelligence. As a result, cognitive architectures only partially include emotions in their design, if at all. Recently, though, it was discovered that emotions and cognition are in fact inter-dependent systems. Consequently, before being able to fully replicate human-level intelligence, it is necessary to understand the concept of emotions and its many roles within the brain. In this thesis, working around the lack of definition for emotion, I show that emotions inform the brain as to the nature of a given situation and guide the decision-making process, to increase the survival potential of virtual agents. In particular ProtoEmo, an architecture replicating the circuits found at the base of the forebrain, is shown to have the ability to detect stimuli relevant to the survival of virtual agents. Hence, it outperforms other emotional agents in terms of survival capabilities, which are measured by the size of the remaining population at the end of a resource foraging task. PrimEmo, the architecture born from the integration of ProtoEmo with standard models of the reward and decision-making systems in the brain, displays survival capabilities similar to the advantage actor-critic algorithm. PrimEmo also shows promises for supporting primitive emotions characterized by their level of arousal and valence. After further refinement, PrimEmo could replace the core decision-making module of a cognitive architecture, such as ACT-R or SOAR. Not only would it confer survival capabilities to the architecture, it would also allow for the possibility of investigating full-fledged emotions, and even emotional expression.
Degree
thesis:*- Name dc:type.qualificationname
- Ph.D
- Level dc:type.qualificationlevel
- doctoral
- Grantor dc:publisher.institution
- University of Sheffield
- Year dc:date.issued
- 2020
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Caspar, Luc
- Advisor dc:contributor.advisor
-
- Moore, Roger
- ORCID person:identifier.orcid
Rights
- Language dc:language.dcterms_iso639_2
- eng
Identifiers
dc:identifier.*- Identifier
- uk.bl.ethos.823924