Abstract
Affective state is an integral component of the way animals, including humans, perceive and interact with their environment. Animals can be biased by their affective state in the way they act to receive rewards or avoid punishments.The Judgement Bias Task (JBT) is a decision-making task aimed at measuring biases in the interpretation of ambiguous information. In chapter 2, I initially ran replication studies (N=15 male rats) of past findings involving pharmacological manipulations of clinical importance, such as ketamine and amphetamine. After failed attempts to replicate published ketamine results, I designed a novel JBT variant to address shortcomings of the original task around the ambiguity of the test stimulus and the frequency of negative feedback and trial presentations. Pilot studies on this variant (N=16 male rats) revealed a different type of perceptual bias in the animals’ responses. I discuss how this bias confounded the interpretation of results and how it relates to the original task design.
By collating data from past JBT studies, I conducted a large-scale analysis, which revealed that factors relating to past trials were important in determining animals’ actions. Therefore, I designed statistical models that were able to account for these factors and any other biases in the animals’ behaviour. Inference by model parameters, instead of summary statistics of actions, grants a more detailed view into the animal’s decision-making process and reveals differences between the effects of ketamine and amphetamine.
Subsequently, I designed a novel foraging task, where animals were free to acquire reward or flee to avoid an imminent threat. Different versions of the task were tested in a pilot study (N=16 male rats). A statistical model reveals individual differences that become apparent when the threat was least predictable, as signaled by the constant presence of an odor in the operant chamber.
Finally, I present a theoretical model, based on reinforcement learning (RL) theory, which incorporates biases due to affective state. Simulated environments with naturalistic elements were also proposed. The model was compared to classical RL models within these environments to assess the benefits of affective biases. Overall, this thesis offers approaches to improve on how inference of affective state is performed, in addition to a hypothesis about why affective state is important for the survival of an animal.
| Date of Award | 6 Dec 2022 |
|---|---|
| Original language | English |
| Awarding Institution |
|
| Supervisor | Conor Houghton (Supervisor) & Emma S J Robinson (Supervisor) |