Mood Bias in AI Text: The Psychology of AI Large Language Models such as ChatGPT, GPT-4 and Bard.
Geoff Davis and Mick Grierson
Director of Studies: Professor Mick Grierson
Four psychometric text experiments to find and explore Mood Bias - positive emotional tone in AI text generation
Experiments
This is a multi-mode four part study to analyse the emotional tone of generated texts, with:
Robot Rorschach - image captioning techniques, based on Rorschach’s famous inkblots
stage setting or jailbreaking used to place the AI into therapeutic scenarios
narrative continuation prompts
humanistic counselling approaches used for text prompts
Aims of experiments
Analyse emotional terms used in long form language of recent AI large language models, in order to gauge the emotional tone of the outputs
Use Natural Language Processing NLP to analyse sentiment and details to find patterns
Find variations when using different generators
For more details, email geoffdavis5 AT gmail com