Artificial Intelligence has been used to map the emotional expressions of Prime Minister Anthony Albanese and Opposition Leader Peter Dutton's during the final Leader's Debate on the ABC on 16 April.
Using advanced AI analysis of the debate transcript, a research team across Monash University, UNSW and Flinders University text-mined and then quantified the emotional patterns of the leaders' expressions during the debate.
The emotion scores were generated using a newly developed and a specially fine-tuned version of Meta AI's Llama 3 large language model.
Monash Business School Professor of Marketing, Stephan Ludwig, said the analysis shows Mr Dutton's emotional expressions leaned heavily on defensive emotions, including fear, shame and worry, while Mr Albanese's performance reflected significantly higher levels of optimism, with appeals to trust and commitment.
"Mr Dutton's expressions of anger (31 per cent of all speech turns during the debate), discontent (38 per cent) and worry (41 per cent) dominated his narrative relative to Albanese," said Professor Ludwig.
Professor of Marketing, Peter Danaher, who collaborated on the project, elaborated on these findings.
"Although Mr Dutton touched lightly on positive emotions such as contentment (22 per cent) and optimism (25 per cent), the overall emotionality was negative," said Professor Danaher.
"The majority of his comments were negative, with 12 per cent more expressions relating to distrust and 14 per cent more negative (rather than positive) sentiment.
"The expressions during his speech paint a picture of an Australia facing growing dangers, at home and abroad."
Supported by the Australian Research Council (ARC), the model was trained on real survey responses where participants both described their experiences and directly rated the emotions they felt, ensuring the model learned to detect emotions based on genuine human self-expressions.
Using this high-quality training data, the model was fine-tuned to accurately recognise specific human emotions and evaluations like trust, commitment and sentiment from any text.
Due to its sophisticated training, it outperforms major models like GPT-4 Turbo or DeepSeek in accuracy, achieving more than 80 per cent predictive performance across all emotions, say researchers.
For the debate, each candidate's individual speaking turns were analysed using the large language model, systematically extracting and quantifying their expressed emotions.
The resulting emotion use was then averaged across all speech turns each candidate had. The results tell a story as striking as the leaders' words themselves.
In contrast to Mr Dutton's emotional expressions, Professor Ludwig said the large language model detected Mr Albanese as presenting a more optimistic vision.
"His emotional expression palette featured significantly higher levels of optimism (46 per cent) and lower use of all negative emotions, except for worry, which was similarly expressed by both candidates at about 40 per cent of all of their speech turns," said Professor Ludwig.
The analysis found Mr Albanese overall made substantially more expressions signalling trust (14 per cent) and commitment 46 (per cent). His overall positive sentiment of plus-14 per cent stood in stark contrast to Mr Dutton's. Mr Albanese spoke of Australia's potential, emphasising unity and opportunity.
Interestingly, both leaders showed minimal joy and excitement, suggesting the electorate is being asked to choose between two very serious, if emotionally distinct, futures: one perhaps more anchored in confronting immediate dangers, and another perhaps more about a cautiously optimistic renewal.
As Australians head toward the polls, the choice may not just be about policies, it may also be about the emotional blueprint each leader offers for the nation's future.
"In 2025, emotions might just matter as much as economics," said Prof Ludwig.
