The influence of AI has been one of the most talked about topics ahead of elections. With general elections underway in India, the discourse surrounding AI’s impact has been largely limited to its downsides, especially with the spate of misinformation and harmful deepfakes. However, now new research has thrown light on how AI can affect positive outcomes in future elections in India.
The research, ‘An AI experiment on Augmented Democracy for India’, conducted by the new-age think tank, Policy 4.0 brings forth an array of possibilities with AI. The research shows how AI could prove to be a great means to help voters make informed decisions. The new research not only explores the power of AI in encouraging data-driven voting behaviour but also sees it as a way to mitigate political polarisation. The researchers created AI twins of sample voters to conduct the experiment.
“Discussions around elections and AI are heavily focused on deepfakes and misinformation, highlighting the negative implications of AI. However, there are also interesting positive use cases that have been explored for some time in policy phases, but no Indian experiment has been conducted on them. This is what we set out to do. This experiment ran parallel to the election. It is essentially an experiment on augmented democracy,” Tanvi Ratna, founder of Policy 4.0. and lead architect of the study, told indianexpress.com.
Ratna explained that Augmented Democracy or (AD) is the idea of using digital twins to expand the ability of people to participate directly in a large volume of democratic decisions.
What is the research about?
The researchers deployed large language models (LLMs) to create digital twins, also known as virtual representations of voters based on their political preferences. The creation of these AI twins is at the core of this experiment. The twins essentially simulated personas of voters using data collected from the survey responses. Later, the AI twins analysed and processed vast datasets which included historical economic performance data and policy outlooks of political parties. This was later used to come up with personalised recommendations that were aligned with the voters’s beliefs and concerns.
“We are seeing significant information asymmetry, disinformation, and misinformation due to the widespread use of the internet. This is especially evident during political campaigns when distinguishing between true and false information is crucial for decision-making, like voting. I believe AI can be beneficial in organising data and helping people make informed decisions, despite its current negative perception,” said Prabhu Pradhan, AI scholar at Policy 4.0 and lead researcher, while explaining the motivation behind the research.
According to the team, the findings of the study are particularly relevant for India considering that it is the world’s largest democracy. The report highlights that over 15 million first-time voters were added to the electoral roll for the 2019 general elections in India. This shows the importance of exploring AI’s role in enhancing the involvement of younger tech-savvy citizens in elections.
One of the unique aspects of the research is the AI twin’s ability to mimic voter choices with accuracy. Based on the report, the AI models showed a 90 per cent success rate in predicting how a voter would cast their ballot based on the set parameters such as core beliefs, political ideology, and issue priorities. It is to be noted that this high level of accuracy indicates AI’s potential to understand and interpret complex voter preferences and this makes way for more personalised and tailored electoral information and recommendations.
Further, the study also found how AI twins could potentially solve the problem of political polarisation. The researchers said that by analysing the confidence levels of voters and their AI twins, they were able to observe a significant decrease in polarisation among high-confidence voters – those voters who are usually assumed to be more prone to partisan biases. During the experiment, the AI twins showcased a tendency to moderate extreme viewpoints, offering more balanced recommendations that could allow voters to consider diverse perspectives and somehow ease the influences of misinformation and echo chambers.
Another aspect of the research was the study of ‘homo economicus’, another variant of the AI twin. This variant of AI twin focusses on economic priorities and performance data offering insights into how voters may approach decision-making if guided with the help of economic indicators.
What are the prospects?
The research shows the potential of AI technologies like LLMs to create personalised ‘AI twin’ assistants for voters. In the future, this could lead to the development of AI-powered voter advice applications and personalised election information systems. These tools will be able to aggregate data on a candidates’ backgrounds, their policy positions, and fact-check and may offer customised overviews to reach voters based on their beliefs and priorities. If executed, this can counter misinformation and help voters make choices that are aligned with their values.