Harvard scholars propose AI-driven political polling to improve accuracy
A team of experts from Harvard's Ash Center for Democratic Governance and Innovation, has suggested that artificial intelligence (AI) could be used to enhance the accuracy of political polling. The proposal is an answer to the low response rate of political polling calls, which currently stands at just 6%, according to Pew Research Center's 2019 poll. The scholars believe this figure presents an opportunity for an algorithmic future in the polling industry.
Potential and current limitations
The Harvard researchers have acknowledged potential inaccuracies in AI, such as instances of chatbots "hallucinating." However, they believe these issues will diminish "over time" as AI improves its ability to anticipate human responses, and recognize when it is likely to be most wrong or uncertain. A study published by the Harvard Data Science Review last year found that ChatGPT generally responded to typical polling questions in a manner similar to humans.
AI polling not fundamentally different from traditional methods
The Harvard experts argue that AI polling is not fundamentally different from traditional polling methods. They stated, "while AI polling will always have limitations in accuracy," it is similar to traditional polling which also faces challenges, such as reaching large enough sample sizes, and dealing with nonresponse and inauthentic responses. The researchers believe that these challenges can be addressed as AI technology continues to evolve.
Potential misuse of AI in political polling
Despite the potential benefits of using AI in political polling, there are concerns about its misuse during elections. This includes deepfakes and disinformation campaigns, similar to those seen during Prime Minister Narendra Modi's re-election. The Harvard scholars acknowledge these concerns but maintain that AI polling is a viable option for improving accuracy in political surveys.