Objective: The aim of this work was to analyze discussions on Twitter related to Montreal AI’s second AI-debate and to investigate the emotions towards AI (including deep learning, and symbolic AI) and the debate.
Methods: All tweets written in English, before and after the AI-debate (December 14, 2020 until January 2, 2021) were included. The tweets which had the terms “aidebate”, “aidebate2” or “ai debate” were extracted using the Twitter API. The tweets by the organizers were removed to reduce the potential bias. The tweets related to another AI debate which occurred on December 14, 2020 were also removed. The analysis was done on the remaining of the tweets. All the raw data and processed tweets were saved in a password protected laptop.
Emotion analysis and geographic analysis of the included tweets were conducted. The word cloud of the tweets was also developed to get a visual estimation of the frequency of the words. We applied machine learning methods (including recurrent neural networks) to analyze the collected data. Emotion-classifier models of Colneric et al. [1] (Ekman's six basic emotions and Plutchik's eight basic emotions) were used.
Results: There were a total of 5,520 likes, 379 replies, and 4,469 mentions in tweets during the study timeframe. Out of 2,741 tweets analyzed, emotion analysis classified the following emotions consecutively as the highest ones during the study timeframe: (1) joy, (2) trust, and (3) fear (according to Plutchik's emotion model, Figure 1); and (1) joy, (2) fear, and (3) surprise (according to Ekman's emotion model, Figure 2). A similar work was conducted last year (link: https://github.com/rahimi-s-lab/Twitter-AI-Debate). The number of tweets this year were more (n= 2,741 comparing to n= 450), and we obtained more information pertaining to each tweet as we used the Twitter API (Figures 10-11). Further details on our analysis as well as details on the comparison of the debates (AI debate 1 and AI debate 2) provided below.
Authors: Arka Mitra, Dr. Samira Rahimi