Misinformation or Collaborative Cognition? Examining the Role of Bots in Manipulating Political Thought in Twitter Discourse

doi:10.21203/rs.3.rs-4824255/v1

Download PDF

Research Article

Misinformation or Collaborative Cognition? Examining the Role of Bots in Manipulating Political Thought in Twitter Discourse

https://doi.org/10.21203/rs.3.rs-4824255/v1

This work is licensed under a CC BY 4.0 License

You are reading this latest preprint version

Artificial intelligence (AI) enhances human communication but also complicates information sharing in online political discourse. This paper empirically investigates AI-generated tweets’ impact on political discourse on Twitter. It analyzed 4,582 tweets on a contentious Canadian political topic in 2019, segmented into bot and human sub-corpora. Using WMatrix5, each corpus underwent semantic tagging across 21 discourse fields and 232 sub-classifications for statistical analysis. Spearman’s rank correlation coefficient showed strong positive correlations between bot-generated and human-generated tweets on March 14, 2019 (r(8) = 0.87, p = .001), March 28–29, 2019 (r(8) = 0.87, p = .001), and April 8–9, 2019 (r(8) = 0.91, p < .001). A linear regression model demonstrated that the bot-generated corpus significantly predicted variance in the semantic content of human-generated tweets, suggesting predictive influence of bot posts on human discourse.

interactive alignment theory

political discourse

social media

misinformation

hashtags

Twitter

tweets

corpus analysis

As social beings, we are inherently inclined to calibrate our emotions, thoughts, and behaviours with those around us, especially when we share an affinity with one another through common beliefs, values, and worldviews. Human minds are designed for social cognition and are thus highly adept at a collective behaviour called ‘shared intentionality,’ which is our unique ability to engage in collaborative interactions through which we share psychological states with one another (Gilbert, 1989; Searle, 1995; Tuomela, 1995; Tomasello 2007). In fact, Vygotsky proposed that what distinguishes human cognition from that of other species is not our individual intellectual abilities, but rather to our capacity to cognate collaboratively with others and to enhance our intellectual agility through that collaboration (noted by Tomasello, 1999; Tomasello et al., 2005; Tomasello, Kruger & Ratner, 1993). Our socially oriented mode of thinking is something that we share in common with primates like the chimpanzee, but human attention goes well beyond simply noticing and looking toward what others in our proximity are gazing at; we also want to share our intention with others by ‘being in the moment with them.’ Joint attention is not only about experiencing the same thing with others at the same time, but also about our awareness that we are doing it together (Tomasello, 1995). This unique ability has been critical to the success of our collaborative activities, like using language to communicate, inventing sophistical symbols to perform sophisticated mathematical calculations, inventing complex social institutions to instill greater degrees of intersubjective cooperation, among many other human accomplishments across history. Our ability to share common psychological ground with others has been paramount to our strength, fitness, and success as a species.

The invention of machine learning and generative AI are part of our intellectual and collaboratively inventive legacy as a highly sentient species, but it is a technological invention that, paradoxically, is currently challenging our inherent collective strength as highly cooperative and collaborative beings. Machine learning has been developed and adapted to assess users’ personality according to their digital footprint, and research conducted at the University of Cambridge has found that “private traits and attributes are predictable from digital records of human behaviour” (Kosinski, 2013). The machine learning component of assessing personality is informed by a long-standing technique for measuring our psychology called ‘psychometrics,’ which involves measurement of people’s emotional and mental traits, including intelligence, personality, skills. Psychometric machine learning is designed to make statistical predictions about the relationship between our online activities and our personality, and generative AI factors that data into its own algorithmic protocol when curating content and information that is attuned to our unique personality profile (Heuer, 2020). The integration of these digital activities in online spaces are not inherently problematic, but they become problematic when they are “used to covertly exploit weaknesses in their character and persuade them to take action against their own best interest” (Matz et al., 2017).

Psychometric machine learning and generative AI have become conventional instruments in electioneering toolboxes used for personalizing political communication and for micro-targeting voters based on predictions about the personality. While these digital tactics are certainly disconcerting, they are only part of the explanation for why they are so effective in circulating ‘fake news’ and misinformation while generating filter bubbles and echo chambers. The discrete presence of these digital activities in online environments makes discerning the difference between content created by generative AI and from content produced by fellow human peers exceedingly difficult, and our social brains become vulnerable to the seductive allure of like-minded representations that appear to be aligned with our values, beliefs, and worldviews.

Increasingly, political ideologies and public opinion are shaped online amidst the automated digital processes that operate beneath our awareness on social media platforms like Twitter. Campaigning candidates have capitalized on the strategic value of integrating psychometric machine learning and generative AI into their online electioneering strategies. This approach allows them to inject targeted messaging discretely into online discussions based on predictions about its receptivity generated by machine learning. In Canada, both the Liberal and Conservative Parties have utilized the services of data analytics firms during electoral campaigns to automate the circulation of targeted messaging according to constituents’ unique personality profile (Kingston, 2017). The challenge, of course, is not that we are exposed to the digital electioneering techniques of psychometric machine learning and generative AI, but that our social psychology is designed to respond favourably to others who share common values, beliefs, and worldviews, and generative AI can be designed to simulate human activity within online spaces that are statistically informed by our individualized psychometric profile. The sense of affinity that is cultivated between us and the content curated for us by a generative AI helps calibrate our attention and intention toward a shared political goal or agenda, especially when we experience a sense of belonging together.

This affinity is often referred to as ‘interactive alignment’ among social psychologists, and it is used to describe the ways in which individuals attune their communication behaviours and styles to those with whom they interact (Garrod & Pickering, 2006, 2007, 2009, 2011, 2013, 2014; Gallotti et al., 2017; Menenti et al., 2012; Rassenberg et al., 2020). In fact, our social predilection for interactive alignment is so pronounced that the phenomenon has also been observed in human-computer interactions (Branigan & Pearson, 2006; Branigan et al., 2010; Branigan, Pickering & Pearson, 2003; Shen & Wang, 2023). The significance of this intersubjective behaviour to our everyday interactions and interpersonal success may be most pronounced in the context of a job interview, during which time we are often inclined to mirror the speed, volume, and tone of our interviewer’s speech in effort to strike a connection. We may also adapt our non-verbal and emotional cues, such as eye contact, open body language, and expressing enthusiasm for the company to cultivate a sense of mutuality and trust. We engage in interactive alignment with others so effortlessly it can feel as natural as breathing, and this propensity, anchored in our social psychology, makes us cognitively and emotionally vulnerable to electioneering tactics used on social media that leverage the statistical operations of psychometric machine learning and generative AI.

According to a 2017 Oxford working paper on computational propaganda, McKelvey and Dubois (2019) note that political parties often employ machine learning and generative AI during elections to micro-target constituents with personalized communications, to craft online messaging strategies based on the psychometric profile of targeted voters, and to artificially enhance perceived popularity of specific ideological messages. Political representatives frequently use social media to propagate political ideologies, and politicians are especially active on Twitter. In Canada, 90% of them post tweets monthly, with 55% posting weekly (Hughes, 2019). They often employ bots to disseminate content that supports their agendas or to criticize opposing points of view. On the receiving end, 30% of Canadians aged 18-35 rely on online news sources (Maru Group, 2022), and 62% of those aged 15-24 get news specifically from social media sites (Statistics Canada, 2023). Research from Toronto Metropolitan University’s Social Media Lab (Dubois et al., 2018) reveals that 34% of young Canadians (18-24) and 28% of Canadians (25-34) share their political opinions on social media monthly. Given that social media interactions between politicians and constituents are heavily mediated by AI algorithms, political ideologies emerge at the intersection of political strategy, algorithmic computation, and interpersonal dynamics on social media, fostering online environments that are likely to induce collaborative cognitive among humans and AI-mediated bots.

This paper examines potential instances of interaction alignment between human users and bots – automated software programs designed to engage in online interactions that simulate human behaviour on social media. With a specific focus on Twitter (now operating at X), the central hypothesis investigated in this paper posits that the psychometric machine learning and generative AI underlying bot activity on social media platforms like Twitter compel human users to interactively align with the conceptual value of political content posted by Twitterbots.

It also considers the potential impact of integrating AI technologies for disseminating political information on each citizen’s freedom of conscience and judgment, as outlined in the Canadian Charter of Rights and Freedoms (Canadian Charter of Rights and Freedoms, 1982, s 15). Supreme Court Justice Antonio Lamer, for instance, emphasized that these rights are crucial to our democratic tradition, stating, “…at the heart of our democratic political tradition. The ability of each citizen to make free and informed decisions is the absolute prerequisite for the legitimacy, acceptability, and efficacy of our system of self‑government” (Dickson, 1985: Supra, n. 10 at 353-354; 361-362). In light of this observation, this paper argues that covert manipulation of political thought through algorithmic computations threatens the foundational principles of Canada’s democratic institutions, which uphold the legitimacy, acceptability, and effectiveness of our electoral processes.

Interactive alignment theory has informed research on the influence of computer-generated text on human users but has not yet been used to study the impact of machine learning and generative AI on political thought on social. Moreover, the phenomenological link between interactive alignment and collaborative cognition as drivers of AI-mediated bots’ success in influencing political thought on social media has yet to be fully investigated.

Given that interaction alignment and collaborative cognition are latent variables dependent on quantifiable data, this study employs computational corpus analysis tools to categorize linguistic units in tweets into quantifiable semantic categories. These units are used to assess the degree of association between statistically significant language use patterns embedded in tweets posted on Twitter by humans and bots using the same hashtag. This paper thus seeks to explore the following research questions:

Q1. Do tweets posted by Twitterbots compel human users to interactively align with that content?

Q2. Can interaction alignment theory elucidate the collaborative cognitive processes
between human users and AI algorithms in shaping political discourse on social media?

Q3. What are the ethical implications of AI-mediated political influence on social media and its implications for democratic principles, as framed by the Canadian Charter of Rights and Freedoms?

Given increasing concern that exposure to AI-mediated misinformation online undermines public trust in media and confidence in institutions (Statistics Canada, 2023), empirically analysis exploring these questions using a mixed-methods approach will contribute significantly to existing bodies of literature as well as our general understanding of the political and psychological properties of bot-generated content on social media.

Textual content is the data most commonly used for studying human-computer interactions on social media and is frequently studied to reveal insights about psychological processes that unfold in relation to language production (Hancock, Woodworth, & Porter, 2011). Scholars typically employ diverse methods for this line of research, including the implicit association test (Caliskan et al., 2017), word frequency analysis (Stone & Hakan Can, 2020), sentiment analysis (Rodríguez-Ibánez et al., 2023; Rahman et al., 2023), and computational corpus analysis (Breeze, 2015; Walker et al., 2023) to examine how language use represents underlying psychological, social, and political phenomena.

Hashtag analysis is another prominent method used alongside computational corpus analysis to help reduce the effect of ‘noisy’ or irrelevant data collected online and to catalogue digital traces of human and bot activities on social media (La Rocca & Boccia Artieri, 2022; Nur et al., 2024). This is a common method used to classifying tweets into common political themes – a process termed ‘topic modeling’ – while at the same time, allowing for the quantification of those themes as abstract variables over time (Freelon et al., 2016). Henry et al. (2018), for instance, have developed a data cleaning method while using hashtags for social media research that involves a computational contextual filtering technique that examines the relationship between a target hashtag and other keywords and hashtags appearing in proximity to the target in order to identify outliers and filter out irrelevant information.

Researchers also often employ computational corpus analysis and statistical models to quantify the semantic value, meaning, and sentiments of social media data (Breeze, 2015, 2019, 2020), allowing for more empirical analysis of online political discourse. For example, computational tools like WMatrix apply machine learning to semantically tag each word making up a corpus of text, which, as Ruth Breeze has noted (2019) “…is a useful aid for analysing large samples of data, and makes it possible to quantify phenomena reliably using normalised frequencies, and run comparisons between corpora.” In a seminal paper, Breeze (2020) applied WMatrix4’s semantic tagging feature to examine the tweets of prominent politicians in the UK to assess the emotional tenor of the language used in those tweets, finding that ‘anger’ was a significant emotion that evokes high power/control valence that often spurs people into action, especially in populous discourse.

Statistical models for content analysis also make the quantification of trends in the use of linguistic units across AI-mediated political communications possible, enhancing robust identification of temporal variations in themes across a corpus of social media texts (Stromer-Galley et al., 2021) using regression analysis to validate findings (Barberá et al., 2015). Methods for sentiment analysis and opinion mining, for instance, have been optimized by Latent Dirichlet Allocation (LDA), which is a statistical model used to reveal important thematic structures and patterns in large corpora of texts that may have otherwise gone unnoticed (Blei & Lafferty, 2012).

These studies have greatly enhanced our understanding of AI-mediated infrastructures, the value of linguistic data for machine learning and generative AI, and the political behaviours that shape public discourse about important political events on social media. Yet, gaps remain in our understanding how interactions with AI-mediated content affect public perception of political information at the cognitive level. This paper advocates for a focused analysis of interaction alignment and collaborative cognition — the interaction between human minds and AI on social media – within human-computer interactions on Twitter. As Joanna Bryson (2015), an expert on AI ethics, has emphasized, there is a need to explore how AI impacts political relationships and governance (Gallotti & Huebner, 2017), highlighting the significance and relevance of this research.

3.1. Broadcasting Political News

Digital media disseminates information in vastly different ways compared with traditional broadcast media. During earlier eras across the twentieth century, political news flowed unidirectionally from a broadcasting source (e.g., television, radio, newspapers, etc.) to the masses who had little to no input. In the digital era, political information converges within an interactive confluence of data emanating from several disparate sources, including human users, political actors, newscasting agencies, and, of course, AI-mediated algorithms. In recent years, a subset of machine learning called ‘deep learning’ has been incorporated into several digital processes involved in the mediation of information on social media platforms like Twitter. These subsets of machine learning use virtual ‘neural networks’ that model the complex problem-solving capacities of human brains, which accommodate a host of automated processes on social media, including content recommendations, image and video recognition, natural language processing, user engagement and interactions, trend and influencer analysis, content creation, and affective analysis (among others). Though deep learning algorithms lack the same level of flexibility in thinking that humans possess and, thus, have yet to replicate human thought (Eysenck & Eysenck, 2021), the output generated by AI-algorithms within online environments closely resembles the kind of output that human minds produce.

3.2. AI and Human Cognition

In the 1950s, British mathematician Alan Turing (1950) introduced the ground-breaking ‘Turing Test’ to gauge machines’ ability to simulate human intelligence, sparking significant interest and debate across disciplines like psychology, cognitive science, computer science, and philosophy. Interest surged but later waned by 1970, when Marvin Minsky and Seymour Papert (1969) detailed the challenges in achieving fully automated AI, citing limitations in learning nonlinear patterns and representing complex hierarchical structures. These computational hurdles gradually lessened in their significance in the 1980s and 1990s with advancements like expert systems, which introduced specific semi-automated tasks and data retrieval through programmed code, protocol, and rules. By the late 1990s, enhanced computational power facilitated sophisticated software developments like video games and internet technologies, supplying vast sources of data for real-time computational analysis (Hosanagar, 2019). These developments revived interest in the potential for AI to simulate human thought by the 2010s.

3.3. AI Technologies and Human Thought

Since the 1950s, numerous scholars, including Douglas Hofstadter (1979), Steven Pinker (1997), Ray Kurzweil (1999), Nick Bostrom (2014), and Eliezer Yudkowsky (2015), have contemplated AI’s ability to emulate human neural networks. More recent research examines how AI-mediated interactions on social media influence identity and social relationships (Turkle, 2015, 2017, 2023) while others contemplate the philosophical implications of AI decision-making in economics and governance (Bostrom & Yudkowsky, 2018; McAfee & Brynjolfsson, 2017). However, the concepts of ‘collaborative cognition’ and ‘interaction alignment’ between humans and AI remains relatively unexplored. This paper fills this gap by investigating how human interactions with bots on social media shapes public opinion and ideology, enhancing our understanding of how AI-mediated communication are used to effectively manipulate political information among constituents in the digital age.

4.1. Interaction Alignment & Collaborative Cognition

Interactive alignment theory, pioneered by Garrod and Pickering in the early 2000s, explores how human-to-human interactions incite human interlocutors to generate a sense of mutual affinity and belonging by mimicking one another’s communicative behaviours. We accomplish this kind of intersubjective conviviality in several ways, but we are especially adept at aligning linguistic patterns to help calibrate the meaning of shared experiences and their mental representations. Interaction alignment is thus critical to successful social information processing (2006, 2007, 2009, 2011, 2013, 2014).

Rooted in social psychology and psycholinguistics, the theory builds upon earlier work by H. Clark (1991), who observed that interlocutors gradually synchronize their mental representations through language to achieve mutual understanding and enhance cooperation within social contexts. Numerous scholars have applied Garrod and Pickering’s framework to investigate whether individuals similarly align with computer mediated texts on a messaging platform, finding that digital interactions often amplified the effect of interactive alignment (Shen & Wang, 2022, 2023; Cohn et al., 2020; Cappellini et al., 2022; Placiński & Żywiczyński, 2023).

Interactive alignment theory posits that underlying priming processes, involving the unconscious activation of preceding responses to stimuli, are manifested in linguistic expressions (Placiński & Żywiczyński, 2023). Successful communication, according to this theory, hinges on the use of shared mental representations among conversational partners, facilitating alignment and focusing collective attention toward shared objectives. This cognitive process entails reciprocal priming of mental representations during dialogue, operating automatically and subconsciously (Pickering & Garrod, 2004). This study uses interactive alignment theory to inform the hypothesis that human users who engaged with the #SNCLavalin Twitter discourse of early 2019 would be compelled to interactively align with the political content of tweets posted by bots. Moreover, interactive alignment theory helps elucidate how tweets generated by bots likely induce human users into a state of collaborative cognition with bot-generated content, resulting in alignment with the semantic value of tweets posted by bots.

Collaborative cognition is a social process through which people use their intellectual abilities alongside others toward resolving a problem while relying on one another expertise and knowledge. Collaborative cognition also entails the negotiation of meaning within intersubjective engagement, resulting in shared mental models and conceptual frameworks that enhance consensus building while minimizing misunderstanding (Norman et al., 2021). Interaction alignment and collaborative cognition are, thus, intimately related concepts in that effective alignment of communication and behavior among individuals supports and enhances collaborative cognitive processes (Rasenberg et al., 2020). By synchronizing their interactions, individuals can exchange knowledge, coordinate their actions, and achieve common goals, which has been a fundamental human strength underlying many of our innovative and inventive successes.

The integration of automated psychometric profiling into political electioneering raises several ethical concerns but it also prompts several methodological and theoretical considerations for the Digital Humanities. Research has shown that traces of our digital activity can be utilized by deep learning AI to infer predictions about our beliefs, behaviors, and motivations (Kosinski et al., 2013; Youyou et al., 2015; Kosinski, 2023). Matz et al. (2017) have noted that machine learning and generative AI can be used to potentially exploit vulnerabilities in our character and incite us to take actions against our own best interests (Matz et al., 2017).

This observation alludes to the idea that AI technologies could be designed to influence some fundamental aspect of our psychology in ways that enhance our susceptivity to manipulation online. This study contends that our social psychological strengths – namely, our ability to cognate collaboratively and share intentionality through joint attention – make us vulnerable to the influence of machine learning and generative AI on social media. Our innate propensity to interactively align with one another creates blind spots in our thinking about politics when we are exposed to bot-generated content on social media platforms like Twitter.

To test the hypothesis that humans were likely incited to interactively align with political ideologies propagated by bots on Twitter across the #SNCLavalin Twitter discourse of 2019, this study first used Spearman’s phi-coefficient to assess the degree of correlation between tweets posted by bots and humans. High to very high correlations were identified in three temporal case studies (tweets from March 14, March 28–29, and April 8–9, 2019), supporting the hypothesis that interaction alignment between bots and human users did take place across the #SNCLavalin Twitter discourse of 2019. A linear regression analysis was then conducted to examine the relationship between bot and human-generated content, finding that the bot corpus accounted for a significant degree of the variance in human tweets.

This is a unique approach to analyzing the impact of AI technologies on human thinking in the digital era, but it also presents new considerations for long-standing questions about the relationship between the human brain and its capacity to invent technology that simulates its own neural and cognitive functions.

The text of tweets making up the corpus studied for this paper was collected using the hashtag #SNCLavalin, resulting in a corpus of 4,582 tweets. Using a computational tool called Botometer, which is designed to determine the likelihood that a Twitter account is operated by a bot rather than a human, bot-generated tweets were identified and segmented into a separate sub-corpus consisting of 628 tweets. The sub-corpus of 3,954 human tweets was then further segmented into 3 temporal samples representing tweets posted on March 14, between March 28–29, and between April 8–9, 2019 (this distribution is outlined in Diagram 1). Every word belonging to each tweet in every corpus was then semantically tagged using WMatrix5, which uses statistical classification of words according to 21 major discourse fields and 232 sub-classifications built into the software package (Rayson, 2021). These sub-classifications were then quantified for both bot and human corpora and the phi-coefficient was applied to measure the degree of association between the two categories (e.g., bots and humans) followed by a linear regression analysis to assess whether and to what degree bot-generated tweets influenced the content of tweets posted by humans.

6.1. Data Collection

Using the hashtag #SNCLavalin as a search parameter on Twitter, a corpus of 4,582 tweets was collected and segmented into two subsets: tweets posted by bots (n = 628) and tweets posted by humans (n = 3,954). Human tweets were further divided into three temporal sub-samples: tweets from March 14 (HT1: n = 1,230), tweets between March 28–29 (HT2: n = 1,288), and tweets between April 8–9 (HT3: n = 1,436) in early 2019. Diagram 1 illustrates the distribution and organization of all datasets.

6.2. Organization of Data

WMatrix5 tagged each word in the corpora (BOTS, HT1, HT2, HT3) by parts-of-speech (e.g., nouns, verbs, adjectives) and further classified them according to 21 semantic categories. This research focused specifically on words tagged as belonging to the “government and public” (G) and “emotions” (E) discourse fields, as defined by WMatrix5 (see Archer et al., 2002), since they were the most appropriate to the study. Each word was also assigned a sub-categorical tag from a 232-tag subset. For instance, a word tagged as E3 indicated it belonged to the emotions field, specifically denoting calm, violent, or angry emotions depending on additional qualifiers. Semantic tags like G2.2 signified aspects of general ethics in politics, with a ‘+’ or ‘-’ indicating positive or negative connotations. For example, G2.2- indicated a negative aspect of general political ethics (Archer et al., 2002). Table 1 shows the frequency of such tags across all corpora.

Table 1

Breakdown of semantic tags per corpora for E sample
TAG	BOTS	HT1	HT2	HT3
E1	0	1	24	1
E2+	13	101	56	87
E2-	2	12	12	5
E3+	2	4	9	8
E3-	38	277	106	169
E4.1+	2	16	29	30
E4.1-	10	87	17	11
E4.2+	0	0	7	7
E4.2-	0	2	3	7
E5+	2	0	1	13
E5-	6	59	20	11
E6+	1	10	18	5
E6-	8	21	46	46

Table 2

Breakdown of semantic tags per corpora for the G sample
TAG	BOTS	HT1	HT2	HT3
G1.1	77	230	203	399
G1.1-	3	9	27	5
G1.2	72	221	256	315
G2.1	87	196	379	493
G2.1+	17	205	40	30
G2.1-	9	52	52	305
G2.2	3	5	9	19
G2.2+	6	31	48	54
G2.2-	83	336	257	360
G3	9	42	23	116

6.3. Analysis

A one-way ANOVA was conducted for tagged words belonging to both the ‘emotions’ (E) and ‘government and public’ (G) semantic fields. Results indicated no significant difference in the means among groups in the E sample (F = 1.45, p = .239), but significant differences were found in the G sample (F = 3.02, p = .042). Post hoc tests using Bonferroni adjustments revealed that the BOTS - HT3 pairwise comparison was statistically significant (p < 0.05), suggesting distinct differences between these groups. Levine tests confirmed unequal variances across groups (E sample: p = 0.009; G sample: p < .001). Spearman’s phi Coefficient assessed the strength and direction of association between semantic tags in bot-generated and human-generated tweets. Finally, linear regression analysis evaluated the impact of the BOT corpus on the three human corpora.

This analysis revealed a strong positive correlation between the bot-generated corpus and each of the human-generated corpora, both in the emotions (E) and government and public (G) samples. In the G sample, Spearman’s correlation coefficients indicated very high positive correlations between BOTS and HT1 (r(8) = 0.87, p = .001) on March 14, 2019, BOTS and HT2 (r(8) = 0.87, p = .001) between March 28–29, 2019, and BOTS and HT3 (r(8) = 0.91, p < .001) between April 8–9, 2019. Refer to Table 3 for detailed pairwise correlation analysis results.

Table 3

Standard error of estimates for words tagged as belonging to the ‘Government & Public’ semantic field.
Pairwise Groups G	R	R²	Adj R²	Standard error of the estimate
BOTS - HT1	0.87	0.76	0.73	61.61
BOTS - HT2	0.96	0.92	0.91	39.94
BOTS - HT3	0.87	0.76	0.74	94.31

Linear regression analysis for the G sample showed that the BOTS corpus accounted for 75.6% of the variance in HT1 (F = 24.8, p = .001, R² = 0.756), 91.9% in HT2 (F = 90.38, p < .001, R² = 0.919), and 76.5% in HT3. In contrast, results for the E sample indicated that the BOTS corpus explained 97.01% of the variance in HT1, 84.8% in HT2, and 89.65% in HT3. The E sample demonstrated higher precision, as indicated by standard errors of estimates, suggesting closer alignment to the regression line compared to the G sample. Tables 3 and 4 provide comprehensive details of these regression outcomes. The significance of these results can be examined by revisiting this paper’s central research questions:

Table 4

Standard error of estimates for words tagged as belonging to the ‘Emotional Actions, States, and Processes’ semantic field.
Pairwise Groups E	R	R²	Adj R²	Standard error of the estimate
BOTS - HT1	0.98	0.97	0.97	14.02
BOTS - HT2	0.92	0.85	0.83	11.68
BOTS - HT3	0.95	0.9	0.89	16.08

Q1: Do tweets posted by Twitterbots compel human users to interactively align with that content?

Results outlined above suggest that bots have a significant influence on the emotional tenor of human posts about political topics on Twitter. Their impact is especially strong in discussions that evoke strong negative emotions, where bot and human word usage is most closely aligned. This means that what bots post on Twitter could heavily affect what human users think about important political issues. Given that the identify of bots is not apparent to human users on Twitter, and provided that political brokers are known to use machine learning for psychometric profiling and generative AI for targeted messaging, it is reasonable to conclude that human users are likely compelled to interactively align, and thus cognate collaboratively with bots on Twitter.

Q2. Can interaction alignment theory elucidate the collaborative cognitive processes

between human users and bots in shaping political discourse on social media?

Interactive alignment is an innate social impulse among human interlocutors, which enhances mutual information processing, cooperation, and understanding among individuals. It is a subconscious process that enables us to align our mental representations with those around us as we interact with them, and it is a crucial for shared objectives and successful communication (Garrod & Pickering, 2006, 2007, 2009, 2011, 2013, 2014). Research has shown that interactive alignment is not only evident in human-computer interactions, but digital communication can also even amplify interactive alignment effects among human users (Branigan & Pearson, 2006; Branigan et al., 2010; Branigan, Pickering & Pearson, 2003; Shen & Wang, 2023). Collaborative cognition underpins several facets of interaction alignment; thus it can be concluded that bots likely influence the nature of political discourse on social media by inducing interaction alignment among human users, who consequently cognate collaboratively with bots.

Q3. What are the ethical implications of AI-mediated political influence on social media and its implications for democratic principles, as framed by the Canadian Charter of Rights and Freedoms?

Since findings suggest that human users are compelled to replicate political content posted by bots on Twitter, and provided that research has already supplied compelling evidence of interaction alignment within human-computer interactions (Branigan & Pearson, 2006), it may be concluded that the use of machine learning and generative AI for electioneering purposes does, indeed, interfere with Canadians’ freedom of conscience and judgment, as outlined in the Canadian Charter of Rights and Freedoms (Canadian Charter of Rights and Freedoms, 1982, s 15). In the sphere of contemporary political communication on social media and its implications for constituents’ ability to think independently during democratic elections, understanding interaction alignment between constituents and AI-mediated discourse is critical. Democratic processes are increasingly challenged, giving rise to autocratic tendencies and populism (Bernhard & O’Neill, 2023). Understanding collaborative cognition in AI-mediated communication is essential to informed policy decision that safeguard citizens’ fundamental rights to think about important political issues free from the interference of electioneering agendas.

Positive correlations observed between tweets posted by bots and humans support the hypothesis that Canadian Twitter users likely interactively aligned with AI-mediated bot content across the #SNCLavalin discourse from March 14th to April 9th, 2019, thus suggesting that collaborative cognition within human-computer interactions took place. These findings are consistent with previous studies on interaction alignment in human-computer interactions (Branigan et al., 2003, 2006, 2010, 2011; Cowan et al., 2015; Shen & Wang, 2023; Suzuki & Katagiri, 2007; Koulouri et al., 2015; Nuñez et al., 2023), indicating that humans are compelled to unconsciously mimic linguistic structures, whether they are produced by peers or computer programs (Branigan et al., 2011). This research also aligns with existing theories highlighting imitation as fundamental to human communication, socialization, and cooperation (Bargh & Chartrand, 1999).

This study advances our understanding of AI’s influence on human thought and political ideology online through the lens of interactive alignment theory and collaborative cognition. Previous research has explored alignment in syntax (Cowan et al., 2015), lexical choices (Shen & Wang, 2022; Nuñez et al., 2023), prosody (Suzuki & Katagiri, 2007), vocal alignment with virtual assistants (Cohn et al., 2020), and thematic alignment (Riordan, 2013). However, semantic alignment in online communication has remained relatively underexplored. This study addresses this gap by analyzing semantically tagged words in bot- and human-generated tweets. The findings support the claim that interaction alignment occurred between bots and human users during the #SNCLavalin Twitter discourse in 2019. This alignment was detectable in the semantic construction of the linguistic data in the tweets, demonstrating that semantic alignment is evident in human-computer interactions, as proposed by Branigan et al. (2011).

This research also employed computational corpus analysis and corpus linguistic methods to explore online human behavior, focusing on the formation of political ideologies during elections. While previous studies have examined the affective, stylistic, syntactic, and lexical aspects of political discourse and their spread among constituents (Sukhanov, 2018), few have investigated how this discourse influences the formation of political ideologies on Twitter. This study clarifies how humans are compelled to interactively align with bot-generated content on Twitter and how collaborative cognition between humans and bots in AI-mediated political communication shapes political ideologies through interaction alignment.

This research faced some limitation. The relatively small size of the BOT corpus compared to the three human corpora could have limited the statistical power, potentially leading to random variability and outliers. To address these limitations, Bonferroni post hoc tests and Levene’s tests were conducted. However, studies suggest that even a small presence of bots (5–10% of the discourse) can significantly influence public opinion (Chen et al., 2020; Chang-Feng et al., 2021), affirming the significance of our findings despite the BOT corpus’s small size.

Future research should replicate these findings across various politically contentious topics to verify the consistency of these dynamics and to provide further validation for this paper’s findings. Using the same approach and protocol on different topics could further enhance our understanding of how AI-mediated technologies and the presence of bots on social media influence political discourse. Continued research in this area could also assist policymakers in understanding how the manipulation of human-computer interactions distorts public knowledge on critical social and political issues. This distortion potentially infringes on individuals’ rights to freedom of thought and judgment as outlined in the Canadian Charter of Rights and Freedoms, thereby compromising the principles of free and democratic elections.

Author Contribution

First and Only Author: Sophia Melanson Ricciardone

Acknowledgement

I would like to thank my dissertation supervisors, Dr. Jamin Pelkey and Dr. Dr. Stéphanie Walsh Matthews, as well as my committee members, Dr. Steven C Bailey, Dr. Leslie Atkinson, and Dr. Ruth Breeze.

Data Availability

Data Availability Statement: The data underlying this article are available via Open Science Framework, at https://osf.io/ft6b2/?view_only=bafcf074d5b3447b826b2d8560dea5b7

Ahmed, Y. A., Ahmad, M. N., Ahmad, N., & Zakaria, N. H. (2019). Social media for knowledge-sharing: A systematic literature review. Telematics and Informatics, 37, 72–112. https://doi.org/10.1016/j.tele.2018.01.015
Archer, Dawn, et al. Introduction to the USAS Category System. 2003.
Barberá, P., Jost, J. T., Nagler, J., Tucker, J. A., & Bonneau, R. (2015). Tweeting From Left to Right: Is Online Political Communication More Than an Echo Chamber? Psychological Science, 26(10), 1531–1542. https://doi.org/10.1177/0956797615594620
Bargh, J. A., & Chartrand, T. L. (2014). The Mind in the Middle: A Practical Guide to Priming and Automaticity Research. In C. M. Judd & H. T. Reis (Eds.), Handbook of Research Methods in Social and Personality Psychology (2nd ed., pp. 311–344). Cambridge University Press. https://doi.org/10.1017/CBO9780511996481.017
Bernhard, M., & O’Neill, D. (2023). Looking Backward, Looking Forward. Perspectives on Politics, 21(3), 805–809. https://doi.org/10.1017/S1537592723001925
Blei, D. M. (2012). Topic modeling and digital humanities. Journal of Digital Humanities, 2(1), 8-11.
Bostrom, N. (2014). Superintelligence: Paths, Dangers, Strategies. Oxford University Press.
Bostrom, N., & Yudkowsky, E. (2018). The Ethics of Artificial Intelligence. In Artificial Intelligence Safety and Security. Chapman and Hall/CRC.
Branigan, H. P., Pickering, M. J., Pearson, J., & McLean, J. F. (2010). Linguistic alignment between people and computers. Journal of Pragmatics, 42(9), 2355–2368. https://doi.org/10.1016/j.pragma.2009.12.012
Branigan, H. P., Pickering, M. J., Pearson, J., McLean, J. F., & Brown, A. (2011). The role of beliefs in lexical alignment: Evidence from dialogs with humans and computers. Cognition, 121(1), 41–57. https://doi.org/10.1016/j.cognition.2011.05.011
Branigan, H. P., Pickering, M. J., Pearson, J., McLean, J. F., & Nass, C. (2003, July). Syntactic alignment between computers and people: The role of belief about mental states. In Proceedings of the 25th annual conference of the cognitive science society (Vol. 31, pp. 186-191). Hillsdale, NJ, USA: Lawrence Erlbaum Associates.
Branigan, H., & Pearson, J. (2006). Alignment in human-computer interaction. How people talk to computers, robots, and other artificial communication partners, 140-156.
Breeze, R. (2015). “Or so the government would have you believe”: Uses of “you” in Guardian editorials. Discourse, Context & Media, 10, 36–44. https://doi.org/10.1016/j.dcm.2015.07.003
Breeze, R. (2020a). Angry tweets: A corpus-assisted study of anger in populist political discourse. Journal of Language Aggression and Conflict, 8(1), 118–145. https://doi.org/10.1075/jlac.00033.bre
Breeze, R. (2020b). Exploring populist styles of political discourse in Twitter. World Englishes, 39(4), 550–567. https://doi.org/10.1111/weng.12496
Breeze, R. (2015). Teaching the vocabulary of legal documents: a corpus-driven approach. ESP Today, 3(1), 44-63.
Bryson, J.J. (2015). Artificial Intelligence and Pro-Social Behaviour. In: Misselhorn, C. (eds) Collective Agency and Cooperation in Natural and Artificial Systems. Philosophical Studies Series, vol 122. Springer, Cham. https://doi.org/10.1007/978-3-319-15515-9_15
Canada (Government of), Statistics Canada (2024, February 27). The Daily — Online hate and aggression among young people in Canada. https://www150.statcan.gc.ca/n1/daily-quotidien/240227/dq240227b-eng.htm (accessed June 25, 2024).
Canada. Parliament. Canadian Charter of Rights and Freedoms. 1982. Available at: https://laws-lois.justice.gc.ca/eng/const/page-12.html (accessed June 25, 2024).
Caliskan, A., Bryson, J. J., & Narayanan, A. (2017). Semantics derived automatically from language corpora contain human-like biases. Science, 356(6334), 183–186. https://doi.org/10.1126/science.aal4230
Chang-Feng, C., Shi, W., Yang, J., & Fu, H.-H. (2021). Social bots’ role in climate change discussion on Twitter: Measuring standpoints, topics, and interaction strategies. Advances in Climate Change Research, 12(6), 913–923. https://doi.org/10.1016/j.accre.2021.09.011
Chen, X., Zhou, X., Chan, J., Chen, L., Sellis, T., & Zhang, Y. (2022). Event Popularity Prediction Using Influential Hashtags From Social Media. IEEE Transactions on Knowledge and Data Engineering, 34(10), 4797–4811. https://doi.org/10.1109/TKDE.2020.3048428
Cheng, C., Luo, Y., & Yu, C. (2020). Dynamic mechanism of social bots interfering with public opinion in network. Physica A: Statistical Mechanics and Its Applications, 551, 124163. https://doi.org/10.1016/j.physa.2020.124163
Clark, H. H. & Schober, M. F. (1991) Asking questions and influencing answers. In: Questions about questions: Inquiries into the cognitive bases of surveys, ed. J. M. Tanur, pp. 15–48. Russell Sage Foundation.
Cohn, M., Jonell, P., Kim, T., Beskow, J., & Zellou, G. (2020, July). Embodiment and gender interact in alignment to TTS voices. In CogSci.
Cowan, B. R., Branigan, H. P., Obregón, M., Bugis, E., & Beale, R. (2015). Voice anthropomorphism, interlocutor modelling and alignment effects on syntactic choices in human−computer dialogue. International Journal of Human-Computer Studies, 83, 27–42. https://doi.org/10.1016/j.ijhcs.2015.05.008
Crockett, M. J. (2017). Moral outrage in the digital age. Nature Human Behaviour, 1(11), 769–771. https://doi.org/10.1038/s41562-017-0213-3
Disinformation or Mass Persuasion Dataset* Melanson Ricciardone, Sophia (2024), Open Science Framework, https://osf.io/ft6b2/?view_only=bafcf074d5b3447b826b2d8560dea5b7
Dubois, E., Gruzd, A., Mai, P., & Jacobson, J. (2018). Social Media and Political Engagement in Canada (SSRN Scholarly Paper No. 3299155). https://doi.org/10.2139/ssrn.3299155
Eysenck, M. W., & Eysenck, C. (2021). AI vs Humans. Routledge. https://doi.org/10.4324/9781003162698
Freelon, D., McIlwain, C. D., & Clark, M. (2016). Beyond the Hashtags: #Ferguson, #Blacklivesmatter, and the Online Struggle for Offline Justice (SSRN Scholarly Paper No. 2747066). https://doi.org/10.2139/ssrn.2747066
Gallotti, M., & Huebner, B. (2017). Collective intentionality and socially extended minds. Philosophical Psychology, 30(3), 251–268. https://doi.org/10.1080/09515089.2017.1295629
Garrod, S., & Pickering, M. (2014). Interactive alignment and language use. In T. M. Holtgraves (Ed.), The Oxford Handbook of Language and Social Psychology (pp. 131–140). Oxford University Press.
Garrod, S., & Pickering, M. J. (2009). Joint Action, Interactive Alignment, and Dialog. Topics in Cognitive Science, 1(2), 292–304. https://doi.org/10.1111/j.1756-8765.2009.01020.x
Garrod, S., & Pickering, M. J. (2013). Dialogue: Interactive Alignment and Its Implications for Language Learning and Language Change. In P.-M. Binder & K. Smith (Eds.), The Language Phenomenon: Human Communication from Milliseconds to Millennia (pp. 47–64). Springer. https://doi.org/10.1007/978-3-642-36086-2_3
Hancock, J. T., Woodworth, M. T., & Porter, S. (2013). Hungry like the wolf: A word‐pattern analysis of the language of psychopaths. Legal and Criminological Psychology, 18(1), 102–114. https://doi.org/10.1111/j.2044-8333.2011.02025.x
Hansson, S., Fuoli, M., & Page, R. (2024). Strategies of Blaming on Social Media: An Experimental Study of Linguistic Framing and Retweetability. Communication Research, 51(5), 467–495. https://doi.org/10.1177/00936502231211363
Henry, Didier, et al. “Filter Hashtag Context through an Original Data Cleaning Method.” Procedia Computer Science, vol. 130, Jan. 2018, pp. 464–71, https://doi.org/10.1016/j.procs.2018.04.050.
Heuer, Hendrik. Users & Machine Learning-Based Curation Systems. Staats- und Universitätsbibliothek Bremen, 2020.
Hofstadter, D. R. (1979). Gödel, Escher, Bach: An Eternal Golden Braid. Penguin.
Hosanagar, K. (2020). A Human’s Guide to Machine Intelligence: How Algorithms Are Shaping Our Lives and How We Can Stay in Control. Penguin.
Hughes, S. W. and A. (2019, April 24). Sizing Up Twitter Users. Pew Research Center. https://www.pewresearch.org/internet/2019/04/24/sizing-up-twitter-users/
Kingston, A. (2017, October 12). Inside the “progressive” think tank that really runs Canada. Macleans.Ca. https://macleans.ca/politics/ottawa/inside-the-progressive-think-tank-that-really-runs-canada/
Kosinski, M. (2023). Computational Psychology. https://doi.org/10.31234/osf.io/tkhg4
Kosinski, M., Stillwell, D., & Graepel, T. (2013). Private traits and attributes are predictable from digital records of human behavior. Proceedings of the National Academy of Sciences, 110(15), 5802–5805. https://doi.org/10.1073/pnas.1218772110
Koulouri, T., Lauria, S., & Macredie, R. D. (2016). Do (and Say) as I Say: Linguistic Adaptation in Human–Computer Dialogs. Human–Computer Interaction, 31(1), 59–95. https://doi.org/10.1080/07370024.2014.934180
Kurzweil, R. (2001). Homo sapiens: Leben im 21. Jahrhundert - was bleibt vom Menschen? Econ-Taschenbuch-Verlag.
La Rocca, G., & Boccia Artieri, G. (2022). Research using hashtags: A meta-synthesis. Frontiers in Sociology, 7. https://doi.org/10.3389/fsoc.2022.1081603
Maru Group (2022). Canadian News Consumption. Retrieved May 28, 2024, from https://www.marugroup.net/public-opinion-polls/canada/canadian-news-consumption
Matz, S. C., Kosinski, M., Nave, G., & Stillwell, D. J. (2017). Psychological targeting as an effective approach to digital mass persuasion. Proceedings of the National Academy of Sciences, 114(48), 12714–12719. https://doi.org/10.1073/pnas.1710966114
McAfee, A., & Brynjolfsson, E. (2017). Machine, Platform, Crowd: Harnessing Our Digital Future. W. W. Norton & Company.
McKelvey, F., & Dubois, E. (2017). Computational propaganda in Canada: The use of political bots. https://ora.ox.ac.uk/objects/uuid:cb1b7ea7-41ac-4de2-9c05-b35a04049788
Menenti, L., Garrod, S., & Pickering, M. (2012). Toward a neural basis of interactive alignment in conversation. Frontiers in Human Neuroscience, 6. https://www.frontiersin.org/articles/10.3389/fnhum.2012.00185
Minsky, M., & Papert, S. (1969). Perceptrons; an Introduction to Computational Geometry. MIT Press.
Norman, Utku, et al. Studying Alignment in a Collaborative Learning Activity via Automatic Methods: The Link Between What We Say and Do. arXiv:2104.04429, arXiv, 14 Apr. 2022, https://doi.org/10.48550/arXiv.2104.04429.
Nuñez, T. R., Jakobowsky, C., Prynda, K., Bergmann, K., & Rosenthal-von der Pütten, A. M. (2023). Virtual agents aligning to their users. Lexical alignment in human–agent-interaction and its psychological effects. International Journal of Human-Computer Studies, 178, 103093. https://doi.org/10.1016/j.ijhcs.2023.103093
Nur, Wan & Sakinah, Aida & Wan Jusoh, Wan Nur & Jobar, Norfaizah & Nizam, Zahril & Md Yusoff, Md Zahril Nizam & Mahat, Hanifah. (2024). Exploring corpus linguistics via computational tool analysis: key finding review. Indonesian Journal of Electrical Engineering and Computer Science. 34. 1042-1052. 10.11591/ijeecs.v34.i2.pp1042-1052.
Pickering, M. J., & Garrod, S. (2004). Toward a mechanistic psychology of dialogue. Behavioral and Brain Sciences, 27(2), 169–190. https://doi.org/10.1017/S0140525X04000056
Pickering, M. J., & Garrod, S. (2006). Alignment as the Basis for Successful Communication. Research on Language and Computation, 4(2), 203–228. https://doi.org/10.1007/s11168-006-9004-0
Pickering, M. J., & Garrod, S. (2007). Do people use language production to make predictions during comprehension? Trends in Cognitive Sciences, 11(3), 105–110. https://doi.org/10.1016/j.tics.2006.12.002
Pinker, S. (1997). Words and rules in the human brain. Nature, 387(6633), 547–548. https://doi.org/10.1038/42347
Placiński, M., & Żywiczyński, P. (2023). Modality effect in interactive alignment: Differences between spoken and text-based conversation. Lingua, 293, 103592. https://doi.org/10.1016/j.lingua.2023.103592
R. v. Big M Drug Mart Ltd., [1985] 1 S.C.R. 295.
Rasenberg, Marlou, et al. “Alignment in Multimodal Interaction: An Integrative Framework.” Cognitive Science, vol. 44, no. 11, 2020, p. e12911, https://doi.org/10.1111/cogs.12911.
Rahman, M. M., Khan, N. I., Sarker, I. H., Ahmed, M., & Islam, M. N. (2023). Leveraging machine learning to analyze sentiment from COVID‐19 tweets: A global perspective. Engineering Reports, 5(3), e12572. https://doi.org/10.1002/eng2.12572
Rayson, P. (2021). WMatrix5. https://eprints.lancs.ac.uk/id/eprint/156972/
Riordan, E. (2013). Online reflections: the implementation of blogs in language teacher education. https://researchrepository.ul.ie/articles/chapter/Online_reflections_the_implementation_of_blogs_in_language_teacher_education/19882612/1
Rodríguez-Ibánez, M., Casánez-Ventura, A., Castejón-Mateos, F., & Cuenca-Jiménez, P.-M. (2023). A review on sentiment analysis from social media platforms. Expert Systems with Applications, 223, 119862. https://doi.org/10.1016/j.eswa.2023.119862
Shen, H., & Wang, M. (2022). Improving Interaction Experience through Lexical Convergence: The Prosocial Effect of Lexical Alignment in Human-Human and Human-Computer Interactions. International Journal of Human–Computer Interaction, 38(1), 28–41. https://doi.org/10.1080/10447318.2021.1921367
Shen, H., & Wang, M. (2023). Effects of social skills on lexical alignment in human-human interaction and human-computer interaction. Computers in Human Behavior, 143, 107718. https://doi.org/10.1016/j.chb.2023.107718
Stone, J. A., & Can, S. H. (2020). Linguistic analysis of municipal twitter feeds: Factors influencing frequency and engagement. Government Information Quarterly, 37(4), 101468. https://doi.org/10.1016/j.giq.2020.101468
Stromer-Galley, J., Rossini, P., Hemsley, J., Bolden, S. E., & McKernan, B. (2021). Political Messaging Over Time: A Comparison of US Presidential Candidate Facebook Posts and Tweets in 2016 and 2020. Social Media + Society, 7(4), 205630512110634. https://doi.org/10.1177/20563051211063465
Sukhanov, Y. Y. (2018). Political discourse as object of linguistic analysis. RUDN Journal of Language Studies, Semiotics and Semantics, 9(1), 200–212. https://doi.org/10.22363/2313-2299-2018-9-1-200-212
Suzuki, N., & Katagiri, Y. (2007). Prosodic alignment in human–computer interaction. Connection Science, 19(2), 131–141. https://doi.org/10.1080/09540090701369125
Turing, A. M., & Anderson, A. R. (1950). Minds and machines.
Turkle, S. (2015). Reclaiming Conversation: The Power of Talk in a Digital Age. Penguin Press.
Turkle, S. (2017). Alone Together: Why We Expect More from Technology and Less from Each Other. Basic Books.
Turkle, S. (2023). Always-On/Always-on-You: The Tethered Self. In Social Theory Re-Wired (3rd ed.). Routledge.
Walker, R. M., Zhang, J., Chandra, Y., Dong, B., & Wang, Y. (2023). Revisiting the academic–practitioner divide: Evidence from computational social science and corpus linguistics. Public Administration Review, 83(6), 1599–1617. https://doi.org/10.1111/puar.13724
Wischnewski, M., Ngo, T., Bernemann, R., Jansen, M., & Krämer, N. (2024). “I agree with you, bot!” How users (dis)engage with social bots on Twitter. New Media & Society, 26(3), 1505–1526. https://doi.org/10.1177/14614448211072307
Youyou, W., Kosinski, M., & Stillwell, D. (2015). Computer-based personality judgments are more accurate than those made by humans. Proceedings of the National Academy of Sciences, 112(4), 1036–1040. https://doi.org/10.1073/pnas.1418680112
Yudkowsky, E. (2016). The AI alignment problem: why it is hard, and where to start. Symbolic Systems Distinguished Speaker, 4, 1.
Zaher, Z., Mazid, I., Brost, L. F., & Wallace, A. A. (2023, March 1). Identifying the Determinants of Retweeting Behavior: A Computational Study of Heuristic-Systematic Model and Brand Communication on Twitter. | Kentucky Journal of Communication | EBSCOhost. https://openurl.ebsco.com/contentitem/gcd:164141630?sid=ebsco:plink:crawler&id=ebsco:gcd:164141630

Table 1: Breakdown of semantic tags per corpora for E sample

TAG	BOTS	HT1	HT2	HT3
E1	0	1	24	1
E2+	13	101	56	87
E2-	2	12	12	5
E3+	2	4	9	8
E3-	38	277	106	169
E4.1+	2	16	29	30
E4.1-	10	87	17	11
E4.2+	0	0	7	7
E4.2-	0	2	3	7
E5+	2	0	1	13
E5-	6	59	20	11
E6+	1	10	18	5
E6-	8	21	46	46

Table 2: Breakdown of semantic tags per corpora for the G sample

TAG	BOTS	HT1	HT2	HT3
G1.1	77	230	203	399
G1.1-	3	9	27	5
G1.2	72	221	256	315
G2.1	87	196	379	493
G2.1+	17	205	40	30
G2.1-	9	52	52	305
G2.2	3	5	9	19
G2.2+	6	31	48	54
G2.2-	83	336	257	360
G3	9	42	23	116

Table 3. Standard error of estimates for words tagged as belonging to the ‘Government & Public’ semantic field.

Pairwise Groups G	R	R²	Adj R²	Standard error of the estimate
BOTS - HT1	0.87	0.76	0.73	61.61
BOTS - HT2	0.96	0.92	0.91	39.94
BOTS - HT3	0.87	0.76	0.74	94.31

Table 4. Standard error of estimates for words tagged as belonging to the ‘Emotional Actions, States, and Processes’ semantic field.

Pairwise Groups E	R	R²	Adj R²	Standard error of the estimate
BOTS - HT1	0.98	0.97	0.97	14.02
BOTS - HT2	0.92	0.85	0.83	11.68
BOTS - HT3	0.95	0.9	0.89	16.08

No competing interests reported.

Download PDF

Reviewers agreed at journal
04 Sep, 2024
Reviewers agreed at journal
27 Aug, 2024
Reviewers agreed at journal
06 Aug, 2024
Reviewers invited by journal
06 Aug, 2024
Editor assigned by journal
06 Aug, 2024
Submission checks completed at journal
06 Aug, 2024
First submitted to journal
29 Jul, 2024

You are reading this latest preprint version

Misinformation or Collaborative Cognition? Examining the Role of Bots in Manipulating Political Thought in Twitter Discourse

Status:

Version 1

Abstract

Figures

1. Introduction

2. Related Work

3. Background

3.1. Broadcasting Political News

3.2. AI and Human Cognition

3.3. AI Technologies and Human Thought

4. Theoretical Framework

4.1. Interaction Alignment & Collaborative Cognition

5. Approach to Research

6. Methods

6.1. Data Collection

6.2. Organization of Data

6.3. Analysis

7. Results and Discussion

8. Conclusions

Declarations

Author Contribution

Acknowledgement

Data Availability

References

Tables

Additional Declarations

Status:

Version 1