Identifying and Mitigating Online Bias in Information Search

Part 2 in a series of research perspectives in human-computer interaction.

Since the 1950s, research in psychology has achieved major insights into human cognition and behavior by adopting a view of human learning as an information-processing system of limited capacity [1]. Within these bounds, heuristics provide efficient solutions to cognitively demanding tasks without the impossible analysis of all available information. In realistic settings that involve time constraints, complexity, and incomplete information, judgments made by heuristics may be necessary or even rational [2, 3, 4]. Internet users face large amounts of uncertainty as they navigate and process information on the web, where they may be particularly vulnerable to false information, misinformation, restrictions of content or time, and ignorance. Where uncertainty exists independently of the user’s own knowledge, the dependence on heuristics in such an environment can lead to biased behavior that strays far from making objective judgments or well-informed decisions.

Drawn from a survey of more than 10,000 U.S. adults in 2018, the Pew Research Center reports that 81% of Americans rely on their own research to gather information before making major life decisions — much more than those who rely on their friends and family (46%) or on professional experts (31%) [5]. A majority of those who rely on their own research, either “a lot” or “a little”, turn to digital tools including the internet, Google, forums or comments, online reviews, and social media. I will explore the last three applications in greater detail in the following post; here, I focus on search engines and search systems. Online search tools and resources certainly afford users a diversity of information and perspectives, but it does not guarantee the diversity of a user’s own knowledge. Content exposure and credibility assessment during information consumption are, in large part, under the user’s selective control over search terms and page navigation, and a biased user may depart considerably from veridicality as they seek content and judge credibility.

In a study on online credibility evaluation, Metzger, Flanagin, and Medders (2010) showed that participants employ selective filters on information, avoiding information that is contrary to existing beliefs and ending their search upon finding information that agrees with those beliefs [6]. This finding is consistent with other studies on selective exposure that show that Internet users tend to favor attitudinally consistent information over inconsistent information [7, 8, 9] and exhibit confirmation bias as they search for information on topics ranging from health [10, 11, 12, 13] to politics [14, 15]. In greater context, selectivity in information consumption based on preexisting beliefs and implicit biases have led users to online “echo chambers” of isolated in-group ideologies or viewpoints [16, 17, 18]. The persistence of echo chambers has led to increased polarization on issues and has diminished plurality of thought, which is integral to information flow and formative debates or discussions.

Seminal work in psychology and behavioral economics have explained biased assessments in terms of cognitive heuristics with functional value, made in the face of uncertainty [19]. However, the heuristics explained in the classic work have been critiqued for their vagueness and their inability to produce testable constraints on the cognitive decision-making processes that are claimed to be only loosely labeled [20, 21, 22]. Further theories have been qualified by understanding the decision-maker as a part of an environment of limited information, in which uncertainty emerges not only from the bounded rationality of an individual’s own mental structure but also from ecologically-valid cues and choice criteria found in the environment. This interpretation suggests that the complexity of environments affects an individual’s capacity to acknowledge their uncertainty, and that one’s own cognition and experience determine the situations they become uncertain about [4]. The specification that “the rationality of heuristics is not logical, but ecological” and that its definition is relative to the environment [23] allows bias to be better tasked and understood through its interaction with controllable environmental parameters. Since online environments are mediated through technology that can be directly programmed, simulated, or observed; the interpretation that uncertainty is a property of the relationship between users and online systems promotes better description of the mechanisms through which it causes bias and to what effect. This qualifies the understanding of bias, beyond a mere recognition of its existence.

Accordingly, users’ beliefs are not only influenced by their initial biases, but also by the design and results of search engines. Jonas, Schulz-Hardt, Frey, and Thelen (2001) showed that confirmation bias resulting from selective exposure of information is stronger due to sequential presentation as opposed to sequential processing of information. This expansion on dissonance theory highlights the importance of the order that users see and select information. The implication in this study is that a user, at each step, makes a decision to confirm, and thereby strengthen, their belief or consider disconfirming it [24]. Bias in search activity can decrease accuracy and increase skewness in the resulting content, and strategies to mitigate further reinforcement of existing user bias depend on the design and structural elements of the search system [25].

In order to understand how belief dynamics and confidence are impacted by the search engine and the content of its results, White and Horvitz (2015) assessed people’s behavior in online search and retrieval in the domain of health information. Their findings showed that search activity is affected by interactions between presearch, or pre-existing, beliefs and the nature or positioning of information. Strongly held beliefs may be difficult to update; but the authors suggest that the determination of clear search paths, led by strong beliefs, can be useful in capturing user viewpoints for the personalization of search experience or in detecting erroneous beliefs for the transmission of corrective information. Information disagreeing with prior beliefs resulted in longer dwell times, and strong evidence placed near the beginning of pages were more likely to result in belief revision following review, suggesting that presearch beliefs and content position and salience collectively shape how people examine those pages as well as the potential for updating their beliefs. In addition to confirmation bias, searchers also exhibited a bias towards positive, or “helps-related”, health information as opposed to information that “does not help” [26]. For tasks that require information from trustworthy authorities like medical decision-making, this selection bias for effective information can provide designers a goal for supporting ways to transmit factual, accurate information rather than validating beliefs or catering to personalized preferences. The topic of personalization warrants further discussion, which I cover in [4/6] of this series.

Information search is particularly vulnerable to confirmation bias and selection bias, due to the inherent gaps in user knowledge as well as the cognitive heuristics employed to deal with the sheer complexity and uncertainty of the internet search space. These biases that influence search can lead internet users to reinforce misperceptions or make ill-informed decisions, especially when biases are exacerbated further by external online actors or motives. Strategies to mitigate such forces on user biases require more selective combative strategies than only search engine adjustments, such as machine learning tools that detect fake news [27]. I enumerate these algorithmic strategies further in [4/6].

Extensions and applications of information foraging theory can guide developers to construct information structures that foster a healthy information diet, which requires an understanding of a user’s mental model of the information environment as well as inquisition into what search engines should explain at all [28]. Unharnessed algorithms are capable of propagating dubious claims or skewed perceptions, and bias mitigation measures require more than applying another algorithmic tool. The biases explored here involve the individual user in information search; however, biases in online activity also extend much further in social contexts and in interactions with recommender systems. Not only is information shaped by bias in its coverage or distribution through social media and in the sentiments expressed from the content, but it is also prey to “gatekeeping” through selective exposure, or the preference for specific outlets, despite the expected diversity and density of connections in social networking platforms [29]. How bias narrows these avenues and exerts blockades on communication is explored in the next part of this series on social media.

References

1. Proctor R.W., & Vu K-P. L. (2012). Human Information Processing. In: Seel N.M. (eds) Encyclopedia of the Sciences of Learning. Springer, Boston, MA. https://doi.org/10.1007/978-1-4419-1428-6_722
2. Canellas, M. C., & Feigh, K. M. (2014). “Heuristic decision making with incomplete information: Conditions for ecological rationality”. 2014 IEEE International Conference on Systems, Man, and Cybernetics (SMC), San Diego, CA, pp. 1963–1970. https://doi.org/10.1109/SMC.2014.6974209
3. Gigerenzer, G. (2002). Adaptive thinking : Rationality in the real world. ProQuest Ebook Central https://ebookcentral.proquest.com
4. Kozyreva, A., & Hertwig, R. (2019). The interpretation of uncertainty in ecological rationality. Synthese. https://doi.org/10.1007/s11229-019-02140-w
5. Turner, E., & Rainie, L. (2020, March 5). Most Americans rely on their own research to make big decisions, and that often means online searches. Pew Research Center. https://www.pewresearch.org/fact-tank/2020/03/05/most-americans-rely-on-their-own-research-to-make-big-decisions-and-that-often-means-online-searches/
6. Metzger, M. J., Flanagin, A. J., Medders, R. B. (2010). Social and Heuristic Approaches to Credibility Evaluation Online, Journal of Communication, 60(3), 413–439. https://doi.org/10.1111/j.1460-2466.2010.01488.x
7. Fischer, P., Jonas, E., Frey, D., & Schulz-Hardt, S. (2005). Selective exposure to information: The impact of information limits. European Journal of Social Psychology, 35, 469–492. https://doi.org/10.1002/ejsp.264
8. Johnson, T., Zhang, W., & Bichard, S. (2011). Revived and Refreshed: Selective Exposure to Blogs and Political Web Sites for Political Information. Networked Sociability and Individualism: Technology for Personal and Professional Relationships. 196–217. http://doi.org/10.4018/978-1-61350-338-6.ch010
9. Schmidt, A. L., Zollo, F., Del Vicario, M., Bessi, A., Scala, A., Caldarelli, G., Stanley, H. E., & Quattrociocchi, W. (2017). Anatomy of news consumption on Facebook. Proceedings of the National Academy of Sciences of the United States of America, 114(12), 3035–3039. https://doi.org/10.1073/pnas.1617052114
10. Meppelink, C. S., Smit, E. G., Fransen M. L., & Diviani, N. (2019). “I was Right about Vaccination”: Confirmation Bias and Health Literacy in Online Health Information Seeking. Journal of Health Communication, 24:2, 129–140. https://doi.org/10.1080/10810730.2019.1583701
11. Moreno-Fernández, M. M., & Matute, H. (2020). Biased Sampling and Causal Estimation of Health-Related Information: Laboratory-Based Experimental Research. J Med Internet Res, 22(7). https://doi.org/10.2196/17502
12. Schweiger, S., Oeberst, A., & Cress, U. (2014). Confirmation bias in web-based search: a randomized online study on the effects of expert information and social tags on information search and evaluation. Journal of medical Internet research, 16(3), e94. https://doi.org/10.2196/jmir.3044
13. Schweiger, S., & Cress, U. (2019). How Confidence in Prior Attitudes, Social Tag Popularity, and Source Credibility Shape Confirmation Bias Toward Antidepressants and Psychotherapy in a Representative German Sample: Randomized Controlled Web-Based Study. Journal of medical Internet research, 21(4), e11081. https://doi.org/10.2196/11081
14. Knobloch-Westerwick, S., Johnson, B. K., & Westerwick, A. (2015). Confirmation Bias in Online Searches: Impacts of Selective Exposure Before an Election on Political Attitude Strength and Shifts, Journal of Computer-Mediated Communication. 20(2), 171–187. https://doi.org/10.1111/jcc4.12105
15. Westerwick, A., Johnson, B. K., & Knobloch-Westerwick, S. (2017). Confirmation biases in selective exposure to political online information: Source bias vs. content bias, Communication Monographs, 84:3, 343–364. https://doi.org/10.1080/03637751.2016.1272761
16. Boutyline, A., & Willer, R. (2017). The Social Structure of Political Echo Chambers: Variation in Ideological Homophily in Online Networks. Political Psychology, 38: 551–569. https://doi.org/10.1111/pops.12337
17. Iyengar, S., & Hahn, K. S. (2009). Red Media, Blue Media: Evidence of Ideological Selectivity in Media Use. Journal of Communication, Volume 59, Issue 1, March 2009, Pages 19–39, https://doi.org/10.1111/j.1460-2466.2008.01402.x
18. Williams, H. T. P., McMurray, J. R., Kurz, T., Lambert, F. H. (2015). Network analysis reveals open forums and echo chambers in social media discussions of climate change. Global Environmental Change, 32: 126–138. https://doi.org/10.1016/j.gloenvcha.2015.03.006
19. Tversky, A., & Kahneman, D. (1974). Judgment under Uncertainty: Heuristics and Biases. Science (New York, N.Y.), 185(4157), 1124–1131. https://doi.org/10.1126/science.185.4157.1124
20. Fiedler, K., & Sydow, M. (2015). Heuristics and Biases: Beyond Tversky and Kahneman’s (1974) Judgment under Uncertainty. https://www.researchgate.net/publication/280981235_Heuristics_and_Biases_Beyond_Tversky_and_Kahneman's_1974_Judgment_under_Uncertainty
21. Gigerenzer, G. (1991). How to Make Cognitive Illusions Disappear: Beyond “Heuristics and Biases”. European Review of Social Psychology, 2:1, 83–115. https://doi.org/10.1080/14792779143000033
22. Gigerenzer, G. (1996). On narrow norms and vague heuristics: A reply to Kahneman and Tversky. Psychological Review, 103(3), 592–596. https://doi.org/10.1037/0033-295X.103.3.592
23. Gigerenzer, G. (2008). Rationality for mortals: How people cope with uncertainty. Oxford University Press. https://doi.org/10.1016/j.jebo.2008.12.007
24. Jonas, E., Schulz-Hardt, S., Frey, D., & Thelen, N. (2001). Confirmation bias in sequential information search after preliminary decisions: An expansion of dissonance theoretical research on selective exposure to information. Journal of Personality and Social Psychology, 80(4), 557–571. https://doi.org/10.1037/0022-3514.80.4.557
25. White, R. W., & Hassan, A. (2014). Content Bias in Online Health Search. ACM Trans. Web 8, 4, Article 25, 33 pages. https://doi.org/10.1145/2663355
26. White, R. W., & Horvitz, E. (2015). Belief Dynamics and Biases in Web Search. ACM Trans. Inf. Syst. 33, 4, Article 18 (May 2015), 46 pages. https://doi.org/10.1145/2746229
27. Meel, P., Vishwakarma, D. K. (2020). Fake news, rumor, information pollution in social media and web: A contemporary survey of state-of-the-arts, challenges and opportunities. Expert Systems with Applications, 153. https://doi.org/10.1016/j.eswa.2019.112986
28. Thomas, P., Billerbeck, B., Craswell, N., & White, R. W. (2020). Investigating Searchers’ Mental Models to Inform Search Explanations. ACM Trans. Inf. Syst. 38(1), 10. https://doi.org/10.1145/3371390
29. Saez-Trumper, D., Castillo, C., & Lalmas, M. (2013). Social media news communities: Gatekeeping, coverage, and statement bias. In Proceedings of the ACM International Conference on Information and Knowledge Management (San Francisco, CA, Oct. 27-Nov. 1). ACM Press, New York, 16791684. http://www.dcs.gla.ac.uk/~mounia/Papers/cikmSM_app.pdf