Within the information ecosystem, Smart IRS is designed to effectively retrieve information based on users’ specifications, reflected in their search queries. However, the retrieved results are prone to bias, which could amount to skewed or misrepresented facts. My previous study revealed that the algorithm bias in Smart IRS could undermine the system’s credibility due to its potential to produce biased outcomes. This implies that the information being consumed by system users is filtered through certain biases, shaping their perceptions in subtle yet profound ways. Algorithm bias is just one of the several biases in Smart IRS that influence the information at people’s disposal. Stakeholders in the information industry, like library and information science (LIS) professionals, are now advocating for transparency in information retrieval models. By this transparency, users will be aware of how the information they consume is generated, giving them an amount of control in their information-seeking process.

Let’s take a simple illustration from any of the social media platforms. Have you observed that when you watch or engage with a few videos from a particular page on Facebook, subsequently, you see content from that page even without you following the page? Your past actions are now used to shape or predict future information delivery (interaction bias), thus constraining your information consumption pattern. If my academic search queries over time are limited to a geographical region, in the future, the IRS would prioritize content from that region, which will provide a geographically biased perspective on issues.

When an information user queries an IRS, results are ranked hierarchically from top to bottom based on certain conditions like perceived relevance to keywords in the search queries, previous users’ engagement, visibility of source, perceived quality, and other technical considerations. System users are more likely to consider and use the top-ranked results without critically engaging the results to prioritize quality, introducing positional bias. With this type of bias, system users could have a false belief of the precision level (or quality) of top results while undermining possible relevant and quality results that are lower ranked. Consequently, system users could likely miss out on quality content that is lower ranked. This behavior can trigger the use of poor-quality information in addressing real-life issues. This has implications for information professionals in educating system users on critical information literacy skills to evaluate sources beyond their positional value. Perhaps we can also retrain the algorithm by shifting preference away from top-ranked results by occasionally engaging (clicking and downloading) low-ranked results perceived to be of good quality.

In addition, most IRS prioritizes content from sources deemed to be of more authority and reputation and that are more trustworthy. It is, however, subjective how the IRS determines reputable and trustworthy sources to prioritize in response to the user’s query. This could stimulate people to consume information emanating from public figures and organizations considered to be of authority, regardless of scientific validation. This was one major cause of misinformation during the COVID-19 pandemic. Similarly, information resources in globally ranked outlets with better visibility are favored against those not in outlets with lower visibility, regardless of quality. As such, information users are programmed (behavioral alteration) to consume information on the grounds of visibility rather than quality. This could contribute to why one research article could receive way more citations than another of similar quality. The IRS’s tendency to retrieve more popular results will influence the way users consume information, which over time could lead to users’ behavioral bias. This could explain why the perceived relevance of an academic paper and its consideration for engagement are predominantly based on citations and alternative metrics, rather than on quality.

What constitutes quality for an IRS should go beyond the system’s ability to provide so many relevant materials in the vast amount of information resources within a short period of time to the degree of fairness with which the results are generated, eliminating biases to their minimum. While the IRS has been trained to act in a certain way that causes prejudice and unfair representation and/or misrepresentation of facts, information users are expected to play their part in asserting an amount of control over the system. Clearing account history, browser history, and cookies; using private mode; and turning off personalization settings could help information users reduce interaction bias in IRS on their personal devices. Also, engaging in deep search queries containing detailed and specific search phrases in order to specifically define the parameters of search results will give users an amount of control over the search process.

Information users want to be in charge of the information they consume. In achieving this, they must understand the internal workings of IRS models, seek balanced perspectives on issues, and critically engage search results. This has implications for revamping their critical information literacy skills, enabling them to evaluate how information is produced before they utilize such information.