Bringing Artificial Intelligence (AI) into Health Information Seeking Behavior: A Study of AI and Information Seeking Research

Bringing Artificial Intelligence (AI) into Health Information Seeking Behavior: A Study of AI and Information Seeking Research Brady D. Lund, Nishith Reddy Mannuru, Malavika Katta, Sesha Sai Leela Madhuri Hota, Akshaya Pamukuntla, Sravya Uppala, Sai Madhav Kola, and Aashrith Mannuru Abstract This paper explores the impact of artificial intelligence (AI) on information seeking behavior research and practice, including the need to scrutinize existing information seeking theory, challenge the understood behavioral norms, and consider redefining information literacy and information retrieval education. The historical examination spans from the 1950s to the present, with a specific focus on recent developments in health information seeking and the evaluation of medical information sources. Key to this exploration are ongoing debates in healthcare, ethics, and AI and information literacy education, which represent important dimensions of the impact of emerging technology on information-seeking behavior. The insights provided by this research can be useful for both researchers and practitioners, aiding them in navigating the evolving landscape shaped by AI technology. Introduction Artificial Intelligence (AI) has significantly transformed various aspects of life, influencing how individuals search for, evaluate the credibility of, and utilize information. In contemporary society, AI serves as the foundation for numerous commonplace technologies, ranging from virtual assistants to personalized medical recommendations. As AI tools and platforms evolve to become more conversational and sophisticated, seamlessly integrating into daily activities, it becomes crucial to assess and comprehend their impact on the way people engage with health data and knowledge. This area of study is commonly referred to as "human information behavior" (Wilson, 2000). This holds particular importance in the critical field of health, where misinformation and insufficient eHealth literacy are already obstacles to achieving optimal medical outcomes (Paige et al., 2017). AI plays a pivotal role in driving various innovations that grant the public direct access to health insights. For instance, chatbots can conduct initial symptom checks (Arellano Carmona et al., 2022), algorithms can sift through databases to deliver personalized therapy or diet recommendations (Parekh et al., 2023), and social media platforms leverage AI to promote public health campaigns and reliable information sources to users (Song et al., 2021). However, concerns persist regarding the potential for spreading misinformation and compromising user privacy. The current generation of AI chatbots still grapples with providing accurate and safe medical advice. This underscores the necessity for additional guidelines and oversight to ensure the accurate and safe usage of advanced AI chatbots, such as ChatGPT, for medical purposes as these technologies continue to advance (Roumeliotis & Tselikas, 2023). A growing number of individuals are relying on conversational chatbots and Internet-based tools to self-diagnose, locate specialized health resources, and determine whether professional care is necessary (Denecke et al., 2019; Stellefson et al., 2018). These AI systems function by processing user inputs, accessing extensive datasets of medical knowledge, and delivering customized suggestions or answers to queries. However, a key question remains regarding the impact of these human-AI healthcare interactions on fundamental aspects of human information behavior. This encompasses understanding users' needs, their strategies for information seeking, their ability to effectively utilize responses, and ultimately, the health decisions and outcomes that ensue. Despite the widespread adoption of these AI systems, individuals often struggle to correctly interpret AI-generated medical advice, underscoring the necessity for clear explanation systems. Ensuring the clinical safety and transparency of generative AI models is crucial, especially as these technologies become more prevalent in medicine and healthcare (Zhang & Kamel Boulos, 2023). The focus should be on ensuring the reliability and interpretability of the models themselves rather than relying solely on individuals' ability to interpret their outputs (Ryan, 2020). Analyzing human information behavior is crucial for optimizing the advantages of AI while prioritizing ethical considerations and individual choices. When AI is well-designed, it holds the potential to aid individuals in making informed health decisions independently. Furthermore, it can serve as inspiration for the development of new policies aimed at improving public awareness of AI and establishing oversight mechanisms to uphold ethical AI practices. This study explores the historical, current, and future influences of artificial intelligence on health and medical information-seeking behavior. It aims to uncover the implications these influences may have on individuals and their decision-making processes. Implications of AI for Information Seeking Theories Historically, information seeking involved the active evaluation of resources – assigning value based on relevance and quality of information. This is what Wilson (1981) called “success” or “failure,” Kuhlthau (1991) called “exploration” and “formulation,” and Bates (1989) captured in her “Berrypicking” concept. This process of evaluating resources, and the knowledge and skill acquired from it, are replaced by the judgment of an AI model. The entire search process is truncated. Information literacy, the ability to judge quality and find information to satisfy a need, is supplanted. AI literacy is more important for retrieving information. The ability to critically assess new knowledge supplied by an AI model is vital. This manifestation of information retrieval technology is fundamentally unique from that of a search engine. Search engines provide a variety of information sources – though the ordering of those resources in the search results may be due to a biased algorithm. This is not the case with AI tools, where the information the seeker receives is curated by the tool based on its own relevance criteria and no differing perspectives on a topic are offered. Typically, the best for which one can hope is an acknowledgement that the AI model is not perfect. Furthermore, consider the role of information exchange or transfer among individuals, which occurs with an artificially intelligent interface rather than other humans. For example, consider Bates’ (1989) Berrypicking model, which suggests that individuals scour for information, carefully selecting just the information from a broad range of resources that best satisfies their narrow needs. When interacting with an AI model, AI is doing the searching for you such that there is little opportunity for “berrypicking” information. Information must, necessarily, be collected in a more linear manner. What impact does this innovation have on the individuals? It conflicts with the natural behavior that humans engage in to successfully obtain information. It places the success or failure squarely on the capacity of the AI model to understand the information needs and effectively and accurately collect the information needed to address that need. Even when working with intermediaries in the past – such as certain databases or librarians – information seekers had considerable latitude in evaluating a variety of information sources. Another potential issue arises when the user seeks information that the AI model may be trained to avoid providing. Some models have been trained to avoid providing misinformation or information pertaining to conspiracy theories or harmful topics. While this may be wellintentioned, there are legitimate reasons why these topics may be explored by individuals, so censoring or filtering responses through the lens of “this topic is inappropriate” can be problematic and could fundamentally alter the success or failure of information seeking in a similar way to book bans in libraries (Marsoof et al., 2023). Effective engagement with an AI large language model for information retrieval hinges on the quality of communication with the system. Figure 1 depicts a fundamental, linear process of seeking information, drawing inspiration from Shannon's (1948) communication model. This diagram serves as a representation of the information-seeking experience when interacting with an AI large language model. It is crucial to recognize that, in this scenario, the AI model functions as a black box, concealing from users the specifics of where and how the information was obtained. Figure 1. Linear Information Seeking Process Historical Overview of AI in Information Seeking The transformation instigated by AI in information seeking is underscored by the shift from active human evaluation to reliance on AI models. The roots of artificial intelligence in information seeking can be traced back to the 1950s, when researchers began exploring the use of computers for processing and evaluating data. In the subsequent decade of the 1960s, the development of natural language processing (NLP) techniques enabled technology to comprehend and respond to human language. As early as 1968, Robert S. Taylor, in his seminal article titled "QuestionNegotiation and Information Seeking in Libraries," envisioned artificial intelligence as a potential future information retrieval technology that could disrupt the negotiation process in seeking information, even though it was not feasible at that time. This foresight is later acknowledged in AI-based solutions proposed to support information retrieval during the 1970s (Oddy, 1977). The 1970s and 1980s marked a monumental shift in the realm of artificial intelligence (AI) in information seeking research. Information retrieval systems have undergone significant evolution, employing AI approaches, including machine learning and natural language processing (NLP), to enhance both the accuracy and value of search results. In an early paper, Smith (1976) defined key concepts in the role of artificial intelligence for information retrieval systems, laying the foundation for subsequent developments. Javelin and Repo (1982) emphasized the crucial role of understanding information needs and seeking behavior in advancing artificial intelligence. Later, Gruppen (1990) identified a specific niche for artificial intelligence systems – summarizing information from diverse sources related to common information needs. Jones (1991) anticipated a paradigm shift in information retrieval due to AI, acknowledging limitations at the time that no longer persist in the 2020s. Saracevic et al. (1988) suggested that the study of questioning, a common theme in information seeking behavior research, would become increasingly important in the development of artificial intelligence technology. This highlights the growing significance of questioning as a fundamental aspect in the evolution of AI-driven information retrieval. Since the advent of the worldwide web in the 1990s, the landscape of information seeking has undergone significant transformation. Search engines, many of which are powered by AI algorithms, played a crucial role in categorizing and extracting information from the vast expanse of the internet, albeit contributing to the proliferation of misinformation (Jacobs et al., 2017). In response to these changes, new information seeking models and interface design concepts were developed to support the evolving information retrieval processes (Burnett & McKinley, 1998; Marchionini & Komlodi, 1998). Rose (2006) observed a notable shift in information seeking behavior with the rise of web interfaces, particularly those supported by AI-powered search engines. During this period, intelligent agents and specialized recommender systems, precursors to modern AI advancements like large language models, were designed to enhance web-browsing for information seeking and retrieval (Detlor & Arsenault, 2002; Konstan et al., 2006). These advancements significantly reshaped health information seeking behavior, with a growing number of individuals turning to sources other than qualified medical professionals to gather information about health conditions. This trend led to the emergence of challenges like "information overload," where the abundance of health information online made it challenging to discern and identify reliable sources (Revere et al., 2007). In the realm of medical health records and information retrieval, Kannampallil et al. (2013) discovered that an AI-enhanced records system yielded greater information gain compared to a traditional paper record system. As the twenty-first century unfolded, machine learning algorithms based on deep learning became pivotal in enhancing the capabilities of information retrieval systems. Artificial intelligence (AI)- powered recommendation engines have brought about a revolution in how users access content, providing tailored recommendations based on their preferences and actions. This aspect has become a central focus of contemporary information behavior studies, specifically examining how AI influences information retrieval through improved recommendations. Kiester and Turp (2022) explore this impact in relation to PubMed's Best Match algorithm, while Carmona et al. (2022) highlight the transformative effect of AI-driven symptom checkers on health information seeking among patients. Additionally, recent studies have delved into the ways interaction with chatbots may shape individuals' information-seeking behavior. Koman et al. (2020) investigated how physicians perceived the use of a chatbot for health information seeking, revealing that reliability and trustworthiness were major criteria influencing the adoption of the technology in practice. Park et al. (2023) pointed out that information seeking involving AI conversational agents could potentially lead to harmful health outcomes, especially for adolescents and other individuals susceptible to misleading information. On the contrary, Nadarzynski et al. (2019) found that the acceptance of AI chatbots for health information seeking is relatively high, particularly among those with moderate-to-advanced IT skills. Forecasting How AI Will Impact Health Information Seeking As we advance into the era of the fourth industrial revolution, the impact of artificial intelligence on information seeking becomes more evident. The studies discussed below contribute to our comprehension of how information seeking is evolving in response to emerging technologies. These insights underscore the growing emphasis on individual differences in behavior and emphasize the essential need for the responsible design and development of literacy programs that integrate both AI literacy and traditional information literacy skills. Lueg (2002) observes that the problem-solving approaches of artificial intelligence (AI) technologies closely resemble how humans engage in information seeking. Specifically, Lueg describes the AI problem-solving process, where it starts by encountering a "problem" (such as a user's prompt in the case of ChatGPT). Subsequently, the AI transforms this problem into a "plan," a method to locate relevant information (such as referencing training data to understand relationships among concepts). Finally, this plan guides the creation of "knowledge," representing an idea of how to solve the problem (as demonstrated in the solution ChatGPT provides back to the user). Similarly, humans typically approach information seeking by identifying a "gap" or problem, using Brenda Dervin's terminology (1998). They then formulate a plan for seeking the necessary information and engage in the information-seeking process, attempting to retrieve the required information to fulfill their needs—whether successfully or unsuccessfully. Considering the parallel problem-solving processes of AI and the human brain provides insights for information researchers on the role of AI technology in the informationseeking process and society as a whole. It's worth noting that discussions on these topics have been ongoing for several decades and continue to evolve, particularly as conversational AI becomes more widespread among the general public. Investigating the impact of an individual's health consciousness on online health informationseeking behavior, Ahadzadeh et al. (2018) unveil a crucial relationship between attitudes, perceptions, and engagement in seeking health information online. Their findings emphasize that heightened health consciousness strengthens the link between a positive attitude towards online health information seeking and the actual intention to engage in such behavior. Particularly noteworthy is the observation that women prioritizing health exhibit more favorable attitudes towards utilizing the internet for health-related purposes, contributing to increased rates of online health information search behavior. This highlights the significance of acknowledging individual difference factors, such as health consciousness, when examining how attitudes and the influence of artificial intelligence translate into tangible outcomes, including information-seeking intentions and behavior. Arellano Carmona et al. (2022) investigate the usage of Buoy Health, an AI-powered symptom checker and health information tool, in a survey involving 2437 users. The study explores user motivations and perceptions of Buoy's recommendations, revealing that AI-driven personalized health information and care recommendations contribute to reduced anxiety and encourage appropriate help-seeking behaviors. The research also uncovers variations in users' intentions to follow Buoy Health's recommendations, with a greater inclination towards adhering to self-care suggestions compared to in-person care recommendations. Notably, minority groups express higher intentions to discuss AI-generated recommendations with healthcare providers, highlighting the complex dynamics of diverse populations interacting with AI tools. This study significantly contributes to ongoing discussions on AI in healthcare, particularly in the context of health information-seeking systems. While symptom checkers demonstrate the potential for AI to enhance health information seeking, concerns about biases and insufficient evidence on behavioral impacts prompt ethical considerations regarding equitable access and effectiveness. In exploring the application of AI-powered chatbots in supporting adolescents' informationseeking on youth mental and sexual health, researchers have highlighted the potential value of these conversational agents in delivering confidential and non-judgmental resources on sensitive topics (Park et al., 2023). Despite this potential, challenges exist, such as rule-based chatbots lacking tailored guidance and large language models (LLMs) potentially producing offensive or dangerous content. The deployment of AI technologies, especially for vulnerable populations, necessitates careful consideration and accountability. Caution in responsible use, particularly in sensitive areas like youth mental and sexual health, is essential. Urgent research is needed to design AI-powered conversational agents that safely address the information needs of youth, particularly on topics like mental health and sexuality. Companies developing such AI technologies for public use must prioritize fairness and safety, aiming to benefit individuals of diverse backgrounds without exposing them to potential risks or inappropriate content. Enhanced design and research are crucial to ensuring that AI, intended to aid health decisions, does not inadvertently create barriers or unintended negative consequences (Singh et al., 2024). Soboleva and Mills (2023) discovered that a general understanding of how AI tools operate is a significant predictor of increased acceptance of these tools as sources of medical information. Other studies also indicate that the acceptance and adoption of AI innovations are heavily influenced by individuals' backgrounds, resulting in diverse information behaviors among those who do and do not adopt AI for information seeking (Canziani & MacSween, 2021). In their research, Hernandez et al. (2023) examine various factors linking the usage of ChatGPT, an AI chatbot, to information-seeking intentions and behavior. Convenience emerges as a substantial factor moderating ChatGPT usage, while ethical considerations, perceived ease of use, perceived usefulness, social influence, herding, and trust play comparatively less significant roles. The study identifies a strong relationship between the use of ChatGPT and engagement in AI for information seeking. Consistent with prior studies, Hernandez et al. (2023) suggest that individuals more engaged with emerging AI technology may exhibit distinct patterns of information seeking. Hironen et al. (2023) present a novel framework to examine the role of emerging artificial intelligence technologies in the information ecosystem. In everyday information seeking, artificial intelligence is poised to reshape human behavior, with its influence extending to search engines, social media, and nearly every aspect of the Internet. While this transformation may broaden users' access to diverse resources due to the extensive training data behind AI models, it also introduces limitations, restricting access to outputs that may be biased or more confined compared to traditional skimming of all available resources on a topic. As this technology grows, new challenges arise in information seeking, emphasizing the necessity for AI literacy training that extends beyond, yet incorporates elements of, traditional information literacy training. Individuals must grasp that the information available on a topic extends beyond what an AI model provides and discern when it is essential to seek additional sources. Given the critical nature of health information in shaping decisions, validation against alternative sources becomes imperative before acting on information supplied by an AI model. Discussion The introduction of ChatGPT marked a significant turning point, signifying the widespread acceptance of artificial intelligence technology among the general public. Similar to the adoption of search engines and the Internet in the past, AI technology is here to stay, becoming increasingly integrated into daily life and playing a pivotal role in how people seek health information (Walker et al., 2023). Acknowledging this reality is crucial for grasping how to effectively support users and patients as they navigate the information provided by AI models. Reliable health information is of paramount importance, as it can determine outcomes ranging from life and death to overall well-being or illness. Unlike traditional search engines and web indices that present a variety of information resources for users to choose from, AI tools like ChatGPT offer a single response based on what the model deems most relevant to the user's query (Ray, 2023). This shift necessitates a transition from information literacy skills, such as assessing the quality of various information sources, to AI literacy skills. This involves the ability to appropriately prompt an AI model and assess the quality of its response. The conventional process of evaluating resources and acquiring knowledge undergoes a transformation when utilizing AI models like ChatGPT, which act as judgment mechanisms, streamlining the entire search process. This stands in contrast to search engines, underscoring that AI tools curate information based on their own relevance criteria, often lacking diverse perspectives on a given topic. Notably, the linear nature of AI-driven information collection challenges traditional models like Bates' Berrypicking, which observe a more nonlinear "berrypicking" behavior in human information seekers. The success or failure of information seeking now hinges on the AI model's ability to understand and accurately gather information, thereby limiting user autonomy in evaluating diverse information sources. Critically assessing AI outputs, especially when seeking vital health-related information, becomes imperative. Information professionals will likely find themselves entrusted with the responsibility of creating instructional and educational resources to cultivate AI literacy skills. This involves merging expertise in AI tools with conventional information literacy skills, including critical evaluation and the collection of information resources (Lund, 2023). Achieving a seamless integration of these skills necessitates a comprehension of the changing patterns in information seeking and the influence of AI in engaging with information sources. Recognizing the implications of these factors on consumer health information seeking becomes particularly crucial for these professionals, given its potential impact on decisions that can be a matter of life or death. As information and record services companies, including prominent databases and indices, adopt AI technology, a fundamental transformation in user behavior is bound to occur. In anticipation of these changes, it is instructive to draw parallels with historical shifts, such as the advent of the Internet. While the integration of AI will undoubtedly bring about significant alterations, it's important to note that transformative shifts are not entirely new. For instance, electronic health systems revolutionized the storage and retrieval of health information, providing a precedent for the kind of comprehensive changes we can anticipate now (Häyrinen et al., 2008). Even as we brace for substantial changes, the fundamental approach to examining and understanding these shifts should, to a large extent, remain consistent.