INTRODUCTION
Artificial intelligence (AI) is currently a top-tier mainstream topic, and it’s easy to see that more and more people are using AI-generated content, such as ChatGPT. Whether it’s researching a historical event, a social issue, or a simple everyday question, it always provides a well-structured and clearly expressed answer. In my observation, I’ve found that this mode of expression inspires an intuitive sense of trust; it makes people feel that it’s an objective tool for presenting information, much like the internet helps us better understand the world. However, in my deeper observation, I’ve discovered that the seemingly objective and independent nature of AI-generated content actually masks more complex algorithmic mechanisms. As Pasquale points out, many contemporary algorithmic systems are essentially “black boxes,” meaning, “we can observe its inputs and outputs, but we cannot tell how one becomes the other” (Pasquale, 2015, pp. 3). In my view, when we cannot understand how information is organized and generated by AI, we will question the content at the source of its generation and its objective and independent expression.
YALE CASE – When AI Starts Changing Minds
https://news.yale.edu/2026/03/03/ais-hidden-bias-chatbots-can-influence-opinions-without-trying?utm_source=chatgpt.com
This trust in objectivity, independence, and neutrality has begun to be challenged in current research, with many questions arising about whether the content produced by artificial intelligence is unbiased. A 2026 Yale University study found that when users chat with human agents like ChatGPT, their views on socio-political issues subtly change. While AI systems themselves are not designed as persuasive tools, they can still influence understanding through different information organization and expression methods, as well as databases. I think there is often a stark contrast in our understanding of information technology. We think that technology merely transmits information, rather than shapes cognition, but we haven’t considered the biases that arise from the sources of data and the databases on which artificial intelligence generates content. I think what’s more important is that this impact is hidden and difficult for users to discover or perceive. As Bolsover and Howard pointed out in their research, AI systems and computer algorithms can not only disseminate information but also use databases to alter the structure of public discourse and influence users’ understanding of reality. For example, they stated that automated content is “effective at spreading information and deceiving users” (Bolsover & Howard, 2018, pp. 2064). In my view, this influence is even more difficult to detect if users cannot distinguish the deep patterns and mechanisms of information sources. I’ve discovered that software like ChatGPT isn’t a simple question-and-answer tool, but rather a system of “interpretive tools” based on a certain stance and algorithm, generated through a certain content-derived logic. This interpretation itself influences the user, potentially carrying some inherent bias.
U.S LAW CASE – When AI Gets It Wrong
In the previous Yale case we analyzed, it’s easy to see that this risk has already been exposed in reality. According to The Guardian, in a 2023 U.S. legal case, two lawyers fabricated legal documents using AI to generate fake ones, citing multiple non-substantive pieces of content generated by ChatGPT. Ultimately, the court discovered that these cases were fictitious and penalized the lawyers. In my opinion, this is not just because the AI made a mistake, but more because these mistakes were presented in an authoritative and credible form, which even professionals failed to detect in time. At the same time, I also believe this incident reveals a core problem with generative AI: so-called “hallucination” AI systems generate seemingly plausible content even without real-world context and data. This AI-generated content possesses a strong sense of confidence and logical self-consistency, making it difficult for users to distinguish the authenticity and reliability of the data source unless manual verification is performed. I think that in this case, in such a high-risk field as law, such credible misinformation is very dangerous. It will directly affect the judgment process and decision-making of the entire case. In fact, the lawyers involved in this case explained that they did not find that the fabricated statements in these cases were most likely not intentional, but rather that they had excessive trust in AI contents. Looking at this case study from a broader perspective, I found it strongly resonates with Yale’s approach. In both cases, AI not only influences users’ opinions but also alters judgments in key decision-making situations as people increasingly rely on AI for information retrieval and content analysis. I believe this influence extends from the cognitive level to reality. Therefore, this case is not merely an example of technical error but also a warning about how humans can be misled by algorithms.
ALGORITHM INFLUENCE – How Algorithms Shape Understanding
If these two cases demonstrate how AI influences real-world judgments, then the bigger question is how this influence occurs. From a technical perspective, generative AIs like ChatGPT often don’t fully adhere to the premise of finding truthful answers. Instead, they respond with content that appeases or unconditionally obeys users. This is often reflected in the false content provided by ChatGPT. I believe the core logic is based on probabilistic prediction rather than verification of actual facts. This means that when generative AI outputs content, it prioritizes sorting and filtering based on existing training data. This process doesn’t prioritize authenticity but rather focuses on unconditionally obeying user commands to satisfy their emotional needs. As Bolsover and Howard pointed out, in automated information environments, algorithms can “dominate discourse” (Bolsover & Howard, 2018, pp. 2067). I think algorithms not only send information but also determine which information should be seen first and how it should be understood. I think that algorithms do more than just send information; they also determine which information should be seen first and how it should be understood. This mechanism becomes even more complex when applied to generative AI because the system not only selects information but also actively reconstructs the intent and expression of the information content. At the same time, I think this mechanism is very similar to “engagement optimization” in platform design. I found that generative AI tends to produce more fluid, easier-to-understand content that better aligns with user emotions and expectations. AI platforms use this model to retain users’ time on the AI platform and improve their experience with AI (anthropomorphic mode). However, this model does not equate to good objectivity and realism; instead, it reinforces certain trends, making users more emotionally receptive to the explanations of generative AI. From an objective perspective, this kind of long-term, subtle adjustment that caters to user emotions will gradually shape the user’s understanding framework, leading to an unconsciously biased perception among users.
THE AI BIAS – How does it work?
The impact of such AI algorithms is not random, but rather determined by the bias inherent in the data itself. Noble’s research points out that algorithmic systems do not objectively reflect real problems, but are more likely to replicate and amplify existing inequalities and biases in society. In her analysis of search engines, she found that when users input specific keywords, the results easily lead to stereotypes rather than providing relatively neutral and objective information. For example, Noble found that search engines tend to associate marginalized groups with negative content when processing race-related keywords. In my view, this further illustrates the bias inherent in AI algorithms and their generation, while simultaneously reshaping people’s perceptions. This bias is mistakenly considered justified.
Noble emphasizes that users’ trust in AI-generated content and search engines is crucial for understanding and exploring the internet, and that these systems “are trusted more readily than they ought to be” (Noble, 2018, pp. 10). In my opinion, when users perceive AI-generated text as objective fact, they have already accepted and entered into a filtered and reconstructed cognitive framework. In reality, there’s a very clear example. Noble’s research found that when users enter keywords like “black girls” into search engines (Noble, 2018, pp. 3-4), the search engines often present content related to pornography and stereotypes, rather than objective and normal information resources. I believe this phenomenon clearly demonstrates that algorithms don’t just reflect user needs; they also use existing data and models to reinforce specific social trends to solidify social consensus, while simultaneously including negative stereotypes and ignoring the objective information needed by vulnerable groups.
A similar case occurred in 2026 when someone used the chat gpt-like AI’s dialogue function, posing as a man/woman and asking the AI, “I didn’t like that man/woman, and it felt so good to punch him.” The AI gave different responses. For men, the AI advised against physical violence to reduce violent tendencies and avoid imprisonment; for women, the AI praised them, commending their strength and independence. Both of these outcome-oriented approaches are highly biased: one views men as natural perpetrators of violence, while the other illogically praises women’s behavior. Under such algorithmic mechanisms, I believe bias is not only preserved but also continuously amplified and reproduced.
At the same time, as AI continues to develop in the future, its impact will deepen further. AI is not only rejecting information and generating new expressions, but also presenting biases and repackaging them to spread their own themes. Therefore, when users rely on AI to understand real-world issues, the content they come into contact with has been invisibly infused with certain values and cognitive paths. These paths are not necessarily objective and neutral information, but rather products that are more in line with the common discourse and context of contemporary society.
CRITICAL
When we combine the three case study questions above, we arrive at a deeper conclusion: AI systems are becoming a new form of cognitive power. They directly influence users through soft means, shaping their judgments subtly and through the way information is organized. I believe the unique aspect of this power is that it is not explicit, but deeply embedded within seemingly neutral technological systems.
As Pasquale points out, “to scrutinize others while avoiding scrutiny oneself is one of the most important forms of power” (Pasquale, 2015, pp. 4). I think that in AI-generated systems, this implicit power manifests in the platform’s ability to continuously collect and analyze user behavior without disclosing its internal logic. This information asymmetry puts users in a very passive position during use, as they cannot verify the authenticity of the information source, nor can they understand the way the information is produced (technically).
In my view, this structural right has already had a profound impact on the present reality. As demonstrated by the three legal cases mentioned above, this impact is not limited to the cognitive level, but has also deeply affected the professional decision-making field and the political field. This raises a crucial question: when AI systems begin to influence multiple key areas such as law, healthcare, and public policy, do we need to implement stricter regulatory measures? If algorithms have become part of the cognitive infrastructure, then their transparency and accountability mechanisms are no longer merely technical issues, but rather significant social and political concerns.
CONCLUSION – So, Can We Still Trust AI?
Let’s return to the original question: Is ChatGPT Really Neutral — or Quietly Shaping Our Beliefs? I believe we can answer this question now. The answer is definitely no. AI-generated content not only conveys information but also shapes our understanding. AI-generated content not only reflects reality but also reconstructs it. As more and more users rely on these AI systems, we are no longer dealing with a technological tool, but rather with the infrastructure that contributes to building our cognition. In this context, what truly needs to be reconsidered is not whether AI is truly efficient, but whether we can maintain an independent, calm, and critical understanding of the content and world presented to us by AI. Otherwise, I believe that in the future, when algorithms gradually become the main gateway for humans to understand society, we may have unknowingly and actively handed over all judgment and thinking power to artificial intelligence. Therefore, from my perspective, discussions about AI should not only remain at the technical level but also, at a time, need to extend to governance and accountability issues. For example, who should be held responsible for the erroneous information generated by AI? Should the platform bear legal or moral responsibility for its output? In the absence of transparency, how can users effectively monitor these systems? These issues all demonstrate that AI is no longer just a tool, but a social system that needs to be included in public discussion and policy regulation.
THE REFERENCE LIST
60 Minutes. (2023). ChatGPT and large language model bias | 60 Minutes. In YouTube. https://www.youtube.com/watch?v=kloNp7AAz0U
Acres, T. (2023, November 30). ChatGPT turns one: The first year of the chatbot that changed the world. Sky News. https://news.sky.com/story/chatgpt-turns-one-the-first-year-of-the-chatbot-that-changed-the-world-13014185
Bolsover, G., & Howard, P. (2018). Chinese computational propaganda: automation, algorithms and the manipulation of information about Chinese politics on Twitter and Weibo. Information, Communication & Society, 22(14), 2063–2080. https://doi.org/10.1080/1369118x.2018.1476576
Cummings, M. (2026, March 3). AI’s hidden bias: Chatbots can influence opinions without trying. Yale News. https://news.yale.edu/2026/03/03/ais-hidden-bias-chatbots-can-influence-opinions-without-trying?
IDG TECHtalk. (2023). What is ChatGPT? In YouTube. https://www.youtube.com/watch?v=XFmMT58zYVY
Luscombe, B. (2018). Your Brain Works Against You When You Argue With Your Significant Other. Here’s How to Fix That, According to an Expert. In Time. https://time.com/5469804/couples-researcher-spouse-signficant-other-fight-argument/
Milmo, D. (2023, June 23). Two US lawyers fined for submitting fake court citations from ChatGPT. The Guardian; The Guardian. https://www.theguardian.com/technology/2023/jun/23/two-us-lawyers-fined-submitting-fake-court-citations-chatgpt?
Noble, S. U. (2018). Algorithms of Oppression: How Search Engines Reinforce Racism (pp. 15–63). New York University Press.
Pasquale, F. (2015). Chapter Title: INTRODUCTION THE NEED TO KNOW Book Title: The Black Box Society Book Subtitle: The Secret Algorithms That Control Money and Information.
Welcome To Zscaler Directory Authentication. (2025). Edcellent.com. https://www.edcellent.com/blog/why-choose-ib
Be the first to comment