Information bias (psychology)

For other uses, see Information bias (disambiguation).

Information bias is a cognitive bias to seek information when it does not affect action. An example of information bias is believing that the more information that can be acquired to make a decision, the better, even if that extra information is irrelevant for the decision.

Information Bias in Artificial Intelligence

Information bias in news, media, and online data is fed into Artificial Intelligence (AI) databases. AI then regurgitates the biased information to consumers. Researchers at the University of South Carolina (USC) tested AI databases (ConceptNET and GenericsKB) for bias facts. The results showed that 38.6% of data fed to AI was biased. Since the data AI consumes can have bias due to limited fact checking, this then causes the consumers of AI to then also consume bias data, spreading information bias.

In addition, studies from AIPRM show that people are aware of this information bias as 66% of U.S. adults are highly concerned about inaccurate AI information, as well as 34% of marketers reporting that generative AI produces biased information.

Information bias in AI can lead to harmful, enforced stereotypes. Researchers at the Massachusetts Institute of Technology (MIT) developed a technique for reducing information bias in AI by removing data points that specifically lead to the failure of a model, in order to protect the integrity of the model while removing its bias.

Example

In an experiment (Baron, Beattie & Hershey 1988), subjects considered this diagnostic problem involving fictitious diseases:

A female patient is presenting symptoms and a history which both suggest a diagnosis of globoma, with about 80% probability. If it isn't globoma, it's either popitis or flapemia. Each disease has its own treatment which is ineffective against the other two diseases. A test called the ET scan would certainly yield a positive result if the patient had popitis, and a negative result if she has flapemia. If the patient has globoma, a positive and negative result are equally likely. If the ET scan was the only test you could do, should you do it? Why or why not?

Many subjects answered that they would conduct the ET scan even if it were costly, and even if it were the only test that could be done. However, the test in question does not affect the course of action as to what treatment should be done. Because the probability of globoma is so high with a probability of 80%, the patient would be treated for globoma no matter what the test says. Globoma is the most probable disease before or after the ET scan.

In this example, we can calculate the value of the ET scan by considering 100 patients, of which 80 have globoma. Since it is equally likely for a patient with globoma to have a positive or negative ET scan result, 40 people will have a positive ET scan and 40 people will have a negative ET scan. The remaining 20 patients have either popitis or flapemia. Out of those, the 10 patients with popitis will have a positive ET scan while the 10 patients with flapemia will have a negative scan. Thus, out of the 50 patients with a positive result (40 globoma + 10 popitis), 80% have globoma; likewise, out of the 50 patients with a negative test (40 globoma + 10 flapemia), 80% have globoma. The probability of globoma is therefore entirely unaffected by the result of the test, regardless of how it turns out. The test can provide no information that would affect the decision to treat the globoma, so it should not be carried out. While this example illustrates information bias in a clinical setting, similar patterns can also be observed in everyday contexts such as media consumption.

Information Bias in Media and News Consumption

Information bias can also appear in the news and media, as individuals continue to seek additional information and sources even when it does not change their understanding of the subject. This is extremely common online, where information is easily accessible. Research from the Pew Research Center has illustrated that many U.S. adults perceive news as biased toward one side, which can lead people to seek more information to gain a more balanced view on the subject. Even so, this will not improve understanding as many sources rely on similar perspectives and framing techniques. Bias in news often stem from editorial choices, such as which topics are covered and how that information is presented, rather than distorting the information in an obvious manner. This can reinforce information bias because people may feel that information will improve their understanding, even when it doesn’t change their conclusions.

Consequences of Information Bias

Information bias can lead to several negative outcomes by encouraging people to seek more information than is necessary or useful. One common effect is analysis paralysis, where decision-making slows or stops because individuals feel they must gather additional details before acting. Online environments often intensify this into digital analysis paralysis, since endless search results and algorithmic recommendations create the sense that there is always more to review. Excessive or irrelevant information also increases cognitive load, making it harder to identify what is actually meaningful and sometimes reducing overall decision quality. Information bias can reinforce overconfidence, giving individuals a false sense of accuracy simply because they have collected more data even when that data does not improve outcomes.

In professional settings such as medicine, hiring, or financial planning, this may lead to unnecessary testing, inefficient processes, or misguided judgments. Online spaces further amplify these effects by contributing to information overload and enabling confirmation-driven information seeking, where users selectively engage with content that supports their existing beliefs. Over time, these patterns can slow workflows, entrench misjudgments, and create systemic inefficiencies that could have been avoided with more targeted information gathering.

The Bias Pipeline

The way information bias is created is through what is called the "bias pipeline". The bias pipeline starts with the human mind, cognitive shortcuts that are taken when processing information can shape one's perception. This leads into skewed reporting which effects the news media in where journalists and editors pick and choose what they want to write. As a result, flawed training data is created which leads to AI learning from biased information sources.

Types of Information Bias

There are 4 main types of Information Bias we see frequently; Misclassification, Observer, Recall, and Reporting Bias. Misclassification bias occurs when individuals are assigned to a different category than the one they should be in which can lead to incorrect associations being observed and recorded which can lead to a skewed result in studies. Observer bias is a type of bias that affects assessment when it comes to observational studies, this can be due to observer variation in the way they viewed a value which can be individually different depending on the observer. Recall bias is an error that occurs when participates either do not remember or do not properly remember the previous events/experiences accurately or purposely omit details which will inevitably invalidate the study and its results. The last type of information bias is reporting bias, reporting bias occurs when the information presented is distorted due to selective disclosure or withholding of information from the parties involved.

See also

• Shared information bias
• Cognitive psychology
• List of cognitive biases

Studies

• Baron, Jonathan; Beattie, Jane; Hershey, John C (1988). "Heuristics and biases in diagnostic reasoning" (PDF). Organizational Behavior and Human Decision Processes. 42 (1): 88–110. doi:10.1016/0749-5978(88)90021-0. Archived from the original on 2016-03-05.{{cite journal}}: CS1 maint: bot: original URL status unknown (link)