Humans utilize all provided information regardless of quality due to a specific cognitive bias
the verdict
INSUFFICIENT LEANING
refutedsupported
the weight of evidence
0 sources for · 2 against
The literature on cognitive biases shows that individuals do not use all information indiscriminately; rather, they selectively favor supportive information and demand varying standards of evidence based on prior beliefs.
Collecting large-scale human-annotated datasets via crowdsourcing to train and improve automated models is a prominent human-in-the-loop approach to integrate human and machine intelligence. However, together with their unique intelligence, humans also come with their biases and subjective beliefs, which may influence the quality of the annotated data and negatively impact the effectiveness of the human-in-the-loop systems. One of the most common types of cognitive biases that humans are subject to is the confirmation bias, which is people's tendency to favor information that confirms their existing beliefs and values. In this paper, we present an algorithmic approach to infer the correct answers of tasks by aggregating the annotations from multiple crowd workers, while taking workers' various levels of confirmation bias into consideration. Evaluations on real-world crowd annotations show that the proposed bias-aware label aggregation algorithm outperforms baseline methods in accurately inferring the ground-truth labels of different tasks when crowd workers indeed exhibit some degree of confirmation bias. Through simulations on synthetic data, we further identify the conditions when the proposed algorithm has the largest advantages over baseline methods.
Accounting for Confirmation Bias in Crowdsourced Label Aggregation Meric Altug Gemalmaz , Ming Yin Purdue University fmgemalma, mingying@purdue.edu Abstract Collecting large-scale human-annotated datasets via crowdsourcing to train and improve automated models is a prominent human-in-the-loop approach to integrate human and machine intelligence. How- ever, together with their unique intelligence, hu- mans also come with their biases and subjective be- liefs, which may influence the quality of the anno- tated data and negatively impact the effectiveness of the human-in-the-loop systems.
One of the most common types of cognitive biases that humans are subject to is the confirmation bias, which is peo- ple’s tendency to favor information that confirms their existing beliefs and values. In this paper, we present an algorithmic approach to infer the cor- rect answers of tasks by aggregating the annota- tions from multiple crowd workers, while taking workers’ various levels of confirmation bias into consideration.
The design of crowdsourcing tasks (e.g., what infor- mation is shown to workers in what order) may also have sub- tle impact on workers and trigger their cognitive biases such as the anchoring bias and ambiguity effect [Eickhoff, 2018; Zhuang et al., 2015]. Another common type of cognitive bias that crowd workers are often subject to is their confirmation bias, which refers to people’s tendency of favoring information that confirms their previously existing beliefs and values [Nickerson, 1998].
In so doing, the current crowdsourced label aggregation algorithms might have missed the opportunity to further improve the inference accuracy by explicitly modeling how worker’s cognitive bi- ases have influenced their work quality. Therefore, in this paper, we focus on worker’s confirmation bias and propose a new label aggregation algorithm to ac- count for it. Specifically, we formulate a probabilistic model of the label generation process by assuming that among other factors, worker’s label on a task is influenced by both the val- ues of the worker and the values expressed in the task.
Within a single task, the ways that infor- mation is presented and the order that questions are asked can also result in worker’s cognitive bias which negatively im- pacts the work quality [Eickhoff, 2018]. In addition, workers may exhibit biases in
For example, when considering the left–right political spectrum,ci = 1 (orsj = 1) could mean the values of annotatori (or the values implied by information in taskj) are extremely conservative, while ci = 0 (orsj = 0 ) means the values of annotatori (or the values implied by information in taskj) are extremely liberal. Annotators’ confirmation bias is captured via the distance between ci and sj—holding all other variables equal, the closer ci andsj are to each other, the more likely annotator i will provide the preferable label in taskj (i.e.,P (lij = 0) is larger). We further use the parameterpi2 [0; 1] to characterize the extent to which annotator i is subject to confirmation bias.
Here, pi = 0 means that annotator i is heavily influenced by her confirmation bias, such that she decides her label on tasks (almost) entirely based on how much the information contained in the task aligns with her values. Conversely, when pi = 1 , annotator i is not influenced by her confir- mation bias at all, such that she decides her label on tasks (almost) entirely based on the ground truth label zj of the task, and zj Bernoulli(1) (i.e., the prior probability for a task to have the preferable label as its ground truth is , P (zj = 0) = ).
Con- sidering workers’ annotations on all 12 statements, the aver- age bias scores for liberal, neutral, and conservative work- ers are 0.18, -0.47, and -0.12, respectively, and we find a negative, albeit non-significant, correlation between workers’ stance and their bias scores (Pearson correlation coefficient =0:086;p = 0:374). This means that compared to neu- tral and conservative workers, liberal workers indeed favored information with liberal values slightly more, implying some degree of confirmation bias.
As humans are often subject to various types of biases, the challenge of how to carefully process the crowd- sourced data to minimize the negative impact that people’s biases bring to data quality becomes pressing. In this paper, we focus on confirmation bias, a particular type of cognitive bias, and propose a new label aggregation algorithm based on a quantitative model which characterizes how crowd workers are influenced by their confirmation bias in their annotations. The evaluation results on both real-world data and synthetic data demonstrate the effectiveness of our proposed method.
Confirmation bias (also confirmatory bias, myside bias, or congeniality bias) is the tendency to search for, interpret, favor and recall information in a way that
Confirmation bias (also confirmatory bias, myside bias, or congeniality bias) is the tendency to search for, interpret, favor and recall information in a way that confirms or supports one's prior beliefs, values, or decisions. People display this bias when they select information that supports their views, ignoring contrary information or when they interpret ambiguous evidence as supporting their
Confirmation biases are not limited to the collection of evidence. Even if two individuals have the same information, the way they interpret it can be biased.
A team at Stanford University conducted an experiment involving participants who felt strongly about capital punishment, with half in favor and half against it. Each participant read descriptions of two studies: a comparison of U.S. states with and without the death penalty, and a comparison of murder rates in a state before and after the introduction of the death penalty. After reading a quick description of each study, the participants were asked whether their opinions had changed. Then, they read a more detailed account of each study's procedure and had to rate whether the research was well-conducted and convincing. In fact, the studies were fictional. Half the participants were told that one kind of study supported the deterrent effect and the other undermined it, while for other participants the conclusions were swapped.
The participants, whether supporters or opponents, reported shifting their attitudes slightly in the direction of the first study they read. Once they read the more detailed descriptions of the two studies, they almost all returned to their original belief regardless of the evidence provided, pointing to details that supported their viewpoint and disregarding anything contrary. Participants described studies supporting their pre-exis
Later work reinterpreted these results as a tendency to test ideas in a one-sided way, focusing on one possibility and ignoring alternatives. Explanations for the observed biases include wishful thinking and the limited human capacity to process information. Another proposal is that people show confirmation bias because they are pragmatically assessing the costs of being wrong rather than investigating in a neutral, scientific way. Flawed decisions due to confirmation bias have been found in a wide range of political, organizational, financial and scientific contexts.
In social media, confirmation bias is amplified by the use of filter bubbles and "algorithmic editing", which display to individuals only information they are likely to agree with, while excluding opposing views. == Definition and context == Confirmation bias, previously used as a "catch-all phrase", was refined by English psychologist Peter Wason, as "a preference for information that is consistent with a hypothesis rather than information which opposes it." Confirmation biases are effects in information processing.
Half the participants were told that one kind of study supported the deterrent effect and the other undermined it, while for other participants the conclusions were swapped. The participants, whether supporters or opponents, reported shifting their attitudes slightly in the direction of the first study they read. Once they read the more detailed descriptions of the two studies, they almost all returned to their original belief regardless of the evidence provided, pointing to details that supported their viewpoint and disregarding anything contrary. Participants described studies supporting their pre-existing view as superior to those that contradicted it, in detailed and specific ways.
As participants evaluated contradictory statements by their favored candidate, emotional centers of their brains were aroused. This did not happen with the statements by the other figures. The experimenters inferred that the different responses to the statements were not due to
if the soul is infected with partisanship for a particular opinion or sect, it accepts without a moment's hesitation the information that is agreeable to it. Prejudice and partisanship obscure the critical faculty and preclude critical investigation. The result is that falsehoods are accepted and transmitted. In the Novum Organum, English philosopher and scientist Francis Bacon (1561–1626) noted that biased assessment of evidence drove "all superstitions, whether in astrology, dreams, omens, divine judgments or the like". He wrote: The human understanding when it has once adopted an opinion ... draws all things else to support and agree with it.
According to experiments that manipulate the desirability of the conclusion, people demand a high standard of evidence for unpalatable ideas and a low standard for preferred ideas. In other words, they ask, "Can I believe this?" for some suggestions and, "Must I believe this?" for others. Although consistency is a desirable feature of attitudes, an excessive drive for consistency is another potential source of bias because it may prevent people from neutrally evaluating new, surprising information. Social psychologist Ziva Kunda combines the cognitive and motivational theories, arguing that motivation creates the bias, but cognitive factors determine the size of the effect.
Biased assimilation is a factor in the modern appeal of alternative medicine, whose proponents are swayed by positive anecdotal evidence but treat scientific evidence hyper-critically. Cognitive therapy was developed by Aaron T. Beck in the early 1960s and has become a popular approach. According to Beck, biased information processing is a factor in depression. His approach teaches people to treat evidence impartially, rather than selectively reinforcing negative outlooks. Phobias and hypochondria have also been shown to involve confirmation bias for threatening information.
Everything we examined (2)
This check searched the claim as stated. It did not run a separate search for evidence against it.