Some recent research have pointed out an interesting phenomenon: there is a "bidirectional" relationship between the biases in artificial intelligence models and those in human thinking.
This interaction creates a mechanism that tends to amplify cognitive distortions in both directions..
This research shows that AI systems not only inherit human biases from training data, but when implemented can intensify them, in turn influencing people's decision-making processes. This creates a cycle that, if not properly managed, is likely to progressively increase the initial biases.
This phenomenon is particularly evident in important areas such as:
In these areas, small initial biases can amplify through repeated interactions between human operators and automated systems, gradually turning into significant differences in results.
The human mind naturally uses "thinking shortcuts" that can introduce systematic errors into our judgments. The theory of "double thinking" distinguishes between:
For example, in the medical field, physicians tend to give too much weight to initial hypotheses, neglecting contrary evidence. This phenomenon, called "confirmation bias," is replicated and amplified by AI systems trained on historical diagnostic data.
Machine learning models perpetuate biases mainly through three channels:
One 2024 UCL study showed that facial recognition systems trained on emotional judgments made by people inherited a 4.7 percent tendency to label faces as "sad," and then amplified this tendency to 11.3 percent in subsequent interactions with users.
Data analysis of recruitment platforms shows that each round of human-algorithm collaboration increases gender bias by 8-14% through mutually reinforcing feedback mechanisms.
When HR professionals receive lists of candidates from AI that are already influenced by historical biases, their subsequent interactions (such as their choice of interview questions or performance evaluations) reinforce distorted representations of the model.
A 2025 meta-analysis of 47 studies found that three rounds of human-IA collaboration increased demographic disparities by 1.7-2.3 times in areas such as health care, lending and education.
The framework for measuring bias proposed by Dong et al. (2024) allows for the detection of bias without the need for "absolute truth" labels by analyzing discrepancies in decision-making patterns among protected groups.
The "algorithmic mirror" technique developed by UCL researchers reduced gender bias in promotion decisions by 41 percent by showing managers what their historical choices would look like if they were made by an AI system.
Training protocols that alternate between AI assistance and autonomous decision making prove particularly promising, reducing the effects of bias transfer from 17 percent to 6 percent in clinical diagnostic studies.
Organizations that implement AI systems without considering interactions with human biases face amplified legal and operational risks.
Analysis of employment discrimination lawsuits shows that AI-assisted hiring processes increase plaintiffs' success rates by 28 percent compared to traditional human-driven cases, as traces of algorithmic decisions provide clearer evidence of disparate impact.
The correlation between algorithmic bias and restrictions on freedom of choice requires us to rethink technological development from the perspective of individual responsibility and safeguarding market efficiency. It is crucial to ensure that AI becomes a tool for expanding opportunities, not restricting them.
Promising directions include:
Only through responsible self-regulation of the industry, combined with freedom of choice for users, can we ensure that technological innovation continues to be an engine of prosperity and opportunity for all who are willing to put their skills to work.
.svg)
.svg)
.svg)
.jpeg)
.jpeg)
.jpeg)