Our relationship with artificial intelligence looks suspiciously like a teenage romance. On Monday, we cannot stop talking about it. We ask ChatGPT to plan our holidays, draft our emails, and explain quantum physics “like I’m five.” On Tuesday, it confidently tells us that a famous author wrote a book they never wrote, and we storm off declaring we will never trust it again. By Thursday, we are back, asking it for dinner recipes. It is complicated.
A recent research tries to make sense of this emotional rollercoaster. Their question: why do people swing between trusting AI too much and rejecting it altogether, and what role do our own mental blind spots play?
Two opposite reactions to the same technology
Researchers have given names to these two moods. “Algorithm appreciation” describes the tendency to prefer advice from an algorithm over advice from a person, believing the machine to be more objective or accurate. “Algorithm aversion” is the opposite: avoiding or distrusting an algorithm, often after seeing it make a mistake.
Both have been well documented, which is precisely the puzzle. Some studies find that people readily follow AI and even rate it above human experts. Others find that people lose faith in an algorithm far more quickly than they would in a human who made the same error. In one well-known line of research, people who watched an algorithm make a single mistake went back to their own, less accurate judgements rather than risk the machine’s imperfection. One stumble, and the relationship is over.
Generative AI has made this tension sharper. Tools like ChatGPT and DeepSeek attracted users at record speed, with ChatGPT reaching 100 million users in about two months. At the same time, they have been widely criticised for producing biased or inaccurate content rooted in their training data. So why do so many people keep using tools they also distrust?
Enter willful ignorance and motivated blindness
The authors propose that the answer lies partly in two very human psychological habits, both of which sound like something your grandmother would accuse you of.
The first is “willful ignorance.” This is the art of deliberately not looking at information we would rather not know, often to avoid responsibility or protect a belief we like. It is the person who never checks their bank balance after a shopping spree, or the student who does not open the email with exam results until after the weekend.
The second is “motivated blindness.” This is the tendency to overlook problems, including ethical problems or bias, when noticing them would get in the way of something we want. If an AI tool is saving you three hours a day, you may be remarkably uninterested in hearing about its flaws. As the old saying goes, it is hard to get someone to understand something when their convenience depends on not understanding it.
These ideas are well known in ethics and psychology, but the authors point out that they are rarely named in research about AI bias. Their contribution is to bring these concepts directly into the conversation about how people respond to AI.
The model: what we want from AI, filtered through what we choose to see
The study’s model starts with three qualities that make up what is often called “responsible AI.” The first is accountability, meaning someone takes responsibility for what the system does. The second is fairness, meaning it does not treat people unjustly. The third is transparency, meaning we can understand how it reaches its outputs. In an ideal world, more of these qualities should lead to more appreciation of AI and less aversion.
But people are not ideal processors of information. The authors propose that willful ignorance and motivated blindness change how strongly these qualities matter. For someone with a strong tendency toward motivated blindness, a little fairness or transparency might go a long way in boosting appreciation, because they are already inclined to see the good and skip the bad. Similarly, when an AI falls short on these qualities, people who practise willful ignorance may not become as averse as you would expect, because they simply choose not to dwell on the shortcomings.
In plain terms, two people can look at the same AI tool with the same strengths and weaknesses and come to opposite conclusions, not because the tool is different but because their minds filter it differently.
How the researchers plan to test it
The paper outlines a mixed-methods approach, which means combining conversations with numbers. In the qualitative part, the team is interviewing MBA students from a leading university who use the latest generative AI tools, to understand in their own words how they come to admire or distrust AI. These interviews will be analysed using grounded theory, a method in which patterns are allowed to emerge from what people say rather than being imposed in advance.
In the quantitative part, participants recruited through an online platform will respond to scenarios in which they receive advice from either an algorithm or a human, and decide whether to accept it. A statistical approach that looks at both individual decisions and individual personalities will reveal whether people generally lean toward or away from algorithms, and whether those with stronger blind spots behave differently.
As a work in progress, the paper presents the theory and research design, with results still to come.
Why it matters
For companies building AI, the implications are intriguing. It is tempting to assume that if you make your system more transparent, fair, and accountable, users will respond rationally and reward you. This study suggests that the human mind is not so tidy. Users’ own psychological tendencies can amplify or muffle the impact of responsible design. Understanding those tendencies could help designers communicate more effectively, perhaps by making biases harder to ignore for users who need to see them, or by helping wary users avoid throwing away a useful tool after one error.
For the rest of us, it is a useful mirror. The next time you find yourself blindly trusting an AI answer because it is convenient, or abandoning an AI tool forever because it made one silly mistake, it is worth pausing. Is the problem the algorithm, or the way you are choosing to see it?
Our relationship with AI may always be a little complicated. But understanding our own biases might help us move from a turbulent romance to something more like a healthy, clear-eyed partnership.
Reference
Zhu, Y., Ding, W., & Sarkar, S. K. (2025). Algorithm appreciation or aversion? The role of psychological biases. In ICIS 2025 Proceedings
