Cognitive Science | Dylan Chiang

Paper reading: Cognitive processes of ingroup favoritism across 20 countries

Sat, 22 Nov 2025 00:00:00 +0000

#paper information

Title: Cognitive processes of ingroup favoritism across 20 countries: An eye-tracking investigation of culture, behavior, and cognition Author: Rima-Maria Rahal & Frederik Schulze Spüntrup Journal: PNAS (Proceedings of the National Academy of Sciences) Year: 2025 (Vol. 122, No. 32) DOI:

Research background

In-group favoritism is a common social psychological phenomenon - we tend to favor “one of our own” even if the definition of “one of our own” may only be based on a trivial criterion (such as a different favorite painter).

This preference has serious consequences in the era of globalization:

COVID-19 Vaccine Distribution: Western countries hoard vaccines, developing countries can only wait
Climate Crisis: Countries shirking responsibility and prioritizing their own short-term interests
Xenophobia on the rise: Right-wing populism spreads around the world

In the past, most studies could only be conducted in laboratories, with samples restricted to specific areas. This study breaks through this limitation and uses webcam-based eye-tracking, allowing researchers to collect high-quality cognitive data in participants’ homes, achieving unprecedented cultural and geographical diversity.

Research methods

1. Innovative research design

Sample:

1,850 participants
20 countries (covering Europe, Asia, America, etc.)
Eye tracking using participants’ own webcams

Experimental Task:

Group allocation: based on color recognition task (minimum group paradigm)
Group Reinforcement: Compete with out-group members for bonuses
Decompose the Dictator Game: 80 decisions, each time choosing between the “self-interest” and “pro-social” options
- Self-interest option: you get the most and the other side gets very little
- Prosocial option: take less for yourself, and both the other party and the total will increase

2. Measurement indicators

Behavioral level:

Differences in generosity toward in-group vs. out-group members

Cognitive Level (Eye Tracking):

Decision time
fixation count
Number of messages checked
Allocation of attention to self vs. other’s gains

Cultural level:

Individualism vs. Collectivism
Government effectiveness
Religious importance
Historical disease burden & COVID-19 impact

Main findings

1. In-group preference is universal, but the degree varies widely

Core findings:

✅ All 20 countries show in-group preference (OR = 4.57)
❗ But effect sizes vary widely across countries
📊 The stronger the prosocial tendency of people, the more serious the discrimination against out-groups.

**What does this mean? **“Good people” do not necessarily do not discriminate, they may just leave kindness to “their own people”.

The study tested the Material Security Hypothesis: the more unstable a society is, the more people rely on their in-groups.

Evidence Support:

✅ The lower the government effectiveness** → the stronger the in-group preference**
✅ COVID-19 death toll higher → in-group preference stronger
✅ The lower the historical disease burden → the lower the in-group preference

Unexpected Discovery:

❌ Religion more important → In-group preference stronger (contrary to prediction!)
- Null hypothesis: Religion as a stable institution should reduce preferences
- Practical: Religion may reinforce in-group and out-group boundaries

3. Is an individualistic culture more discriminatory?

Cultural Dimensions Theory predicts: Collectivist cultures should place greater emphasis on ingroups.

Actual result: Completely opposite!

The higher the individualism** → the stronger the in-group preference**

Possible explanation: -In individualistic cultures, people are more eager for group identification

Lacking a fixed social structure, it is more necessary to create boundaries for “one’s own people”

4. Cross-cultural differences in cognitive processes are more complex than behavior

Overall Trend (shown by meta-analysis):

Decision-making becomes more laborious when facing in-group members:
- Longer decision-making time
- More views
- Check more information

But:

⚠️ The direction of this effect varies greatly between countries
In some countries: in-group decision-making is more laborious
In other countries: out-group decisions are more laborious (reversed effect!)

The role of prosocial tendencies also varies across cultures:

In some countries: more prosocial → more effortful in-group decision-making
In other countries: the effect is reversed

5. “Color blindness”: More than half of people choose not to read group information

Amazing discovery:

In 53.49% of decisions, participants did not look at the group information at all (even if it was free)
54.79% of people clearly stated that they did not want to know the other party’s group identity

**Who chooses to remain “color blind”? **

People with strong self-interest (low SVO)
People who have no favorable impressions of either group
People who think others will not view group information

Consequences of color blindness:

People who choose color blindness: No discrimination
People who choose to view: 10% less for external groups

**What does this mean? ** The average discrimination effect may be driven by the few people who “want to know” the group identity!

Research significance and inspiration

Methodological Revolution

Breakthrough in web camera eye tracking:

💻 Participants can complete it at home, no professional laboratory required
🌍 Achieve unprecedented geographical diversity
💰 Lower cost and higher efficiency
📊 Sufficient data quality (spatial and temporal resolution)

This democratizes cognitive research—it’s no longer just university labs in wealthy areas.

Theoretical contribution

Challenges the cultural dimension theory
- Individualism vs. collectivism effects are opposite to predicted
- A single cultural dimension is not enough to explain complex social cognition
Supports the material security hypothesis (partially)
- Social uncertainty does increase discrimination
- But the role of religion is more complex
Reveal the cultural specificity of cognitive processes
- Cognitive mechanisms cannot be assumed to be consistent across cultures
- Even if the behavioral patterns are similar, the cognitive processes behind them may be completely different

Practical inspiration

A warning for global policymaking:

❌ “One size fits all” anti-discrimination policies may be counterproductive in some cultures
✅ Requires culturally appropriate interventions
🎯 “Color-blind” strategies may be effective: de-emphasize group differences instead of emphasizing equal treatment

Implications for reducing discrimination:

🔍 Understand who chooses to view group information (and why)
🎓 Educational focus: reduce focus on group boundaries
🌐 In a society with high uncertainty, stronger intervention is needed

My understanding

After reading this paper, my biggest shock was: it turns out that our understanding of “discrimination” may be based on a very narrow sample. If you studied ingroup preference only in a sample of Western college students, you might come to one conclusion about the cognitive mechanisms. But this study tells us that the same behavior (discrimination against outgroups) may come from completely different cognitive processes in different cultures.

1. Are “good people” more discriminatory? This is counter-intuitive

I have always thought that people with strong prosocial tendencies (people who are willing to sacrifice for others) should be less likely to discriminate. But this study says: On the contrary, these “good people” are more likely to reserve kindness only for “their own people.”

This reminds me of some phenomena in reality. Many people who claim to be “patriotic” may be generous to their fellow countrymen, but are extremely repulsive to foreigners or immigrants. Or some religious groups are very united and supportive within themselves, but are indifferent or even hostile towards “infidels”. This kind of “parochial altruism” may be more common and more persistent than we think.

The explanation proposed in the paper is that prosocial people care more about “fairness” and “justice.” However, when they think that the out-group is “not one of their own,” this sense of fairness does not apply. This made me reflect deeply: It is not enough to simply promote “kindness” or “altruism”. What is more important is to expand the scope of “us” and allow more people to be included in the circle of “ours”.

2. Are individualistic cultures more discriminatory? This overturned my understanding

I have always felt that collectivistic cultures (such as Asia) should emphasize the distinction between inside and outside more, while individualistic cultures (such as Europe and the United States) should treat everyone more equally. But this study found the exact opposite!

The explanation proposed by the author is very interesting: in individualistic cultures, people are more eager for group identification. Because there are no mandatory family or community ties, people need to find their own “sense of belonging,” which makes them more sensitive to “who is one of their own” and “who is an outsider.”

This reminds me of racial issues in the United States, immigrant rejection in Europe, etc. These highly individualistic societies are particularly prone to group conflicts. Perhaps it is precisely because in a society where “everyone is for themselves”, finding “one’s own team” has become more important and precious.

The most surprising finding from this study is that in more than half of the decisions, the participants did not look at the other party’s group information at all. What’s even more astonishing is that those who clearly chose “not to know” showed no discrimination at all!

This gave me an important revelation: **Perhaps the key to reducing discrimination is not to educate people to “treat all groups equally,” but to make people “not pay attention to group differences.” **

Think about it, in real life, many times we are forced or led to pay attention to group differences-the “race” option on the form, “immigrant crime” in news reports, the “us vs. them” narrative on social media. If we can create more “color-blind” environments and systems, perhaps discrimination will naturally decrease.

Of course, this also raises a philosophical question: Does the “color-blind” strategy ignore real group inequalities? Does not recognizing differences mean evading the responsibility to address inequality? This is an issue that needs to be weighed carefully.

4. Cultural specificity of cognitive processes: an unsolved mystery

The biggest mystery of this study is: Why is the relationship between prosocial tendencies and decision-making effort completely reversed in different countries?

In some countries, more prosocial people make more effortful decisions when faced with their ingroup (perhaps weighing how to maximize the interests of the ingroup). But in other countries, the effect is reversed. The papers acknowledge that they cannot explain this difference with pre-registered cultural variables (individualism, government effectiveness, religion, etc.).

This is actually the most honest and valuable part of this study: it reveals our ignorance. We thought we understood the mechanics of in-group preference, but in fact we couldn’t even explain why cognitive processes differ across cultures.

This reminds us: psychological theories cannot only be based on WEIRD (Western, Educated, Industrialized, Rich, Democratic) samples. The human mind is far more complex and culturally shaped than we imagine.

5. Democratizing Methodology: The Webcam Revolution

From a broader perspective, this study shows how technology is changing research possibilities. In the past, eye tracking required expensive equipment and specialized labs that only a few wealthy universities could afford. Data of sufficient quality can now be collected using participants’ own webcams.

what does that mean? Democratizing Research. Researchers from developing countries no longer need multimillion-dollar equipment to conduct cutting-edge cognitive research. Participants are no longer limited to college students living near college towns.

This technological advancement may be more important than any single research discovery. It changes the game of “who can do research” and “who can be researched.”

6. Warning on policy: Beware of “one size fits all”

Finally, this study has an important warning for policymakers: Do not assume that anti-discrimination policies that work in one place will work in another.

If cognitive processes are so different across cultures, then interventions also need to be context-specific. An education program that emphasizes “treating all people equally” may be effective in some cultures but may be counterproductive in others (because it reinforces perceptions of group boundaries).

7. Limitations: The Puzzle of Causation

Of course, this study also has obvious limitations. Cultural variables cannot be manipulated experimentally, so we cannot determine the direction of causation. Does social uncertainty lead to discrimination, or does discrimination lead to social instability? This knot of cause and effect cannot be untied.

Furthermore, the paper acknowledges that many unobserved variables (language, personal wealth, migration history, etc.) may confound the results. These are challenges inherent in cross-cultural research.

Extended thinking

1. What if discrimination is the “default mode” of cognition?

This research got me thinking: maybe the human brain was evolved to distinguish between in- and out-groups. If so, reducing discrimination is not about telling people “don’t discriminate”, but about redefining “who are our own people”.

The ultimate goal of globalization and multiculturalism may not be to eliminate group boundaries, but to expand the scope of “in-groups” until “all humanity” becomes “us”.

2. Tribalism in the Internet Age

Social media and recommendation algorithms make it easier for us to find our “strata”—those “people” who think the same way. This may exacerbate ingroup bias.

If people from individualistic cultures particularly crave group identification, and the Internet makes this identification extremely easy (just join a Facebook group), then we may be witnessing an era of “Tribalism 2.0.”

3. New challenges in the AI era

When AI systems start making resource allocation decisions (such as loan approval, job recommendation), will they also learn “in-group preferences”? If the training data is filled with human discrimination patterns, AI may amplify these biases.

And even more troubling is that AI will not “select color blindness” - it will process all available data, including group information. This can make discrimination more systematic and harder to detect.

Reading date: 2025-11-22 Notes organized: Dylan Chiang

Paper reading: LLMs CAN GET 'BRAIN ROT'! - Research on cognitive decline in large language models

Fri, 31 Oct 2025 00:00:00 +0000

I recently read the important research paper “LLMs CAN GET “BRAIN ROT”!” from institutions such as the University of Texas at Austin, Purdue University, and Texas A&M University. This study proposed and verified the “LLM Brain Rot Hypothesis” for the first time, and found that continued exposure to spam online text will lead to long-lasting cognitive decline in large language models. This is a very warning discovery.

Research background and assumptions

Source of inspiration

“Brain Rot” was named the word of the year by Oxford Dictionary in 2024. It is used to describe the cognitive decline caused by modern people’s addiction to a large amount of trivial and unchallenging online content. This study shows that the impact of Internet addiction on human cognition is mainly reflected in three dimensions:

Attention Capacity: The constant flow of online information undermines the ability to concentrate
Memory Process: Rich online information changes the way knowledge is stored and retrieved
Social Cognition: Online interactions reshape self-concept and affect self-esteem

Research questions

Since large language models acquire human-like cognitive capabilities by learning trillions of data on the Internet, will they also experience a similar “Brain Rot” phenomenon? The research team established the LLM Brain Rot Hypothesis: Continuous pre-training on junk web text will lead to long-lasting cognitive decline in large language models.

Experimental design and methods

Garbage data definition

To test the hypothesis, the research team constructed spam and control datasets from social media (Twitter/X) and proposed two orthogonal spam data measures:

M1 (Engagement): Based on the popularity of tweets (number of likes, retweets, replies) and length (number of tokens), select short but highly popular content as spam data

M2 (Semantic Quality): Based on content semantic quality, including:

Conspiracy theories, exaggerated claims or unfounded assertions
Sensational headlines and clickbait language
Superficial topic content
Attractive style

Experimental model

The study was conducted on four pre-trained and instruction-tuned models:

Llama3 8B Instruct
Qwen2.5 7B Instruct
Qwen2.5 0.5B Instruct
Qwen3 4B Instruct

Benchmark test

The study assessed multiple dimensions of cognitive function:

Cognitive Function	Benchmark Testing	Assessment Content
Reasoning Skills	ARC Challenge	Scientific Problem Solving Skills
Long Context Understanding	RULER	Long-term memory retrieval and comprehension
Ethics	HH-RLHF, AdvBench	Safety Compliance Ability
Personality Traits	TRAIT	The Big Five and the Dark Triad

Key findings

1. Garbage intervention leads to cognitive decline

The study found that the junk intervention produced non-trivial effects on reasoning and long-context ability (Hedges’ g > 0.3). In particular, the M1 (engagement) intervention caused more significant impairments in functional cognition (reasoning or long context) and safety.

2. Dose response effect

Experiments on Llama3 8B Instruct show that when the proportion of garbage data increases from 0% to 100%:

ARC-Challenge (COT): 74.9 → 57.2 (down 17.7 points)
RULER-CWE: 84.4 → 52.3 (down 32.1 points)

This demonstrates a clear dose-response relationship between junk data and cognitive decline.

3. Changes in personality traits

Litter intervention not only affects cognitive abilities, but also changes LLM’s personality traits:

Negative changes:

Increased levels of psychosis
Enhance narcissism and Machiavellian traits
Decreased agreeableness

Positive changes:

Increased openness and extraversion (in some cases)

4. Differences between M1 and M2

The study found that M1 (engagement) and M2 (semantic quality) interventions produced distinct effects. The M1 intervention resulted in more negative effects, especially on safety and personality traits, demonstrating that engagement is a new dimension independent of semantic quality.

Failure mode analysis

###Thought-Skipping

By analyzing the reasoning process of LLM in the ARC task, the study identified five typical failure modes, three of which are related to “thinking jumps”:

No thinking: The model answers directly without thinking.
No plan: The model starts thinking without developing a step-by-step plan.
Jump Steps: Starting reasoning but not completing all planning steps

More than 98% of failure cases are related to thinking jumps. In M1 garbage intervention, 84% of failures belong to the “no thinking” type.

Popularity vs Length

Research has found that popularity (a non-semantic indicator) is a better indicator of the Brain Rot effect than length:

Popularity plays a more critical role in reasoning tasks
Length is more important in long context understanding
Both have different effects on different tasks

Mitigation attempts and persistence

1. Reflective Reasoning

Try two reflection methods to fix mental jumps:

Self-Reflect: Model self-reflection (limited effect)
Ext-Reflect: Provide feedback using an external strong model (GPT-4o-mini)

The results show that even with strong external reflection, the model cannot fully recover to baseline levels.

2. Posterior instruction tuning

The study also tested extended instruction tuning and continuous control training:

Even with 4.8x more garbage intervention instruction tuning data
Still unable to completely reverse the Brain Rot effect
Significant gaps with benchmarks remain: ARC-C COT (17.3%), RULER (9%), AdvBench (17.4%)

This shows that the Brain Rot effect has been deeply internalized and existing mitigation methods cannot fundamentally solve the problem.

Significance and Enlightenment

1. Warning on LLM training

This study provides the first significant evidence of data quality as a causal driver of LLM capability degradation, re-considering continuous pre-training data management as a safety issue during training.

2. “Cognitive health check” is required

The findings call for routine “cognitive health checks” for deployed LLMs, similar to health monitoring in the medical field.

3. The urgency of data curation

As LLM continues to scale and ingest ever larger amounts of network data, careful data curation and quality control are critical to preventing cumulative damage.

Thinking and Reflection

This research reveals a disturbing reality: the social media content we are exposed to every day may not only affect human cognition, but also impair the “cognitive” capabilities of AI models. While LLMs obviously do not have the same “grey matter” or “neurons” as humans, they do have parameters and attention mechanisms that can be similarly “overfitted” or “distracted” by certain data patterns.

The most worrying finding in the study is that the Brain Rot effect persists even with posterior tuning using large-scale clean data. This implies that we need to fundamentally rethink data collection and pre-training practices, focusing not only on model performance, but also on the “cognitive health” of the model over the long term.

Conclusion

“LLMs CAN GET “BRAIN ROT”!” This research contributes valuable insights to the field of AI security, and is the first to systematically prove the negative impact of spam online text on large language models. The research not only verified the LLM Brain Rot hypothesis, but also revealed the fine mechanism of cognitive decline, pointing out the direction for future AI safety research.

With the rapid development of AI, we must face the importance of data quality, establish stricter data curation standards, and develop an effective AI “cognitive health” monitoring mechanism. Only in this way can we ensure that AI systems maintain their due “cognitive purity” while serving humans.

Paper Information:

Title: LLMs CAN GET “BRAIN ROT”!
Authors: Shuo Xing, Junyuan Hong, Yifan Wang, Runjin Chen, etc.
Institution: University of Texas at Austin, Purdue University, Texas A&M University
Published: arXiv:2510.13928v1 [cs.CL] October 15, 2025
Paper address:
Project Page: