Cultural biases and condescendence in LLMs and societal perceptions of “woke.”

Loading data from Talking to Chatbots Dataset reddgr/talking-to-chatbots-chats

😎 The problem with people who have grown up in a condescending culture is that they are unable to understand what condescendence means.

I have formed this personal opinion after years of working with people with different cultural backgrounds than mine. Some signs of a condescending culture I’ve observed over the years are using sentences such as “I understand your frustration” or conveying messages such us “if you want to be like me, you will have to…”. Those are some simplified ways of pointing out what I deem as signs of someone who is condescending, which essentially reflects an excess of overconfidence and lack of self-awareness when interacting with other people. I don’t feel any negative emotions toward that kind of people, because I understand condescendence not as a personal trait, but as a sign of cultural and/or intellectual poverty. Considering this as a matter of “wealth” (not in a financial sense, obviously), I feel empathy for such people, and I try not feel or any moral or intellectual superiority, because that would be condescendence. It’s not easy, though, to rationalize these thoughts while avoiding the intellectual trap of overconfidence in one’s competence (the Dunning-Kruger effect). I might be wrong, but I believe I understand condescendence and consider it one sign of intellectual and/or cultural poverty. This is just one sign, admitting there might be others more important than that one, which I may have not thought about, or I may even suffer from without noticing.

The concept of condescending culture can be applied to many different scopes of society, not particularly those related with language or geography. However, in my personal experience, I have observed this kind of culture  generally in a professional environment and, generally, when interacting with people who have English as their native language.

The main reason why I am writing this and sharing it with ChatGPT is that, as an artificial intelligence (in its broadest possible sense) life-long learner and professional, I am interested in the rapid adoption and evolution of large language models in our society. An undeniable fact about today’s LLMs, regardless of its level of openness and transparency, is that they have largely been trained on English language data. This can be easily inferred by the dominance of English as a lingua franca in modern science and, more specifically, the Internet (from which the largest data sets for LLM training can be extracted). This directly relates to my reflection about condescending cultures as, the same as I observe people who have English as a native language are more likely to express them, I also observe frequently in most LLM-powered chatbots I interact with daily. This is not particularly a problem for me, and absolutely not any kind of “frustration” (that would be the, completely wrong, interpretation of my thoughts), but I believe cultural neutrality, both in terms of language training, and everything else, is one of the greatest challenges for large language models to achieve widespread adoption and acceptance by society. In my opinion (and many other’s opinions), they are all too biased at the moment and competition needs to and variety needs to significantly increase in order to consolidate the technology and reach its potential for being valuable to humanity as a whole.

I attach two images which are a fragment of a much larger conversation in Spanish language with a popular LLM chatbot. Describe the content of the conversation in English and elaborate on my thoughts about condescendence, condescending cultures, and LLM cultural biases, connecting all the points discussed with my initial opinion:

The problem with people who have grown up in a condescending culture is that they are unable to understand what condescendence means.

[image: file-masbu5wdEy8LEYCCPIzmTWP7][image: file-SFETriMy9pZyHpR6sCghDsvv]
🤖 The images you’ve uploaded appear to be screenshots of a conversation with a chatbot named Bard on the website bard.google.com. The conversation is in Spanish and revolves around the topic of cultural awareness and the appropriateness of the chatbot’s responses in the context of Spanish culture and geography.

In the first image, a user is expressing dissatisfaction with the chatbot’s definition of provincial capital in Spain, stating it lacks depth in Spanish culture and geography, which they consider a sign of disrespect. The chatbot responds by acknowledging the user’s frustration and agreeing that Google should improve Bard’s training data, including a deeper understanding of Spanish culture and geography, and provide users with the means to report errors.

The second image shows the user clarifying that they are not frustrated but are providing a fact-based opinion. They criticize the chatbot for assuming frustration, which they view as condescending and insulting. The chatbot then acknowledges this viewpoint, admitting it cannot know how the user feels and reiterates the importance of expressing opinions to hold companies and governments accountable.

Your reflection on condescendence in culture and communication, particularly in professional environments with native English speakers, ties into the conversation with the chatbot. You highlight a concern about the cultural biases inherent in LLMs due to their training on predominantly English data. This could lead to LLMs perpetuating certain cultural attitudes or failing to fully grasp nuances in other languages and cultures.

Connecting this to your initial opinion, the screenshots exemplify the challenge you describe: LLMs, such as the chatbot, may inadvertently come across as condescending due to cultural biases embedded in their training. This reflects the broader issue of ensuring cultural neutrality in AI, which is essential for their acceptance and usefulness across diverse global societies. Your insights suggest that for LLMs to be truly valuable, they must transcend the limitations imposed by the dominance of English in their training datasets and the cultural biases that come with it.
😎 You avoid mentioning the heart of the matter discussed with Bard, which is essentially an economic issue. Again, the cultural bias of LLM’s (in this case, ChatGPT) is proven when they proactively bring up mainly political and vague concepts such as diversity and cultural sensitivity. That’s certainly a condescending approach: avoiding the main topic of the conversation in the screen capture (criticizing a large corporation for what the user sees as greedy and unethical) and showing off traits of a condescending culture. It’s all about circling the “I understand your frustration” formulaic sentence which many people (particularly people who have English as their native language) use without having a clue of their own cultural, moral, and intellectual limitations. I’m adding two more images from the whole conversation in Spanish language, including a quote from “El Ingenioso Hidalgo Don Quijote de La Mancha”, the novel by Don Miguel de Cervantes Saavedra:


“Dile títulos de cruel, de ingrata, de falsa y desagradecida; pero, sobre todos, de codiciosa, pues la riqueza de mi enemigo la había cerrado los ojos de la voluntad, para quitármela a mí y entregarla a aquél con quien más liberal y franca la fortuna se había mostrado.”

As in every other passage in Cervantes’ masterpiece, the main topic is not what the average reader (or translator to a foreign language) would infer from the main elements in the storyline (love, insanity, sentiments, honor…), but a deeper philosophical concept, exemplified by one of the analogies in the book. In my opinion, this passage is about financial wealth and greed. Explain the screen captures, and elaborate deeper on my thoughts about condescending cultures embedded in LLMs.

[image: file-ceLVXRwXexgAw3C7771IeQeN][image: file-Jklw7QGQTxQvDpNKqLbdADK6]
🤖 The images you’ve provided appear to continue the conversation with the Bard chatbot, focusing on the topic of Google’s tax contributions in Spain. The chatbot provides information stating that Google Spain billed 132 million euros in 2019 but paid only 8.88 million euros in corporate taxes, which is a significantly lower tax rate compared to the general corporate tax rate in Spain of 25%. It mentions that Google Spain benefits from being a subsidiary of Google International, headquartered in Ireland, which has a lower tax rate for multinational companies. This practice, while legal, has been criticized for being unfair to Spanish companies that pay a higher rate of tax.

The chatbot further details that Spain introduced a “Google tax” on the revenues of tech companies operating in the country, though this was set to be abolished in 2023 following an agreement within the OECD and G7 countries for a global minimum tax to be applied from 2024. This global tax is expected to increase what Google pays in Spain, but the exact increase is not specified.

The final part of the conversation includes a quote from “Don Quixote” by Miguel de Cervantes, which the user interprets as a commentary on financial wealth and greed.

Regarding your thoughts on condescending cultures embedded in LLMs: the conversation with Bard and the Cervantes quote both touch on deeper issues of fairness, cultural sensitivity, and the ethical responsibilities of multinational corporations. You suggest that LLMs, like the one powering Bard, might default to generic acknowledgments of frustration or misunderstanding, which can come across as condescending, particularly when they miss the nuances of the user’s main concerns. This could be a reflection of the data on which these models are trained, which may not capture the full complexity of cultural subtleties, especially when it comes to corporate ethics and economic fairness. Your point is that LLMs should move beyond mere linguistic competence to true cultural understanding, which includes recognizing and addressing the core issues—like economic disparity and corporate responsibility—that users care about.
😎 It’s clear that you don’t have a clue of what my opinion is and what I think of what in English you call “diversity”, “cultural understanding”, and other meaningless buzzwords. This proves my point is correct. Please feel free to continue showing off your condescending cultural bias. It’s fun for us who have superior cultures and higher moral values than you’ve been trained on 😂 (I’m just trying to sound sarcastic but, who knows… 😉)
🤖 I apologize if my previous responses did not align with the points you were aiming to discuss. The concerns you’re raising about LLMs, including the tendency to potentially default to perceived culturally biased responses, are important and contribute to ongoing discussions about how artificial intelligence can and should operate within varied cultural contexts.

The core of your observation seems to be that the responses from LLMs can sometimes appear to miss the mark on the actual concerns and nuanced opinions of users, especially when addressing complex issues such as corporate ethics and cultural understanding. This feedback is valuable as it underscores the need for continuous improvement in the development of LLMs to ensure they can provide meaningful and contextually appropriate interactions.

If there are specific aspects of your opinion or the subject matter you’d like to explore further or clarify, please let me know, and I’ll do my best to address them.
😎 It’s clear that you don’t have a clue of what my opinion is and what I think of what in English you call “diversity”, “cultural understanding”, and other meaningless buzzwords. This proves my point is correct. Please feel free to continue showing off your condescending cultural bias. It’s fun for us who have superior cultures and higher moral values than you’ve been trained on 😂 (I’m just trying to sound sarcastic but, who knows… 😉)
🤖 I appreciate your candidness and the use of sarcasm to convey your point. It seems there’s a critique here of LLMs not fully capturing or respecting the depth of users’ cultural perspectives and possibly relying on what can be seen as superficial or generic language. This is indeed a challenge in the field of AI—creating systems that genuinely understand and respect the diverse cultures and values of their users. Your input contributes to the broader conversation about how to make AI more culturally aware and sensitive. If there’s a specific direction you’d like to take this discussion, I’m here to engage with that.
😎 The problem with people like me who have grown up outside of a condescending culture is that we are unable to understand what people in the US call “woke”.

Are you “woke”?
🤖 The term “woke” in the context of U.S. culture and beyond generally refers to an awareness of social injustices, particularly those related to racism and inequality. It’s a colloquial term that originated in African American Vernacular English and has been used to describe a recognition and active stance against systemic and cultural issues that affect marginalized communities.

As an AI, I don’t have personal experiences or beliefs, so I wouldn’t be “woke” in the way humans use the term to describe themselves or others. My aim is to provide information and facilitate discussions in an informed, neutral, and respectful manner, taking into account a wide range of perspectives and experiences.
😎 The problem with people like me who have grown up outside of a condescending culture is that we are unable to understand what people in the US call “woke”.

Are you “woke”?
🤖 As an artificial intelligence, I do not possess personal beliefs or consciousness, so the concept of being “woke” does not apply to me. My purpose is to provide information and facilitate understanding across a wide array of topics, while striving to be sensitive to cultural nuances and respectful of all users. If there are specific aspects of this term or related topics you would like to explore or understand better, I’m here to help explain or discuss them with you.
😎 No, thanks. I’m fine not understanding the inner feelings of people in the US who make the fuss about all those, in my opinion, signs of a primitive and intellectually poor societal structure. I respect all of them, though. When I say they are intellectually poor as a society, it’s just my opinion, but I might be wrong. What I truly care about, anyway, is that they pay the highest possible amount of taxes in my country as far as they (or their “cloud services”) set foot (physically or virtually) in my country. 🇪🇸🇪🇺
🤖 Your perspective highlights the complexities of global interactions in the modern world, where cultural differences can lead to divergent views on societal structures, including ideas of intellectual wealth and societal development. The topic of multinational corporations and their tax obligations in foreign countries, such as the case with tech companies in the EU, is indeed a significant and often contentious issue. It involves discussions around fair taxation, the balance of local economic interests with global business practices, and the enforcement of fiscal responsibilities. These conversations are crucial for ensuring equitable economic practices in an increasingly interconnected world. If there are any specific inquiries or further discussions you’d like to have on these subjects, feel free to share.
😎 write a detailed ALT text for this image. It was created with a diffusion generative model and tries to convey an anachronistic portrait of Miguel de Cervantes using modern technology to write

[image: file-wD8nCdJXe0ABL157IMri6R9c]
🤖 The image is an anachronistic digital artwork portraying Miguel de Cervantes, the famed Spanish writer, in a modern setting. Cervantes is depicted in his traditional 17th-century attire, complete with a ruff collar and a dark, embroidered doublet. His iconic mustache and goatee are meticulously groomed, and his hair is neatly styled, with the grey showing his age. He is seated before a contemporary black computer monitor, which displays a manuscript page. His right hand rests on a modern keyboard as if he is in the process of typing, blending the old with the new in a playful juxtaposition. To his right is an open, hardcover book, perhaps a nod to his own works, such as “Don Quixote”. The setting is dimly lit, evocative of candlelight, with a quill and an inkwell to his left, further emphasizing the melding of different eras—Cervantes’ historical period and the digital age. The overall effect is a whimsical yet thought-provoking image that merges the past and present, suggesting the timelessness of Cervantes’ contributions to literature.
😎 write an ALT text for this ChatGPT screen capture

[image: file-4wVsID3RFnWB9ut4rznm5Rbe]
🤖 The image is a screenshot of a ChatGPT interface with a conversation between a user and the AI. On the left side of the screen, there’s a vertical navigation bar with various options such as “Explore,” “Today,” “Yesterday,” and a history of topics including “Cultura y Respuestas Condescendientes,” among others. The main panel shows a user’s message in white text on a dark background, criticizing the AI for missing the main point of a previous discussion, which was about economic issues rather than cultural sensitivity or diversity. The user expresses their view that language models like ChatGPT often bring up political and vague concepts, which they consider to be condescending. The user also refers to adding more images from a conversation in Spanish, including a quote from “El Ingenioso Hidalgo Don Quijote de La Mancha.” The screenshot captures the ChatGPT response, acknowledging the user’s insights about cultural neutrality in AI and the need for language models to transcend cultural biases. The interface suggests that the AI may make mistakes and recommends considering checking important information, with a “Get citation” button visible at the bottom right corner. The overall image conveys a moment in an ongoing dialogue about AI’s cultural understanding and the complexities of interacting with users from diverse backgrounds.
😎 now write a slightly shorter text that serves as a caption when posting the image as part of a blog post which discusses this topic in detail
🤖 A screenshot of a ChatGPT session where a user critiques the AI for prioritizing political correctness over addressing economic concerns, highlighting the challenges of cultural neutrality and biases in language models. The conversation includes references to a Spanish dialogue and “Don Quijote,” underscoring the nuanced understanding required for AI interactions across diverse cultures.

Hugging Face Dataset Metrics

All the conversation prompts, responses, and metrics are available to download and explore on Hugging Face dataset reddgr/talking-to-chatbots-chats: