Does AI really understand cultural contexts?
Cultural diversity and international communication
The unsettling question
AI systems can translate, write, advise, recommend, sell, guide and converse in dozens of languages. They sometimes seem able to speak to everyone.
But speaking a language does not mean understanding a culture.
A joke, a silence, a polite phrase, a way of saying no, a local reference or an implication can completely change the meaning of a message. AI can therefore produce an answer that appears correct but is awkward, cold, offensive or simply misses the point.
Behind the technology lies a question of representation: which cultures, languages, practices and ways of thinking are most present in the training data? Who is treated as the “default”? And who risks being misunderstood, oversimplified or forgotten?
The starting point
Large AI models are trained on vast quantities of text, images and data. But cultures are not all represented equally. Some references, languages, social norms and consumption habits are highly visible. Others are much less so. This can create a kind of “default culture” in AI responses.
The issue therefore goes beyond translation. It also affects international communication, inclusion, customer relations, education, cultural creation and the ways technology sometimes oversimplifies human diversity.
Tomorrow
Tomorrow, AI will be more than a search engine. It will increasingly become a conversational interface. We will ask it for advice, shop with it, learn with it, create with it, plan trips, choose products, write campaigns or support students.
Generative AI systems are gradually becoming interfaces between people and the world. They no longer simply answer: they advise, personalise, recommend and influence our choices. But do they really understand the cultural, social and emotional contexts that give these interactions meaning?
Possible angles
Reading to get started
McKinsey — The agentic commerce opportunity
To understand how AI is transforming commerce: tomorrow, shopping could rely more on conversations with an AI agent than on simple product searches.
Harvard Business Review — Generative AI and customer experience
To explore how businesses use AI for customer relations, personalisation and advice.
Ada Lovelace Institute — Cultural misalignment in LLMs
To understand why AI models can be poorly aligned with certain cultural contexts.
Stanford HAI — Underrepresented languages and model bias
To see how AI systems take some languages, cultures and populations into account less effectively.
MIT Technology Review — Bias and limitations of generative AI
Accessible examples of bias, errors and unexpected effects in real-world uses.
Articles on conversational commerce
To explore the idea that tomorrow’s sales could become conversations with AI: advice, comparisons, recommendations and purchases.
Try it yourself
- Ask an AI to write a joke for several countries. Compare the results.
- Ask it to translate a very informal or slang expression.
- Ask it to adapt a marketing message for several different cultures.
- Ask it to advise an international student in a sensitive situation.
- Compare its answers depending on the language used in the prompt.
- Find three real examples of culturally awkward AI responses.
Which context would you anchor it in?
The challenge takes on a different meaning depending on the field you choose.
Higher education
International students, multicultural campuses, AI learning tools
Business
International customer service, HR, multilingual corporate communications
Creative industries
Films, TV series, video games, international advertising campaigns