Ryan Kolln, CEO and Managing Director of Appen.
AI has impacted the entire globe, but research shows that even the most advanced multilingual systems continue to fall short. What they’re missing is nuance, or the cultural cues within language that convey meaning beyond words.
This becomes more apparent as governments and enterprises adopt multilingual AI at scale, increasingly raising the stakes. AI that misunderstands culture can unintentionally distort meaning, alienate users and even cause harm in extreme cases. Cultural understanding shouldn’t be considered an add-on or a bonus feature in AI. It’s an absolute must if we want AI that’s more trustworthy, safer and more effective in the real world.
Why Does Cultural Intelligence Matter In AI?
Language is often a reflection of the culture that shapes it, impacting tone, idioms, dialects and even silence across regions. Not surprisingly, AI systems trained on narrow or homogeneous data often miss these subtleties and interpret language more literally without the right context.
This is not theoretical. Even something as basic as language translation tools can cause serious harm, as it did in 2017 when a Palestinian man was arrested after Facebook’s translation system changed “good morning” to “attack them” in Hebrew and “hurt them” in English.
Not all examples are this extreme, but literal translation without context can produce wildly inaccurate meaning.
This is a significant issue, with studies suggesting that nearly half (47%) of contextual meaning may be lost through conventional machine translation, especially when systems have low regional variety. In areas like healthcare or public safety, which can deal with actual life-or-death situations, a mistranslated instruction or misinterpreted alert can have dire consequences.
Governments have responded with national AI strategies that emphasize responsible and culturally grounded AI. As AI becomes increasingly vital to delivering public services, it needs to recognize the communities in which it operates.
Strategies For Building Culturally Adaptive AI
Enterprises that want global AI systems to communicate respectfully and accurately across regions and continents should focus on three principles:
1. Invest In Representative, Context-Rich Data
Data must reflect the true diversity of languages and dialects, not just the “standard” forms. This includes slang, regional expressions and cultural behaviors that influence meaning.
2. Prioritize Human Evaluation
People who are familiar with local culture should assess models for factors such as fluency, coherence, cultural relevance, politeness norms and the accuracy of idioms (not just literal correctness). The best AI systems are supported by human judgments and go beyond automated metrics.
3. Design For Full Localization
Culturally adaptive AI aims to interpret meaning in local context. As a basic example, think of it as replacing a phrase like “raining cats and dogs” with a local equivalent rather than translating it literally.
Ultimately, effective AI mirrors how humans naturally communicate. We don’t only speak a language, we navigate cultures, and AI should reflect that as well.
So what does this look like in practice? For one tech company, its large language model was deployed globally, initially supporting 10 dialects across five languages. After incorporating structured feedback from native speakers, including realistic dialogue, ranked outputs and refined cultural alignment, the model ultimately expanded to more than 70 dialects and 30 languages.
The takeaway: culturally rich data and large-scale human evaluation led to unmatched cultural intelligence and, most importantly, a smarter model.
Speak Our Languages, Respect Our Cultures
In AI, speed and performance are table stakes. The emerging generation of systems is being defined by their ability to communicate authentically with people from different communities. Cultural adaptation is an opportunity for innovation and builds trust with users around the world.
If they’re not already, tech leaders need to begin planning for this shift now, with pipelines for culturally rich data, expertise from native speakers and systems that respect regional identity and nuance. These foundations are the baseline for multilingual AI adoption.
You can’t separate language from identity. AI that only understands words will never fully understand people. And the leaders in multilingual AI will be those who treat culture as central to communication.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?
