Artificial Intelligence has made remarkable progress. From transforming customer service to powering strategic decision-making, its influence spans virtually every industry. But while its capabilities have grown exponentially, cultural understanding remains one of AI’s most significant blind spots.
We’ve followed these developments closely. In previous explorations such as AI Meets Culture and our Gemini vs. ChatGPT comparison, we've consistently found that even the most advanced systems fall short of cultural nuance. With the introduction of AI Agents by OpenAI, we revisited the experiment: can custom-trained agents finally bridge the cultural divide?
Understanding AI Agents
AI Agents represent the next evolution of generative AI. These custom-made personas are trained on targeted datasets and governed by detailed behavioural parameters. The goal is to emulate consistent, human-like engagement, offering deeper contextual relevance and more coherent personality traits. While this represents a meaningful technical step forward, and offers much improved performance on many fronts, is cultural understanding one of them?
The Cultural Test: Methodology Recap
To test whether these new agents show improved cultural accuracy, we replicated our original cultural experiment using the following process:
- Each AI Agent was equipped with a thorough explanation of the Six Dimensions of National Culture.
- A number of country-specific cultural scores was integrated into their training sets.
- Each agent was instructed to behave like an “average person” from their assigned country, specifically instructed to avoid stereotypes, extreme views, and caricatures.
- We evaluated responses using Mean Absolute Error (MAE) against validated cultural benchmarks found in our Country Comparison Tool. Read more about MAE in our previous article.
The Results: Minor Progress, Major Limitations
The MAE scores reveal only modest improvements. While Finland showed significantly better alignment, other countries remained far off the mark:
(Lower is better)
- Finnish Agent: 15.83
- Indian Agent: 30.17
- Austrian Agent: 31.17
- American Agent: 31.83
- Overall Average MAE: 27.25
(Previous study average: 30.69)
This data suggests the new Agent framework yields incremental improvements, but is still a long way from capturing cultural nuance. In fact, in several cases, performance even declined.
Why AI Still Misses the Cultural Mark
Culture isn’t just a collection of facts, it’s a lived, collective experience. It’s made up of unspoken norms, historical context, emotional responses, and deeply ingrained values. These are not easily encoded into training data. Despite improvements in language generation and personalisation, today’s AI models remain largely blind to the emotional and situational subtleties that define cultural behaviour.
What This Means for Global Leaders
For international teams and multicultural organisations, misreading culture isn’t a minor oversight, it’s a strategic risk. Miscommunication, eroded trust, and poor collaboration often stem from cultural disconnects. AI Agents may offer useful support, but they cannot yet replace human insight when it comes to navigating cultural dynamics, and should not be used for that purpose.
That's why our Intercultural Competence Programme remains essential. Designed to build real-world cultural fluency, it equips professionals with the tools and understanding AI still lacks.
Final Thought: Let AI Assist, But Keep Culture Human-Led
Despite the rapid evolution of generative AI, it remains fundamentally unfit for evaluating culture or cultural differences. Culture is not a dataset to be decoded, it’s a context-rich, emotionally layered, and deeply human phenomenon.
While AI can support operational tasks, it lacks the lived experience and intuition required to navigate cultural nuance. Relying on it to assess or interpret culture risks oversimplification at best, and serious misrepresentation at worst.
If you aim to leverage culture as a strategic asset, the priority must be people. Equip your teams with the tools, insights, and lived perspectives that AI cannot replicate. Use AI to streamline processes, but when it comes to cultural understanding, keep the leadership firmly human.