Everything in this book so far describes a human problem: a manager reads a gesture wrong, a coworker takes offence at a delay, a customer hears evasion in a word that meant humility. Human misreadings can be corrected. Somebody explains, somebody apologizes, somebody learns.

The situation changes when the misreading is encoded.

Artificial intelligence has become a significant part of employee recruitment, performance measurement, and the interaction between companies and their customers. Service encounters are handled by chatbots. Algorithms screen resumes. Employee productivity is rated by automated systems. The widespread deployment of AI in customer service, recruitment and employee evaluation has changed how businesses communicate with customers and how they evaluate the people who work for them.

That raises a question this research is unusually well placed to ask: can AI systems recognize and value culturally diverse service approaches, or do they propagate bias by treating Western communication norms as universal standards?

The evidence

The question is not speculative. Existing scholarship shows that large language models display methodical cultural bias.

Evaluations of leading models found that they align with the values of English-speaking and Protestant European countries, and are “most distinct from cultural values of African-Islamic countries.” Further evaluation of five widely used generative models found that they valued self-expression, minimal hierarchy, and individualist choices. Those values sit at the opposite end of the scale from the orientations described throughout this book: collectivism, high power distance, deference to authority.

The reason is not mysterious. These systems are trained predominantly on English-language sources produced in Western nations, and those sources overwhelmingly express individualistic communication styles, efficiency-oriented service expectations, and low-context interaction patterns. The assumptions come with the data.

So when a system trained that way encounters an employee who behaves as the participants in this study behave — warmth, deference, structure-seeking, indirect and context-heavy communication — it has a ready category for the behaviour. Inefficiency. Rigidity. Deviation from protocol. The same behaviours this research classified as assets.

Where it bites

Frontline service encounters are exactly where this contrast surfaces.

Western customers, from low-context cultures, generally prefer clear, direct, explicit communication, and evaluate service largely on efficiency and task completion. Arab, high-context cultures tend toward indirect and nonverbal communication that relies on situational context, and place weight on the quality of the relationship between customer and employee. A Western-trained system does not recognize high-context behaviour as effective. It recognizes it as a failure to complete the task quickly.

Chatbots designed for swift task completion may ignore the relationship-building dimension that collectivist employees consider the substance of the job — and may penalize workers for the indirect, context-heavy communication that is, in their own understanding, good service.

Recruitment systems raise the stakes further. AI-driven hiring frequently reproduces bias against non-majority groups, overlooking qualified candidates whose names, language styles, or educational backgrounds fall outside the norms the system was trained on. This is where the story of a man asked to call himself “Mo” stops being an anecdote about one uncomfortable manager. A manager can be persuaded. A résumé filter that has learned which names correlate with past hires cannot be argued with, and it does not know it is doing anything.

The chain runs like this. Humans interpret culturally different behaviour. Organizations build standards on those interpretations. Organizations encode those standards into software. Systems learn the assumptions embedded in the data and the environment. And a misunderstanding that used to happen once, in one shop, between two people, now happens at scale, consistently, invisibly, in every hiring round and every performance cycle.

The workers in this study were sometimes misread by managers. Their children will be misread by systems that never met them.

It is fixable

The same research that documents the bias also shows it is not a fixed property of the technology. Cultural value patterns can be modelled in AI systems, and large language models possess identifiable cultural orientations that can be shifted through targeted prompting or training methods.

One practical approach is to train systems on data explicitly tagged with metadata that identifies the cultural characteristics of service interactions, so that the system learns appropriate behavioural variation instead of treating Western norms as the universal default. Alongside that, multicultural service script libraries can be built to reflect a range of communication styles rather than only efficiency-driven, low-context patterns.

The measurement layer has to change with it. Behaviours reflecting relational warmth — including the long conversation — must be readable as business excellence rather than defect. Performance metrics that incorporate cultural dimensions can recognize strong customer relationships, trust, adaptability and loyalty as components of quality, instead of scoring only speed and brevity.

Acculturation belongs in that design too. People negotiate between heritage and host frameworks in situational, selective and dynamic ways — as every chapter of this book has shown. A worker navigating both Arab hospitality and Canadian formality is exercising a subtle competence, and it should be positioned as an asset. Systems used in recruitment, evaluation and customer service should not penalize individuals who have not yet adapted to host-country norms, and should be able to recognize the value of what newcomers contribute, including their skill at adjusting service behaviour across different organizational cultures, client groups and situations.

One caution belongs here, in the author’s own voice rather than the literature’s: most of this work has been done in controlled settings rather than live organizations. The theoretical groundwork is solid. The practical application is thin. What follows is a proposal, not a report from the field.