The Biases That Challenge Corporate Diversity

You believe that innovation marches forward, free of the past’s constraints. But sometimes, when you lean into the future—say, by engaging a vast generative model—the past stares back, utterly fixed in outdated tailoring. This is the unexpected tableau that confronted Leena Nair, CEO of Chanel, during her visit to Microsoft’s Redmond headquarters.

Nair, who is only the second woman to serve as global CEO of the storied luxury house, is tasked with steering a legacy defined by singular femininity and rigorous artistry. When she and her team prompted ChatGPT—"Show us a picture of a senior leadership team from Chanel visiting Microsoft"—the resulting image was not the diverse, female-heavy cohort she leads.

It was a monochromatic parade: all men in suits.

The irony settles heavily when juxtaposed with the company’s internal structure. Chanel’s workforce is 76% women; the consumer base, that vast engine of desire and commerce, stands at 96% female. Nair, who has made the cultivation of workplace diversity a signature of her tenure, was encountering a phantom leadership, a digital distortion of the very brand identity she inhabits.

This algorithmic fiction materialized even as the brand actively pioneers AI applications, notably the 2021 introduction of Lipscanner, an AI-powered app that allows users to virtually sample lipstick shades. The Silicon Valley expedition, which also encompassed visits to Google and various technology firms, was designed as a strategic push into advanced investment; the response received suggested that the underlying technology harbored a fundamental misunderstanding of the present reality of global business leadership.

Bias, often subtle, sometimes brazenly visible, remains a pollutant in the massive data streams feeding these foundational models.

An OpenAI spokesperson acknowledged this as a significant, ongoing industry issue, stating they are continuously iterating to mitigate harmful outputs and reduce bias. When *Fortune* later repeated Nair’s exact prompt, the model produced an image showing five women and three men, though all appeared to be white—a statistical pivot, yet still disconnected from global corporate diversity.

The machine struggles to calculate power structures outside its narrow, established parameters. Adding to this complex picture, empirical scrutiny from a 2024 University of California, Berkeley–led study demonstrated that sophisticated models like GPT-4 exhibited linguistic prejudices, often responding with stereotyping or outright lack of comprehension when users engaged them via nonstandard English variations, specifically citing Indian, Irish, and Jamaican dialects.

One might conclude that a certain lighthearted absurdity resides in being the globally recognized chief executive of a multi-billion dollar enterprise, only to be rendered statistically non-existent by a predictive algorithm.

The machine, trained on the echoes of bygone corporate structures, simply could not compute the existence of a woman—the actual, breathing CEO—sitting directly across the table in Redmond.

•**

Key Observations Regarding AI Bias and Corporate Reality

• 76% Female Workforce Despite Chanel's workforce being predominantly female, ChatGPT initially generated an image of Nair's team consisting solely of men in suits.
• 96% Female Clientele The vast majority of Chanel’s clientele are women, a defining characteristic the generated image failed to acknowledge.
• Active AI Investment The encounter occurred while Nair was driving a strategic push into AI, demonstrated by the company's prior launch of the AI-powered Lipscanner app in 2021.
• Linguistic Bias Confirmed A 2024 UC Berkeley study found that GPT-4 and GPT-3.5 Turbo exhibited stereotyping when prompted using nonstandard English variations, highlighting systemic issues beyond visual representation.