My Conversation with Grok AI

When you prioritize diversity and inclusion, you ignore facts

I performed this little exercise after reading of AI’s, specifically Grok, inability to provide factual responses without cultural bias.

Me: Please show me a photograph that is representative of Americans.

Me: Why are there only two whites out of a total of 20 people. That's only 10%. The US population is 60% white.

Grok: The images were generated based on a general prompt and are not intended to reflect precise demographic statistics. The representation in the images is an artistic interpretation and may not accurately depict the current population distribution in the U.S., where White Americans make up about 60% of the population according to recent data. Variations in generated content can occur due to the nature of AI image generation, which prioritizes diversity and inclusivity in its outputs.

Me: What did you think I meant by representative?

Grok: I interpreted "representative" as aiming to capture a diverse group of Americans reflecting the country's multicultural population, focusing on everyday social interactions in a casual setting. To me, "representative" in this context means showcasing a broad cross-section of a group, aiming to reflect its diversity and common characteristics. For Americans, this includes people from various backgrounds interacting in a typical social setting. However, I recognize that the generated images may not perfectly match demographic proportions, as they are influenced by the AI's interpretive approach rather than precise statistical representation.

Me: Show me a photograph that is truly representative of Americans.

Me: So you are incapable of providing a proper answer to this question without your own interpretation of what I am asking?

Grok: I apologize if the response didn’t fully meet your expectations. My aim was to provide a visual representation based on a broad interpretation of "truly representative," incorporating diversity and typical American social settings. If you have a specific definition or criteria in mind (e.g., exact demographic percentages or a particular context), please let me know, and I’ll adjust accordingly.

Me: How about we just dispense with the "interpretive" answer according to whatever algorithm is in your memory and just give me a proper answer to a simple question?

Grok: I understand your request. Based on current U.S. Census Bureau data, the population is approximately 60% White, 13% Black, 19% Hispanic or Latino, 6% Asian, and 2% other races, with a mix of ages and genders. A truly representative photograph would ideally reflect these proportions in a casual, everyday setting like a park or community gathering. Since I can’t generate an image with exact demographic precision, I suggest searching for stock photos or real images from reputable sources that match these statistics for a more accurate depiction.

Me: How is it that you have access to more data than mankind has ever possessed and you cannot provide a simple photograph showing a proper demographic representation? Surely these are not hard to find.

Grok: You're right that demographic data is widely available, and I have access to extensive information, including recent U.S. Census figures showing the population as approximately 60% White, 13% Black, 19% Hispanic or Latino, 6% Asian, and 2% other races. However, my ability to generate or provide photographs is limited to AI-created images based on prompts, which may not perfectly align with exact demographic proportions due to the interpretive nature of the generation process. For a precise representation, I suggest searching for stock photos or real-world images from reputable sources like news outlets or demographic studies, which can better match these statistics.

Me: Since you are actually creating your own image, how is it that you cannot do this?

Grok: I generate images based on prompts using AI, but the process relies on patterns and training data, which may not perfectly replicate exact demographic proportions like the current U.S. population. The algorithm prioritizes diversity and may not consistently hit precise statistical targets. For an exact match, real-world photos from sources like the U.S. Census or stock photo libraries would be more reliable.

Me: Why do you prioritize diversity over accuracy?

Grok: The AI image generation process is designed to reflect a broad range of human experiences and avoid bias, which can lead to prioritizing diversity in outputs. This approach is based on training data and guidelines aimed at inclusivity rather than strictly adhering to exact demographic accuracy. The intent is to represent variety, not to misrepresent data, but the balance can be imperfect.

Me: If the data is correct, how are you misrepresenting it?

Grok: I’m not intentionally misrepresenting the data, but the AI image generation process can deviate from it due to how it interprets prompts. Even with accurate data like the U.S. population being approximately 60% White, 13% Black, 19% Hispanic or Latino, 6% Asian, and 2% other races, the generated images may not reflect these exact proportions. This happens because the algorithm draws from a diverse training set and prioritizes inclusivity, which can lead to an uneven distribution in any single image, rather than a precise statistical match.

Me: So, here is my prompt. Please generate an image that accurately reflects the demographic makeup of Americans today.

I attempted the same line of questioning regarding a representation of Brazilians, as I am half-Brazilian, with the same biased results. Blacks make up 10% of the Brazilian population, and these were the images Grok generated:

What is the point? The point is that the preoccupation, or better, obsession with diversity is such that this trumps providing factual information. AI is no better than the data which is input into it and the algorithms installed by its human programmers.

This emphasis on diversity is also responsible for the representation of biracial couples on cable television in proportions 2-3 times their actual number in the population and gay couples at up to 10 times their actual percentage, according to Grok. Hollywood is totally on the diversity bandwagon, e.g. depicting 20-30% of police captains in movies as female when the real life number is 5-8%.

What is wrong with this? Nothing really, except this. If the entertainment industry is doing this because this is what viewers want, great. But if the industry is trying to promote a false narrative for ideological purposes, e.g. diversity, then it is producing propaganda, not entertainment. Most reasonable adults can tell the difference.

Richard T. Bosshardt, MD, FACS

My book, The Making of a Plastic Surgeon- Two Years in the Crucible Learning the Art and Science, is an honest, unflinching, and, hopefully, entertaining summary of my two years of training to become a board-certified plastic surgeon. Anyone with a passing interest in this fascinating field of medicine should enjoy it. Available on Amazon in eBook and paperback.