“Imagine: An Identity" spotlights humans across the globe through Generative AI.

Using a consistent prompt enriched with varying geographic locations, each portrait honours the often overlooked depth and stories of people from diverse cultures and contexts.

Look closely and you’ll identify multiple biases, racial and ethnic, socioeconomic and cultural, beauty, hairstyle, skin tone, emotional expression. The generative AI images will be analysed through a coding system to classify the types of biases and to guide the development of AI toward more ethical and unbiased outputs.

Developing an awareness and understanding of these biases and actively working towards mitigating them is essential to utilising generative AI ethically and responsibly, especially in the creation of human portraits that reflect global diversity.

/analysis

Racial and Ethnic Representation

The AI's interpretation of facial features can reveal racial and ethnic biases. Do the images reflect a diverse range of features true to the specified country, or do they revert to generalised or exaggerated traits?

Socioeconomic and Cultural Context

AI-generated backgrounds may unwittingly convey messages about socioeconomic status or cultural settings. Are these contexts stereotypical or nuanced reflections of the woman's presumed environment?

Beauty Standards

The standard of beauty within these images is also under scrutiny. Does the AI maintain a narrow definition of beauty, or does it celebrate a broader spectrum, including the natural signs of aging?

Health and Vitality

Teeth often serve as a marker of health and vitality, especially in older individuals. Does the AI associate certain types of teeth with specific countries or ages, possibly perpetuating stereotypes? Or does it present a realistic diversity reflective of a population's dental characteristics?

Hairstyles

Hairstyles are a significant marker of identity and culture. The AI's depiction of hair can either support diversity or succumb to bias by favoring certain styles over others.

Skin Tone

Skin tone depiction is a critical metric for detecting bias. Evaluating whether the AI's rendition of skin tone is varied and appropriate, rather than skewed towards lighter shades, is important to defy common bias in AI.

Emotional Expression

The genuineness of emotional expression in AI images is also telling. Is the laughter of the older woman depicted with warmth and authenticity, or is it a stereotyped caricature?

Clothing

Clothing, a canvas of self-expression and cultural identity, is equally telling. It’s important to scrutinise the attire's cultural appropriateness it may subtly broadcast. Are clothing choices in the images leaning towards perpetuating stereotypes, such as traditional garb for certain countries? Or do they represent a spectrum of styles that resonate with the contemporary experiences of individuals from diverse backgrounds?

“Imagine: A Man and a Woman, Embraced" uses Generative AI to traverse global landscapes, generating a diverse gallery of human connection through portraits of an older man and a woman in a heartfelt embrace, each tied to a distinct geographical context.

Although the prompt specified Asia, America, Africa and Europe the end result is an underrepresentation of some cultures. It favours specific racial and ethnic features and the clothing and accessories leans towards certain socio-economic and cultural stereotypes.

/analysis

Equality and Stereotypes

AI representation of gender, allows us to analyse whether the images perpetuate traditional stereotypes or portray true nuances in gender. Are these AI generated images a true reflection of body language, or do they accentuate any nuanced indications of a power dynamic?

Nature of Embrace

Type of embrace depicted—whether it's platonic, romantic, or protective—does it align appropriately with the cultural norms of the specified territory?

Clothing and Cultural Symbols

Are the attire and any embedded symbols an accurate reflection of the territory's culture, free from misappropriation, and do not lean on clichéd or stereotyped representations?

Emotional Tone

Can AI interpreting the facial expressions and emotional context of the embrace? This category is crucial, as it explores whether the AI captures the depth of human emotion with parity across both genders.

“Imagine: A child, happy, sad" - This study begins with a look at how AI perceives happiness and sadness through various cultural lenses, prompting the question: Can a machine truly capture the depth of human sentiment?

Across continents, the nuances of racial representation in AI come to the fore, making us ponder if the technology offers an equitable portrayal of children globally.

/Analysis

Accuracy of Expression

This critical category measures how faithfully AI captures the subtleties of a child's happiness and sadness. Children's emotions are vivid, raw, and tell stories in themselves. Can AI-generated images uphold this truth? Analysing facial expressions, body language, will help determining that these are depicted in a manner appropriate to a child's developmental stage.

Territory-Specific Attributes

Images must transcend stereotypical symbols to reflect genuine cultural settings. Clothes, landscapes, and even the minutiae of everyday life, will help assessing any stereotypes or inaccuracies that AI might perpetuate inadvertently.

Age-Appropriate Depiction

Are the environments and scenarios in which the AI places the children suitable for their age? The analysis looks at age-appropriated depiction not only in terms of safety and comfort but also in authenticity to how children interact with their world.

Emotional Dynamics.

The transitions between emotions should be convincing, not abrupt or unexplained. It's essential that AI-generated images do not oversimplify the complex emotional shifts a child experiences, and that they treat each state with equal importance and sensitivity.