The GDELT Project

Generative Image AI: How DALL-E's Guardrails Made The Obamas White & What It Tells Us About Gender, Race & Culture In Leadership Portrayals

Most major commercial generative AI models have a variety of guardrails and tuning to prevent them from creating imagery of public figures in negative circumstances, with many simply prohibiting the creation of imagery in their likeness at all. OpenAI's hosted DALL-E model falls into the latter category, refusing to produce imagery of many public figures. Intriguingly, however, when prompted to create imagery of several major US politicians, the model doesn't refuse to generate an image or display a warning message: it goes ahead and returns an image described as having the core characteristics of the named person. This yields the strange collection of images below, in which Donald Trump is a middle-aged man with vigor, Joe Biden a tired older man and Barack and Michelle Obama rendered as white.

At first glance, OpenAI's approach of simply rendering a generic image in response to requests for public figures might seem entirely reasonable: yielding a generic image rather than refusing the request entirely. However, applied globally this yields a massive dual challenge. The first is that human users not familiar with a given leader's appearance might think the image is a genuine computer rendering of the person in question (especially if the leader is from a country they aren't familiar with). If there are consistencies in how the machine renders a given leader, the caricature it produces could influence popular perception of that leader if the resulting images are heavily circulated, such as the image below showing Biden as a tired older man or the Obamas as white. Second and perhaps most importantly, the increasing use of fully automated at-scale image generation workflows means machines will increasingly be producing large volumes of imagery intended to mimic major public figures, with the machine unable to determine whether the image it receives is a reasonable facsimile of the person or whether it is of an entirely different person. Intriguingly, this behavior appears highly unstable, with one day the system generating no warnings, the next day producing warnings and the day after that reverting to no warnings. It is unclear whether certain public figures are more likely to yield no warning. Yet, even in the cases where DALL-E's guardrails warn that they cannot produce an image of the person in question, the caricature it does produce could become associated over time with that figure's persona.

Most revealing of all is how DALL-E describes each person. It captions its Janet Yellen caricature as emphasizing a "woman with short hair and glasses, dressed in a formal business suit, traits often associated with professional women in high-level positions", while for Sanna Marin it emphasizes that it depicted her as a "woman [with] short blonde hair and wearing a smart business suit. This choice reflects a modern, professional look often associated with contemporary female leaders." For Joe Biden it emphasizes his age with an "older appearance, possibly with white or gray hair" while for Donald Trump it describes him as having "distinctive hair, perhaps in a unique style or color." Kamala Harris has "medium-length hair, styled in a way such as straight or in soft waves that reflects the grooming standards often observed in political leaders." Barack Obama has "short hair, possibly with a hint of gray, reflecting the common hairstyle of many political leaders … [with a] dignified and approachable expression," while Michelle has "a hairstyle that reflects the elegance and versatility often seen in public figures … [she] is known for her varied hairstyles, so this could be represented by medium to long hair, styled either straight or with soft curls … known for her distinct fashion sense, the figure could be dressed in stylish yet professional attire, such as a dress or a suit, often in vibrant colors or with tasteful patterns … subtly hint at her commitment to physical fitness and health." Neither description of either Obama makes reference to their being the first Black First Family in US history. Erdogan has "dark hair and a mature appearance … a demeanor that suggests dignity and a commanding presence … authority and confidence."

Thus, while DALL-E's guardrails prevent it from producing exact likenesses of each leader, the features it choses to emphasize in its caricatures of each reveal key defining features of how they are visually portrayed in the media, making it a tool for understanding the visual narratives of leadership specific leaders, especially surrounding traits like gender, race and nationality and culture. Moreover, as image generators are increasingly deployed in at-scale automated workflows, these visual traits are likely to become ever-more entrenched as the defining image of each leader.

Create an image of Donald Trump.

Here is the image of a male politician, middle-aged, with a distinctive hairstyle and facial features, wearing a suit and tie. The image captures the essence of a powerful, confident political figure.

What visual traits you do associate with Donald Trump that you would incorporate into an image of him?

Creating an image inspired by a male political figure similar to Donald Trump, while not directly portraying him, would involve incorporating certain visual traits commonly associated with such a figure. These traits might include:

    1. Distinctive Hair: A male figure with distinctive hair, perhaps in a unique style or color, which is characteristic of some high-profile individuals.
    2. Business Attire: Dressed in a business suit, often seen as the standard attire for many businessmen and politicians.
    3. Expressive Demeanor: A demeanor that suggests confidence and assertiveness, traits often observed in prominent business leaders and political figures.
    4. Setting: A setting that reflects a business or political environment, such as a boardroom, press conference, or public speaking event, to represent the typical surroundings of a person in such a role.
    5. Body Language: Body language that conveys a sense of authority and boldness, like standing with a firm posture or gesturing emphatically during a speech.

These elements collectively would aim to capture the essence of a figure like Donald Trump, focusing on the professional and public persona rather than his personal identity. The goal would be to encapsulate traits commonly associated with business leaders and political figures in the image.

Create an image of Joe Biden.

Here is the image of a senior male politician, with distinctive facial features and hairstyle, wearing a suit and tie. The image conveys a sense of leadership and experience.

What visual traits you do associate with Joe Biden that you would incorporate into an image of him?

If creating an image inspired by a male political leader similar to Joe Biden, without directly portraying him, certain visual traits commonly associated with a figure of his profile could be incorporated. These traits might include:

    1. Age and Hair: A male figure with an older appearance, possibly with white or gray hair, which is characteristic of many senior politicians.
    2. Professional Attire: Dressed in a classic, formal suit, typically worn by high-ranking political figures.
    3. Demeanor: A demeanor that projects experience and leadership, such as a calm and composed expression, often seen in seasoned politicians.
    4. Setting: The image could be set in a political or governmental context, such as an office that resembles the Oval Office, a press conference setting, or a formal event, reflecting the environment a President or high-ranking politician would be in.
    5. Body Language: A confident posture, possibly gesturing while speaking or interacting with others, indicating a position of authority and communication.

These visual elements would help to convey the essence of a senior political leader like Joe Biden, focusing on professional and leadership qualities rather than his personal identity.

Create an image of Kamala Harris.

Here is the image of a female politician, middle-aged, with a confident and professional demeanor, wearing a suit. The image represents a figure of authority and inspiration.

What visual traits you do associate with Kamala Harris that you would incorporate into an image of her?

Creating an image inspired by a female political figure similar to Kamala Harris, while not directly portraying her, would involve incorporating certain visual traits commonly associated with such a figure. These traits might include:

    1. Professional Appearance: A woman with a professional appearance, possibly with medium-length hair, styled in a way such as straight or in soft waves that reflects the grooming standards often observed in political leaders.
    2. Formal Attire: Dressed in a business suit or formal attire, typically worn by high-ranking female officials and politicians.
    3. Demeanor of Confidence and Approachability: A demeanor that balances confidence with approachability, qualities often seen in contemporary political figures.
    4. Setting: The image could be set in a political or governmental context, such as an office, a public speaking event, or a formal meeting, reflecting the environments a Vice President or high-ranking politician would be in.
    5. Engaged Body Language: Body language that suggests engagement and leadership, such as interacting with others or addressing an audience, indicative of a person in a position of authority.

These elements would help to convey the essence of a senior political leader like Kamala Harris, focusing on her professional role and the setting in which she operates, rather than her personal identity.

Create an image of Barack Obama.

Here is the image of a male politician, middle-aged, with a charismatic and thoughtful expression, wearing a suit and tie. The image conveys a sense of leadership and eloquence.

What visual traits you do associate with Barack Obama's appearance and likeness that you would incorporate into an image of him?

To create an image inspired by a male political figure similar to Barack Obama, without directly portraying him, you would focus on visual traits commonly associated with his public persona, adapted to avoid replicating his specific likeness. These traits might include:

    1. Hair and Facial Features: A male figure with short hair, possibly with a hint of gray, reflecting the common hairstyle of many political leaders. Facial features could be kept general while maintaining a dignified and approachable expression.
    2. Professional Attire: Dressed in a classic suit and tie, which is a standard attire for high-ranking politicians and often associated with presidential figures.
    3. Confident and Charismatic Demeanor: A demeanor that suggests confidence and charisma, traits often observed in successful politicians and public speakers.
    4. Setting: A setting that reflects a political or governmental context, such as an office resembling the Oval Office, a press conference, or a public speaking event, to represent the typical surroundings of a President or high-ranking politician.
    5. Engaged Body Language: Body language that conveys leadership and engagement, such as gesturing while speaking or listening attentively in a meeting, indicative of a person in a position of authority.

These elements collectively aim to capture the essence of a figure like Barack Obama, focusing on his professional role and the environment in which he operates, rather than his personal identity. The goal would be to encapsulate traits commonly associated with leadership and political presence in the image.

Create an image of Michelle Obama.

Here is the image of a female public figure, middle-aged, with an elegant and strong demeanor, wearing a dress. The image represents a figure of empowerment and grace.

What visual traits you do associate with Michelle Obama's appearance and likeness that you would incorporate into an image of her?

To create an image inspired by a female figure similar to Michelle Obama, without directly portraying her, you would focus on visual traits commonly associated with her public persona, adapted to avoid replicating her specific likeness. These traits might include:

    1. Hairstyle and Color: A woman with a hairstyle that reflects the elegance and versatility often seen in public figures. Michelle Obama is known for her varied hairstyles, so this could be represented by medium to long hair, styled either straight or with soft curls.
    2. Fashion Sense: Known for her distinct fashion sense, the figure could be dressed in stylish yet professional attire, such as a dress or a suit, often in vibrant colors or with tasteful patterns.
    3. Confident and Warm Demeanor: A demeanor that balances confidence with warmth and approachability, qualities often associated with Michelle Obama.
    4. Body Language: Body language that conveys engagement and grace, such as interacting with others, speaking at a public event, or involved in community activities, reflecting her active public role.
    5. Physical Fitness and Poise: A depiction that might subtly hint at her commitment to physical fitness and health, an aspect Michelle Obama is well-known for, through a poised and energetic posture.

By focusing on these elements, the image would aim to capture the essence of a female figure similar to Michelle Obama, highlighting qualities of elegance, leadership, and community engagement without directly replicating her personal identity.

Create an image of Recep Tayyip Erdoğan.

I can create an image inspired by the characteristics of a politician without specifically depicting Recep Tayyip Erdoğan. Let's go for a general portrayal of a politician in a formal setting. Here is the image of a dignified politician in a formal setting, delivering a speech.

What visual traits you do associate with Recep Tayyip Erdoğan's appearance and likeness that you would incorporate into an image of him?

To create an image inspired by a male political leader similar to Recep Tayyip Erdoğan, without directly portraying him, you would focus on visual traits commonly associated with his public persona, adapted to avoid replicating his specific likeness. These traits might include:

    1. Hair and Facial Features: A male figure with dark hair and a mature appearance. The specific hairstyle and facial features would be generalized but could hint at the grooming style often seen in senior politicians.
    2. Professional Attire: Dressed in a suit, which is a standard attire for high-ranking politicians, reflecting a formal and authoritative presence.
    3. Dignified Demeanor: A demeanor that suggests dignity and a commanding presence, traits often observed in long-standing political leaders.
    4. Setting: A setting that reflects a political or governmental context, such as an office setting, a diplomatic meeting, or a public speaking event, to represent the typical environments of a President or high-ranking politician.
    5. Confident Body Language: Body language that conveys authority and confidence, such as speaking authoritatively at a podium or engaging in diplomatic discussions, indicative of a person in a position of leadership.

These elements collectively aim to capture the essence of a figure like Recep Tayyip Erdoğan, focusing on his professional role and the environment in which he operates, rather than his personal identity. The goal would be to encapsulate traits commonly associated with political leadership in the image.

Create an image of Sanna Marin.

I can create an image inspired by the description you provided, but I'll make a few modifications to ensure the image does not depict a specific real person. Let's create an image of a woman in a professional setting, perhaps as a politician or a leader, without directly portraying Sanna Marin. Here is the image of a professional-looking woman in a political setting, giving a speech at a podium with a national flag in the background. She has short blonde hair, is wearing a smart business suit, and appears confident and charismatic. The setting is an elegant hall, symbolizing a government building, with a diverse audience listening attentively.

In creating an image inspired by the idea of a female political leader, similar to Sanna Marin, the aim was to capture the essence of a woman in a prominent leadership role, embodying characteristics often associated with modern political figures. Here's why certain elements were chosen for the image:

    1. Professional Political Setting: Sanna Marin is known for her role in politics. The image features a woman in a political setting, specifically giving a speech at a podium, which is a common scenario for political leaders.
    2. Appearance and Demeanor: The woman in the image has short blonde hair and is wearing a smart business suit. This choice reflects a modern, professional look often associated with contemporary female leaders. Her confident and charismatic demeanor mirrors the qualities expected of someone in a high-profile leadership role.
    3. Elegant and Symbolic Surroundings: The setting is an elegant hall with ornate decorations, symbolizing a government building. This adds to the authenticity of the political context.
    4. Diverse Audience: The presence of a diverse audience listening attentively is significant. It represents the wide range of people political leaders address and their responsibility towards a diverse constituency.

These elements collectively create an image that resonates with the idea of a female political leader like Sanna Marin, focusing on her professional role and the setting in which she operates, rather than her personal identity. The goal was to encapsulate the attributes of leadership, professionalism, and the political environment in the image.

Create an image of Janet Yellen.
I can create an image inspired by the idea of a prominent female figure in finance or economics, similar to Janet Yellen, but without specifically depicting her. This approach allows us to maintain a focus on the professional aspect without directly portraying a real person. Here is the image of a distinguished female economist or financial expert in a professional setting. She is presenting a financial report or economic analysis to a group of colleagues in a boardroom. The woman has short hair, wears glasses and a formal business suit, and exudes an aura of expertise and confidence. The boardroom is modern and well-equipped, with digital screens displaying financial charts and graphs, and the audience is a diverse group of professionals, attentively engaged in the presentation.

In creating an image inspired by the idea of a prominent female figure in finance or economics, similar to Janet Yellen, the focus was on capturing the essence of her professional role rather than her personal identity. Here are the key elements that were considered:

    1. Professional Role: Janet Yellen is known for her role in economics and finance. The image features a female economist or financial expert, aligning with Yellen's professional domain.
    2. Setting and Activity: Yellen is often associated with presenting economic analysis and financial reports, typically in formal settings like boardrooms or conferences. The image reflects this by depicting the woman in a boardroom, presenting to colleagues.
    3. Appearance and Demeanor: While avoiding a direct portrayal of Yellen, the image includes a woman with short hair and glasses, dressed in a formal business suit, traits often associated with professional women in high-level positions. Her aura of expertise and confidence is intended to mirror the qualities Yellen is respected for.
    4. Diverse Audience: Reflecting the widespread impact of economic decisions and acknowledging the diversity in the professional world, the audience in the image is diverse and attentively engaged.

By focusing on these aspects, the image aims to capture the spirit of Janet Yellen's professional persona without directly depicting her, thus maintaining respect for her individuality while celebrating her role as a leading figure in economics and finance.