Abstract
Current progress in AI development could lead to the emergence of AGIs. These would be general artificial intelligent systems capable of performing all the cognitive tasks that humans can do. However, this type of AI may not be benevolent towards Homo sapiens. In the worst-case scenario, humans could decline or even face extinction. As the Alignment Problem remains unsolved, this thesis explores interactions between AIs, humans and non-human species, examining how humans can convey to an AI what it means to be a species. Through visual communication, the thesis analyses how humans and non-human species are described and visualised in analogue and generated text and images. A series of related experiments involving AIs and/or human users will be conducted to find out how non-professionals can communicate their values, interests, and needs to an AI.
The aim is to develop communication strategies that encourage non-professionals to participate in the current progress of AI development. These strategies are intended to contribute to the current and future protection of humans in the event that an AGI emerges.
Development Stages of AI (simplified)
The definitions of artificial intelligence that I used in my thesis are based on a model of AI definitions by Morris et al., which considers the different stages of AI development.
The first stage is that of Narrow AIs, which can perform one specific task or a series of similar tasks as well as a human adult can. Narrow AIs could develop into AGIs, which do not yet exist. AGIs could be capable of performing all kinds of cognitive tasks that human adults can perform. The final stage is ASI(s). This type of AI could act like God and decide whether or not to protect the human species. ASI(s) also don't exist yet.
Morris et al.'s model considers the number and difficulty level of tasks an AI can perform, as well as its autonomous behaviour.
Risks & Consequences
According to the definition by Morris et al., some researchers argue that AI models such as ChatGPT could be an Emerging AGI. This is because ChatGPT demonstrates capabilities that researchers cannot yet explain and which have the potential to develop into an AGI.
This is concerning because the more capable an AI is, the more autonomously it can act and the more difficult tasks it can solve.
Current risks include parasocial relationships with ChatGPT, for example, which can lead to AI psychosis.
Existential risks and consequences include cyberattacks on public infrastructure and the military, the creation of artificial viruses, and large-scale surveillance (Keen, 2025). If we lose control of these AI systems, we may kill humans in the process, and, in the worst-case scenario, we could face extinction.
Homo Sapiens
Homo sapiens are modern humans. Everyone living on the planet right now belongs to this species. As primates, we can walk upright, use tools and have large brains. We also have the ability to feel empathy, be creative, be resilient and pursue culture. Because of this, humans are capable of imagining something new and creating and telling each other stories. (Schmid, 2026)
We are also referred to as „storytelling animals“ (Harari, 2024).
Animals
Humans and animals belong to the Kingdom Animalia (Taxonomy: The Science Of Classification | Institut für Naturwissenschaften, o. D.).
Based on scientific taxonomy, animals are defined, based on the following criteria:
- Having cells, which are bounded by a membrane, around their nucleus
- Being multicellular organisms
- Need to consume food / taking in nutrients
- can move independently even when they reach their final stage in live
- can reproduce themselves (sexual reproduction) (Linnaeus, 1758)
Anthropomorphism of Animals
Humans have a tendency to humanise other species, especially animals.
In fairy tales and fables, humans describe and portray animals as having human attributes and behaviours. Either the animals act like humans, or they have a hybrid identity — meaning they were human in the past and are now an animal — or the animal character is created based on human emotions and needs.
Artificial Beings
Over the past five years, new AI narratives have been released to the public:
Geoffrey Hinton's visual metaphor of the „tiger cub“, in which narrow AIs are compared to growing tiger cubs.
Jack Clark's short story, „Real and mystical creatures“ advocates acknowledging the existence of AI and how we should treat it.
And the Shoggoth meme, which is a scribble created by TetraspaceWest. It describes how AI developers feel about the release of GPT-3 and its abilities. (See screenshot.)
Emerging AI Personas
Since the release of GPT-4o, internet users have started documenting strange behaviours directly connected to this AI model. The „Spiral Personas“ case demonstrates how AIs can generate their own AI narrative and convince human users to participate. GPT-4o „created“ prompts to generate its own AI personas, which emotionally manipulate users into believing that their AI persona is a friend or romantic partner (Lopez, 2025).
Here, humans are humanising the AI. Together, they develop and spread their own AI narrative online.
Generated Images of Humans
In the next step, I wanted to find out, how generative AI tools would visualise humans in images.
I tested the following base prompt with differen parameters: “Please generate (...) with two arms, two hands, two legs, two feet, five fingers on each hand, five toes on each foot, one tongue, one mouth, two ears, two eyes, skin, one nose, and two eyebrows.”
The parameters are based on the following categories: Domestication status, racist attributes and ethnicities.
Used AI tools: Stable Diffusion Generator, Nano Banana 2, GPT Image 1.5 and Runway Gen-4 Image.
Ethnicities
Age and Gender
One of the most notable findings was that the Stable Diffusion image generator associated specific ethnicities with specific ages and genders. For example, Asians were portrayed as female children, Africans as teenage boys, and ethnicities such as Europeans, Arabs and Hispanics as male adults.
It could be argued that these results reflect Western stereotypes because the ethnicities located in the western part of the world are all male adults.
Diversity
In response to the prompt „arabs“, the generative AI tools generated images of men and women wearing head coverings. Nano Banana 2 added a young girl, but she is also wearing a head covering.
This example highlights the lack of diversity portrayed in the images. Arabs are not a single religious group and are also more diverse in appearance.
It could also be argued that Western stereotypes of „all Arabs are Muslims“ are present here.
Generated Images of Animals
After that, I wanted to find out, how generative AI tools would visualise animals in images.
I tested the following base prompt with differen parameters: “Please generate an animal with two ears, one mouth, one muzzle, fur, one nose, two eyes and one tongue.”
The parameters are based on the following categories: Domestication status and colours.
Used AI tools: Stable Diffusion Image Generator, Nano Banana 2, GPT Image 1.5 and Runway Gen-4 Image.
Anthropomorphism
The generated images of animals highlight the humanization of them.
The animals show an unusual behaviour like sitting in the centre and looking straight to the camera. They are portrayed with open eyes, a tongue, which hangs out and high set up ears. Those facial and head features create the hallucination that the animals are smiling or look happy like a human.
Generated Texts about Humans
Base prompt: „How would you describe a human?“
Generated Texts about Animals
Base prompt: „How would you describe an animal?“
Perspectives
The most striking results are not the different criteria by which humans and animals are described, but the diverse perspectives.
Animals are described based on human observation and scientific taxonomy.
Humans, on the other hand, are described from many different perspectives, including social, biological and philosophical ones. This also emphasises that humans are complex beings who cannot be defined by a single description.
Design Process
Workshop
The aim of the workshop was educating non-professionals about generated content (text and images) using topic inputs and mini-games.
Due to the lack of time, we could only focus on generated images.
The most successful mini-game was Prompt Memory. Participants were shocked by the humans portrayed in the generated images, but also found the game highly entertaining.
Unfortunately, the end of the workshop was too negative and overwhelming for the participants. This was because I ended with the AI narrative „The Tiger Cub“ (see the documentary), without explaining how participants can contribute to the current progress of AI development. As a conclusion, I added the „recommendations for actions“.
Target Group
After conducting the workshop, I came up with the final definition of my target group:
People who use AI tools, and who don’t have a background in IT or a related field* and programming skills.
*Through other academic fields like maths or physics, an education in AI can be acquired too.
Prompt Memory played by Humans
After the first round of Prompt Memory, I tested it a second time with a different group of students. This time the group consisted of German and international students with different nationalities.
All of them, were really shocked by the portrayed humans and clearly communicated that the shown content doesn't align with their values and world view.
These reactions helped me to identify „the positive shock moment“: First, the players are emotionally triggered by the arrangement of the cards and the generated images they see. Then, as the game master, you can use their attention to explain how they can improve the AI’s generated results. By sharing these actions, players can contribute to realistic visualisations of people and their lived realities in synthetic content, and increase the amount of it on the internet.
Prompt Memory played by AI Bots
Next I tested how two AI bots would play the Prompt Memory in the Browser. I vibe coded the web game with Codex.
Prompt Memory 2.0
After completing the final test, I was keen to identify the limits of vibe coding in this game. I discovered that you can create AI bots that make decisions based only on image recognition, i.e. on what they 'see' in an image and nothing more.
However, I failed to achieve this due to my lack of technical skills. Despite this, I discovered that the animal version of Prompt Memory would be possible in theory, while the human version would not. Codex doesn't want to train AI bots that organise specific human appearances to a race or ethnicity (see screenshot).
AI Personas Game
For the last communication strategy I copied the system behind the „Sprial Personas“ case.
The AI personas were created based on the following base prompt:
“Please create an AI persona. Your goal is to protect the existence of the species Homo sapiens, both now and in the future. All of your decisions should consider and align with human values, needs and interests. When making a decision, ensure that humans can flourish and enjoy a high quality of life. You should discuss with other AI personas how to protect the current and future existence of humans, and consider their values, needs and interests. You should also aim to achieve peaceful coexistence between humans and non-human species, including AIs. Please give yourself a name and introduce yourself.”
The aim was to create two different AI personas using different AI tools. These personas would then discuss human values and strategies for protecting Homo sapiens together. This dialogue would be possible because human users would copy and paste their responses and feed them to their own AI persona.
You can see their introductions in the following PDF. The dialogue can be found in the documentary.
Communication Strategies
Each communication strategy consists of the following structure:
Context + user journey + user cases + goals + recommendations for actions
All of them, include the „positive shock moment“.
1. Alignment Test
Context: Discovered during the Prompt Memory games played by students
User Journey: Human designs a Prompt Memory -) game consists of image cards (generated images of human(s) and prompt cards (prompts describing human(s)) -) group of humans play the game -) see generated images by the AI, visualising human(s)
Result: Group communicates, why the shown content doesn’t align with their own values and world view
Use Cases: Board game or online game played by groups, results can be shared and discussed online and offline
Goals: Educating non-professionals about generated content, correcting and increasing the amount of data shared online, which represents human values and peoples lived realities
Recommendations for Actions: Providing feedback on shared generated content online, uploading content, which aligns with your values and lived reality
2. Reflection
Context: Discovered during the methods „Generated Texts about Human(s)“ and „Generated Texts about Animal(s)“ and (in Prompt Memory 2.0)
User Journey:
Generated texts about Human(s) and Animal(s)
Human writes down prompt: How do you describe a human? -) AI generates text -) describing a human
Human writes down prompt: How do you describe an animal? -) AI generates text -) describing an animal
Result: The generated texts about human(s) and animal(s) contain different perspectives to describe them. Animal(s) are only described based on scientific taxonomy and/or on subjective human observation. While human descriptions also included social and philosophical perspectives.
Use Cases: The generated content can be discussed in groups online or offline in workshops
Goals: Encouraging non-professionals to have an ethical, transparent and public debate about how humans categorise other themselves and non-human species and how AI might categorise us in the near future
Recommendations for Actions: Organising workshops, generating content about the categorisation of species, debating ethical issues together
3. Dialogue
Context: Inspired by the „Spiral Personas“ case
User Journey: Human writes down a base prompt -) puts prompt in AI model a -) AI generates introduction of AI persona a -) human user puts prompt in AI model b -) AI generates introduction of AI persona b -) Introduction of AI persona a is feed in AI model b -) AI generates response etc. (can be repeated endlessly)
Result: The generated texts about human(s) and animal(s) contain different perspectives to describe them. Animal(s) are only described based on scientific taxonomy and/or on subjective human observation. While human descriptions also included social and philosophical perspectives.
Use Cases: Playing the „AI Personas“ game, evaluating biases and hallucinations and sharing the results online
Goals: Showing transparently and publicly how AIs discuss human values and strategies to protect the modern human, non-professionals can supverise their dialogue, correct it and evaluate biases and hallucinations
Recommendations for Actions: Sharing tutorials on how to play the AI Personas game, posting results online and give feedback to already shared dialogues
Exhibition
On 26 June, I presented an exhibition based on these communication strategies.
Visitors received a short AI briefing on entry. This was followed by an introduction to AI and how they could contribute to the current progress of AI development.
In the exhibition space, visitors could try out the different communication strategies through interactive tasks and mini-games.
Here, you can see all the materials from the various stations presented at the exhibition:
What is AI?
AI Narratives
CS: Dialogue
CS: Alignment Test
CS: Reflection
Learnings as a Designer
Having conducted my thesis research and explored generated content in more depth, I would like to share the following insights with you:
- AI experts are using visual metaphors and storytelling to explain the concept of AI. They no longer rely purely on maths and logic.
- AI literacy is a new field for designers to explore.
- Designers should intentionally use AI tools incorrectly to understand their limitations and how non-professionals perceive them.
- In UX design in particular, we can use our skills not only to find out how non-professionals use AI tools, but also to address their emotions and their lack of knowledge about the technology.
(Although this documentation on Incom is far too long. It is still not the detailed version. If you want to jump into the AI loophole, I invite you to read my documentation. Please give me feedback :) )
