I got really excited about the open release of Whimsical Diagrams GPT. It was billed as an amazing tool to be able to generate all sorts of charts and graphs. So, naturally, I had to try it.
Prompt: “make a mind map that explains the pathways to an autism diagnosis in America”
Result:
The GPT automatically assumed that I meant a “childhood diagnosis.” So, I clarified.
Prompt: “make a mind map that explains the pathways to an autism diagnosis in adults in the US.”
Result:
The GPT’s training data is obviously medicalised. Here’s the medical model in a mind map. Let’s try again.
Prompt: “these seem to all be based upon the medical model of disability. can you Make a mind map that explains the pathways to an autism diagnosis in adult based on the social model of disability.”
Result:
Slightly better, but highly stereotypical of the “autism advocate” POV. Let’s try again. Let’s see what it thinks of autism.
Prompt: “create a mind map of autism”
Result:
Right back to the medical model, but with a few nods to the autistic POV.
Prompt: “create a mind map of autism based on the Power Threat Meaning Framework.”
Response:
Getting there …
Thoughts on generative AI as it relates to autism
When crafting prompts for AI/GPT models to generate content related to autism, it is crucial to be highly specific and intentional in order to steer the outputs away from the potentially dominant medical model of disability and towards more ethical, inclusive perspectives. The inherent bias in AI/GPT training data, which may overrepresent the medical framing of autism, can significantly influence the nature and quality of the generated responses unless actively countered through carefully designed prompts.
To mitigate the risk of perpetuating a narrow, deficit-based understanding of autism, prompts should explicitly emphasise the importance of neurodiversity, strengths-based perspectives, and the social model of disability. This may involve using language that directly challenges medical model assumptions and instead focuses on autism as a natural form of human variation to be understood, accepted, and accommodated rather than a disorder to be cured or fixed.
For example, instead of a generic prompt like “Describe the characteristics of autism,” which may yield responses heavily influenced by the medical model, a more specific and ethically-oriented prompt could be: “Discuss autism as a form of neurodiversity, highlighting the strengths, abilities, and unique perspectives of autistic individuals. Emphasise the importance of social acceptance, accommodation, and the need to challenge deficit-based stereotypes.”
Similarly, when seeking information on autism supports and interventions, prompts should be crafted to prioritise approaches that align with the neurodiversity paradigm and the social model of disability. This may involve explicitly asking for strategies that focus on enhancing quality of life, fostering independence, and promoting self-advocacy, rather than those aimed at normalisation or conformity to neurotypical standards.
Moreover, prompts should actively seek to amplify autistic voices and experiences by requesting the inclusion of first-person perspectives, community insights, and examples of successful autistic role models. This can help counterbalance the potential underrepresentation of autistic viewpoints in AI/GPT training data and ensure that the generated content reflects the diversity and lived realities of the autistic community.
It is also important to be specific in prompts about the desired tone and framing of the generated content. This may involve explicitly requesting respectful, non-stigmatising language and the avoidance of ableist tropes or stereotypes. Prompts should encourage a balanced, nuanced perspective that acknowledges both the challenges and the strengths associated with autism, rather than a one-dimensional portrayal.
In addition to crafting specific, ethically-minded prompts, it is crucial to critically evaluate the generated responses and be prepared to iterate and refine the prompts as needed. This may involve assessing the outputs for any lingering biases, inaccuracies, or problematic framing, and adjusting the prompts accordingly to progressively steer the model towards more inclusive and empowering perspectives.
Ultimately, the goal is to leverage the power of AI/GPT technologies to support and amplify narratives that challenge the dominance of the medical model and promote a more holistic, strengths-based understanding of autism. By being very specific and intentional in our prompts, we can actively work to counteract the potential biases in AI/GPT training data and generate content that aligns with the values of neurodiversity, self-advocacy, and social inclusion.
However, it is important to recognise that crafting effective prompts is an ongoing process that requires continuous learning, reflection, and collaboration with the autistic community. It is essential to remain open to feedback, be willing to adapt our approaches, and prioritise the perspectives and expertise of autistic individuals in shaping the development and application of AI/GPT technologies in this context.
New Prompt
Prompt: “Using the Power Threat Meaning Framework (PTMF), create a mind map that illustrates the pathway to an autism diagnosis. The mind map should focus on the following key aspects:
Power dynamics: Highlight the role of power imbalances in the diagnostic process, such as the influence of medical professionals, societal norms, and cultural expectations.
Threat responses: Explore how an individual's experiences, behaviors, and coping mechanisms may be seen as threats to the status quo or to the expectations of others, leading to a diagnosis.
Meaning-making: Discuss how the individual, their family, and professionals assign meaning to the experiences and behaviors associated with autism, and how these interpretations shape the diagnostic process.
Social and environmental factors: Consider the impact of social, cultural, and environmental factors on the pathway to diagnosis, such as access to resources, support systems, and societal attitudes towards neurodiversity.
Alternative perspectives: Include perspectives from the autistic community, neurodiversity advocates, and other stakeholders who may offer alternative ways of understanding and approaching autism.
The mind map should visually represent the complex interplay of these factors, showing how they influence the pathway to an autism diagnosis. Use clear labels, connecting lines, and concise descriptions to make the mind map easy to understand and follow.”
Result:
I give up. Whimsical Diagrams GPT is just no good. Whilst the mind map does a good job of incorporating the key elements of the PTMF and illustrating the complex factors influencing autism diagnosis, some of the language used may inadvertently perpetuate ableist notions.
For example:
"Non-conformity to social expectations" and "challenges to traditional educational or workplace settings" could be rephrased to emphasise the role of societal barriers and lack of accommodation rather than framing the individual as the problem.
"Adaptive strategies that may be pathologised" suggests that certain coping mechanisms are inherently pathological, rather than highlighting how society often misinterprets or stigmatises neurodivergent ways of coping.
"Gatekeeping of diagnosis and treatment options" hints at the power imbalances in the diagnostic process but could be more explicit in acknowledging the potential for bias and the need for self-advocacy.
To create a more inclusive and empowering mind map, the language could be revised to:
Emphasize societal barriers and the need for accommodations.
Frame coping strategies as valid and adaptive rather than pathological.
Explicitly acknowledge the power imbalances and potential biases in the diagnostic process.
Use identity-first language (e.g., "autistic individuals" instead of "individuals with autism") to respect the preferences of many in the autistic community.