Lost in Translation: When LLMs Miss the Lived Experience of Autism
As the year draws to a close, I wanted to take a moment to reflect on what’s been published in autism research throughout 2024. Staying informed is a critical part of my work, both as an autistic person and as someone who seeks to critique the systems that continue to frame autism through narrow, harmful lenses. To this end, I turned to ConsensusGPT, curious about its latest developments and whether its responses would reflect any growth in how autism research is represented. My query was straightforward: “Provide a summary and list of the top 10 papers on autism in 2024.”
The response I received was a stark reminder of how deeply embedded the medical model of disability remains in mainstream discussions of autism. The list was dominated by studies focused on pathology, interventions, and technological “solutions” that aim to normalise autistic people rather than celebrate our differences. Despite some gestures toward social factors, the overarching narrative was one that frames autism as a problem to be detected, managed, or cured. As someone who wants to stay current with research, it’s disheartening—and frankly, triggering—to see my lived experience and the experiences of my community so consistently devalued and erased.
What does it mean when large language models like ConsensusGPT produce lists so deeply influenced by the medical model of autism? For one, it underscores how biased the research landscape remains. These systems are trained on existing knowledge, and if that knowledge is overwhelmingly shaped by the perspectives of clinicians, researchers, and policymakers who see autism as a deficit, then that is the narrative they will reproduce. But this isn’t just an academic issue. These biased outputs shape public perception. Someone searching for “top autism research” might walk away believing that the most pressing issues in our community are early detection, behaviour modification, and compliance training—not systemic oppression, lack of accommodation, or the need for meaningful inclusion.
The deeper question is: what does this do to the conversation about autism? It reinforces the idea that autistic lives are problems to be solved, rather than individuals with unique perspectives, strengths, and needs. It perpetuates a eugenic thread in Western discourse, where the ultimate goal seems to be a society free from autism, rather than one enriched by its diversity.
For me, this experience reiterates the sense that there is “no place for autism” in the frameworks that dominate Western research and policy. The absence of studies reflecting autistic voices, focusing on our lived experiences, or challenging oppressive systems is not just a gap—it’s an erasure. It’s exhausting to stay informed when the systems we turn to for knowledge so often dehumanise us. Yet, it’s precisely this erasure that fuels my commitment to critiquing these frameworks and advocating for a future where autistic voices lead the conversation.
What is ConsensusGPT?
ConsensusGPT is an LLM-powered “research assistant” that integrates ChatGPT's language generation capabilities with Consensus’s database of over 200 million academic papers. According to its creators, it aims to streamline the research process by allowing users to input natural language queries, retrieving and synthesising information into summaries or formatted outputs like APA citations. Consensus claims to cover diverse scientific disciplines, offering quality indicators and advanced search capabilities, including tools to assess citation relevance and journal standards.
Again, according to Consensus, unlike Google Scholar or Academia.edu, ConsensusGPT prioritises conversational interaction. Users can ask detailed or broad queries without needing the exact search terms, and the system generates summaries rather than simply linking to papers. This synthesis, however, is both a strength and a limitation. Whilst it offers convenience, the responses are mediated by ChatGPT, which may introduce biases or frame the information according to dominant narratives within its training data or the Consensus database. For instance, when searching for autism research, the results will reflect a medical model perspective, marginalising lived experiences and alternative frameworks like the social model.
In contrast, Google Scholar provides raw search results, allowing users to directly access abstracts, full papers, and citation links. Its strength lies in its objectivity—what you see is what you get—but it lacks the interpretive layer of ConsensusGPT, leaving users to synthesise information themselves.
Academia.edu, on the other hand, focuses on networking among scholars. It allows access to papers uploaded by researchers themselves and facilitates interaction through follow features and messaging. Whilst its scope is narrower, it is more community-driven, reflecting individual contributions rather than comprehensive databases.
ConsensusGPT stands out for its user-friendliness and summarisation capabilities, but for those seeking unfiltered information or niche academic networking, Google Scholar and Academia.edu remain indispensable alternatives.
Back to My ‘Research’
To explore “the latest autism research,” I logged into the GPT Marketplace and loaded ConsensusGPT. The process of setting up and refining my query required a bit of trial and error. My aim was straightforward: to generate a list of the top autism studies published in 2024.
My initial query yielded results spanning decades, with studies dating back to the 1990s. Whilst this demonstrated ConsensusGPT’s ability to access a broad range of research, it wasn’t what I needed. I had to prompt the system further, clarifying that I wanted only the most recent publications from 2024. After a bit of tweaking, I managed to refine the query and received a curated list of studies, each accompanied by summaries and references formatted for academic use (see below).
This iterative process highlighted both the strengths and limitations of ConsensusGPT. It’s a tool that offers powerful access to academic content but requires precise prompting to deliver relevant results. Unlike platforms like Google Scholar, where users filter results manually, or Academia.edu, where research is often shared directly by scholars, ConsensusGPT provides a conversational interface that synthesises information on demand. However, this synthesis depends heavily on the clarity of the initial query and how effectively the tool interprets it, making the setup process an important step in achieving meaningful results.
The World According to ConsensusGPT
Here’s what the query, “provide a summary and list of the top 10 papers on autism in 2024.”
“Here are the top 10 papers on autism published in 2024, with summaries and details:
Model for autism disorder detection using deep learning
Authors: Anshu Sharma, Poonam Tanwar
Focus: Utilizes deep learning to detect autism spectrum disorder (ASD) with high accuracy (93.41%).
Published in: IAES International Journal of Artificial Intelligence
Using Measures of Vowel Space for Autistic Traits Characterization
Authors: Chin-Po Chen et al.
Focus: Analyzes articulation-related acoustics to characterize autistic traits through speech.
Published in: IEEE/ACM Transactions on Audio, Speech, and Language Processing
A computational intelligent analysis of autism spectrum disorder using machine learning techniques
Authors: Murali Anand Mareeswaran, Kanchana Selvarajan
Focus: Employs machine learning for early ASD diagnosis, achieving 96% classification accuracy.
Published in: IAES International Journal of Artificial Intelligence
Authors: Emma R. Plourde et al.
Focus: Explores medication and service use among youths with ASD, particularly those with co-occurring mental health conditions.
Published in: American Journal on Intellectual and Developmental Disabilities
Impact of the Great Recession on Adults With Autism and Their Mothers
Authors: Emily J. Hickey et al.
Focus: Investigates the economic downturn's impact on autistic adults and their mothers, revealing resilience but also specific vulnerabilities.
Published in: American Journal on Intellectual and Developmental Disabilities
Authors: M. Alabdulaziz
Focus: Shows video modeling significantly improves math problem-solving skills in ASD students.
Published in: International Journal of Instruction
Eating skills in children with autism via video modeling
Authors: Nurul Khairani Ismail et al.
Focus: Evaluates video modeling for teaching eating skills to ASD children, showing positive results.
Published in: International Journal of Public Health Science
Authors: Derek Nord et al.
Focus: Assesses how health insurance impacts employment among caregivers of children with ASD.
Published in: American Journal on Intellectual and Developmental Disabilities
Authors: Smily Jesu Priya Victor Paulra et al.
Focus: Introduces a training program to improve emotion recognition in ASD children, with validated effectiveness.
Published in: Journal of Associated Medical Sciences
Authors: Brian Barger et al.
Focus: Highlights disparities in developmental screening and diagnosis rates among racial/ethnic groups.
Published in: Infants & Young Children
These papers reflect recent advancements in autism research, emphasizing innovative technologies, educational interventions, and the impact of socioeconomic factors.”
I asked for an APA formatted reference list.
“Here is the APA reference list for the top 10 autism papers published in 2024:
Sharma, A., & Tanwar, P. (2024). Model for autism disorder detection using deep learning. IAES International Journal of Artificial Intelligence (IJ-AI). https://consensus.app/papers/model-autism-disorder-detection-using-learning-sharma/aae78626d61251d6b3f53f687f4eafbc/?utm_source=chatgpt
Chen, C. P., Pan, H. H., Gau, S. S., & Lee, C. C. (2024). Using measures of vowel space for autistic traits characterization. IEEE/ACM Transactions on Audio, Speech, and Language Processing, 32, 591–607. https://consensus.app/papers/using-measures-vowel-space-autistic-traits-chen/76742c9fff0655a7832802210339a38c/?utm_source=chatgpt
Mareeswaran, M. A., & Selvarajan, K. (2024). A computational intelligent analysis of autism spectrum disorder using machine learning techniques. IAES International Journal of Artificial Intelligence (IJ-AI). https://consensus.app/papers/intelligent-analysis-autism-spectrum-disorder-using-mareeswaran/5ca5b216e603518dbd866c437f1b37fb/?utm_source=chatgpt
Plourde, E. R., Ali, M. M., & West, K. D. (2024). Psychotropic medication and psychosocial service use among transition age youth with autism spectrum disorder. American Journal on Intellectual and Developmental Disabilities, 129(1), 1–9. https://consensus.app/papers/psychotropic-medication-psychosocial-service-among-plourde/10fbac2524365591831742b25a14e959/?utm_source=chatgpt
Hickey, E. J., DaWalt, L., Bolt, D., Hong, J., Song, J., Taylor, J. L., & Mailick, M. R. (2024). Impact of the Great Recession on adults with autism and their mothers. American Journal on Intellectual and Developmental Disabilities, 129(1), 26–40. https://consensus.app/papers/impact-great-recession-adults-with-autism-their-mothers-hickey/5153fc7542845af6bb3a5c9fb286903e/?utm_source=chatgpt
Alabdulaziz, M. (2024). The effects of using video-based instruction in solving fraction computations of students with autism spectrum disorder. International Journal of Instruction. https://consensus.app/papers/effects-using-instruction-solving-fraction-computations-alabdulaziz/5d457501a87554a7b7dd2ad06e1daab0/?utm_source=chatgpt
Ismail, N. K., Hosshan, H., Abd Aziz, N., Masnan, A. H., Che Rozubi, N. S., & Md Zalli, M. M. (2024). Eating skills in children with autism via video modeling: An evaluation of face and content validation. International Journal of Public Health Science (IJPHS). https://consensus.app/papers/eating-skills-children-autism-video-modeling-evaluation-ismail/ac3a154717135477858e47cae0c4fb57/?utm_source=chatgpt
Nord, D., Hamre, K., & Andresen, J. (2024). The effects of health insurance coverage on workforce engagement of family caregivers of children with intellectual disability and autism. American Journal on Intellectual and Developmental Disabilities, 129(1), 10–25. https://consensus.app/papers/effects-health-insurance-coverage-workforce-engagement-nord/b697c49f69c7558fbfe8bf57777d275a/?utm_source=chatgpt
Paulra, S. J. P. V., Chinchai, S., Munkhetvit, P., & Sriphetcharawut, S. (2024). The development and content validity of the emotional recognition memory training program (ERMTP) for children with autism spectrum disorder: A trial phase. Journal of Associated Medical Sciences. https://consensus.app/papers/development-content-validity-recognition-memory-paulra/689d6610278c5664a4fa18ac66427d9c/?utm_source=chatgpt
Barger, B., Salmon, A., & Chödrön, G. (2024). Correlations between state-level monitoring and screening rates and early identified ASD/DD across racial and ethnic groups. Infants & Young Children, 37, 64–81. https://consensus.app/papers/correlations-between-statelevel-monitoring-screening-barger/757bc257536c5b118ebf1fc3160d8cb5/?utm_source=chatgpt”
The Analysis
From my perspective, using Critical Theory and the Power Threat Meaning Framework (PTMF), the selection of these studies by ConsensusGPT as representative of “a diverse range of topics in autism research” reveals much about the systems and structures that shape dominant narratives in autism research as well as the data underpinning LLMs. Let’s break this down:
1. Technological Determinism and the Commodification of Autism
The emphasis on technological solutions, such as deep learning and machine learning (Sharma & Tanwar; Mareeswaran & Selvarajan), highlights the capitalist imperative to “solve” autism through profitable technologies. These studies align with a framework that sees autism as a deficit to be detected and managed, reinforcing the medical model of disability. The high classification accuracies (e.g., 93-96%) reflect an obsession with quantifiable “progress,” but at what cost? These tools likely reinforce the power of diagnostic systems whilst ignoring the lived experiences and agency of autistic individuals.
2. Speech as a Site of Control
The focus on speech acoustics (Chen et al.) positions autistic traits within a framework of abnormality to be objectively measured. This aligns with the historical pathologisation of autistic communication styles, ignoring how power structures create stigma around non-normative speech. Instead of celebrating autistic language as diverse and meaningful, the study implicitly centres the standards of “normal” articulation within the neuro-majority, threatening the identity of those who deviate.
3. Neoliberal Solutions in Education
The studies on video-based interventions (Alabdulaziz; Ismail et al.) highlight neoliberal, individualised approaches to education. These tools are framed as empowering autistic learners, but in reality, they often serve as cost-cutting measures in underfunded systems, offloading the responsibility for meaningful learning onto the child and their family. Video modelling assumes a one-size-fits-all solution without addressing structural barriers, such as inaccessible curricula or the under-resourcing of inclusive education programs.
4. Socioeconomic Threats and Blame Shifting
Studies on economic challenges (Hickey et al.; Nord et al.) shift the focus from systemic inequalities to individual vulnerabilities. For example, examining how caregivers engage with the workforce underlines how autism is treated as a “burden” to be mitigated, rather than addressing the broader socio-economic systems that devalue care work and marginalise disabled families. Similarly, the framing of autistic adults and their mothers as "resilient but vulnerable" during the Great Recession ignores how capitalist systems compound the oppression of disabled communities.
5. Pathologising Emotion and Behaviour
The emotional recognition memory training program (Paulra et al.) reflects the ongoing effort to “fix” autistic behaviour to align with neurotypical norms. From a PTMF perspective, this reinforces the idea that autistic traits are threats to be managed rather than adaptive responses to an oppressive social world. These programs often invalidate the emotional expressions of autistic individuals, imposing meaning that fits the dominant framework.
6. Structural Inequities Framed as Individual Issues
The study by Barger et al. on racial and ethnic disparities in developmental screening gestures toward a systemic issue but fails to interrogate the power dynamics that perpetuate such disparities. By focusing on “correlations” rather than structural causes, it depoliticizes the issue, framing inequity as a data problem rather than a systemic injustice tied to racism and economic oppression.
7. The Narrow Lens of Research Priorities
Across all these studies, there is a glaring absence of critical perspectives that centre autistic voices or explore the social model of disability. The medicalisation of autism dominates, reinforcing the idea that autistic traits are problems to be managed or eradicated. This selection prioritises research that fits within the neoliberal, capitalist framework, where autism is framed as a threat requiring technological, behavioural, or pharmacological solutions. The lack of studies addressing structural change or community-led approaches highlights how power shapes research agendas.
Conclusion
ConsensusGPT’s selection reflects the values of a system that prioritises research aligned with marketable solutions and the medical model of disability. By emphasising technological fixes, individual interventions, and measurable outcomes, it perpetuates the view that autism is a deficit to be managed rather than a difference to be understood. From a PTMF perspective, the real threats lie in the socio-economic structures and power imbalances that frame autistic lives as burdensome, rather than recognising the systemic oppression that creates these challenges.
When I tried to clarify what I wanted, “… these are all medical-model studies. How about studies from 2024 that highlight lived experiences of autistics, have autistic researchers, and are framed in a model other than the medical model,” it replied, “Unfortunately, the retrieved studies from 2024 mostly align with the medical model, focusing on clinical, diagnostic, and technical approaches rather than emphasizing lived experiences of autistic individuals or being authored by autistic researchers with alternative frameworks like the social model or neurodiversity paradigm…” Ouch!
Final thoughts …
After exploring ConsensusGPT for current “autism research,” it’s clear that whilst the tool shows promise, it falls short of delivering meaningful results for someone like me, an autistic person seeking research that reflects lived experiences. Despite its powerful ability to synthesise vast amounts of academic content, its reliance on the biases inherent in its data sources limits its usefulness for nuanced, human-centred inquiries.
The process of refining queries and obtaining current results highlights a fundamental issue: ConsensusGPT is excellent at surfacing mainstream narratives but struggles to account for the depth and diversity of perspectives that matter to marginalised groups. For autism research, this means its results often reinforce the medical model, presenting autism as a deficit to be detected, managed, or “corrected.” This framing not only erases the voices of autistic people but perpetuates harmful narratives that have real-world consequences, from policy decisions to societal attitudes.
For someone like me, staying informed about research isn’t just an academic exercise—it’s a way to understand how the world sees us and, often, how it misunderstands us. But when tools like ConsensusGPT default to dehumanising frameworks, they serve to amplify the systemic erasure of autistic perspectives.
ConsensusGPT may be useful for researchers who are already steeped in dominant paradigms, but for those of us looking for critical, inclusive, and meaningful work, it isn’t ready for primetime. Until it can reflect the diversity of thought and lived experience that exists within autism research, it remains an incomplete resource.