Cognitive Differences in Autistic Individuals
Until recently, cognitive processing in autistic individuals tended to confirm deficits in
executive functioning and working memory. Several studies on cognitive deficits in the adult
neurodivergent population, particularly older adults, have found significant impairments in item
and relational memory, executive functioning, Theory of Mind, and attention (Tse et al., 2022)
when compared to neurotypical control samples. Autistic individuals also indicated deficits in
cognitive flexibility, or the ability to perform the same task under changing environments or
rules (Lague et al., 2024).
Strengths
Simultaneously, neurodivergence has positive aspects that are overlooked, even by
autistic people themselves. Despite longer reaction times, older autistic adults performed better
in experiments with greater cognitive load when compared to neurotypical individuals (Tse et al.,
2022). They also have enhanced visual-processing capabilities and higher attention,
characterized by the Monotropic cognitive theory (intense but narrow form of attention in
autistic people) (Lage et al., 2024). Literature also indicates that people with autism experience
less susceptibility to cognitive bias and process information more rationally compared to
neurotypical individuals (Rozenkrantz et al., 2021).
Cognitive Flexibility
Cognitive flexibility is a core domain in Executive Functioning. Hollocks et al. (2023)
defines it as the ability to switch between cognitive processes to produce a context appropriate
behavioral response. The study focused on the following skills: attentional and set shifting,
generativity (how well individuals create spontaneously appropriate novel responses), and
reward sensitivity.
Research into how autism affects cognitive flexibility remains hindered by lack of
adequate measurements of the domain. Furthermore, the authors find that Autism Spectrum
Disorder’s effects on other cognitive domains cloud researchers’ ability to discern between direct
cognitive flexibility deficits and more broad executive functioning issues (Hollocks et al., 2023).
Despite this, in a separate more recent study, the authors confirm that executive functioning
remains a challenge for people with autism (Hollocks et al., 2025).
Attention
Attention is defined by Castra et al. (2023) as a combination of selecting, modulating,
and focusing on stimuli relevant to behavior. The study utilized a combination of tests to
determine attention performance. In the Test of Everyday Attention (TEA), participants are
evaluated on several attention domains found in everyday tasks, including selective attention,
attentional switching, auditory-verbal working memory, and sustained attention. Autistic adults
scored significantly lower on these TEA tests compared to the neurotypical group.
Memory
According to Desaunay et al. (2020), Memory is broken into two separate categories:
short-term memory (STM), or working memory (WM) more recently, and long-term memory
(LTM), divided into the subsystems explicit and implicit memory. The findings indicate
differences between autistic and neurotypical individuals across memory domains. In particular,
studies have shown that autism negatively affects short-term memory, including visual, verbal,
and visuospatial domains. However, fewer difficulties present themselves in long-term memory
compared to short-term memory. Nevertheless, some deficits compared to neurotypical control
groups still presented themselves. Desaunay et al. (2020) found a small effect size for verbal
material, medium effect size for visual material, and no effect size for visuospatial material.
Executive Functioning
Jertberg et al. (2025) authored a study designed to understand the differences in executive
functioning in over 900 autistic adults aged 18-77. They utilized multiple tasks to test inhibition,
cognitive flexibility, working memory, and attentional orienting. Participants had an average or
above average IQ. The authors wanted to correct several shortcomings of prior studies,
particularly sample size and accounting for comorbid conditions like ADHD.
In all four tasks, individuals with autism showed longer reaction times compared to nonautistic individuals (Jertberg et al., 2025), aside from cognitive flexibility. Additionally, autistic
individuals displayed greater efficiency in visual processing. They also exhibit meta-cognitive
differences when compared to neurotypical individuals. This could cause neurodivergent people
to prioritize accuracy over speed. Jertberg et al. (2025) also found the neurodivergent sample had
a latency in coordination and motor tasks.
Despite these findings, the authors conclude that other than reaction time latency in the
performed tasks, there were no explicit differences in any of the participants pertaining to direct
executive function. They point out several reasons for previous studies’ conflicting findings.
Mostly, they state that small sample sizes and self-reporting questionnaires that influenced
previous research. Particularly, autistic individuals statistically suffer from depression disorders
more often than the general population, which could cause them to underestimate their cognitive
abilities (Jertberg et. al., 2025). Another important note found was the fact that a broader
application of the term “Autism Spectrum Disorder” diversified the neurodivergent population
overall when compared to previous studies. Finally, the authors concluded that creating
conducive educational and professional environments, (structure, explicit instructions, and
accommodations for slower processing and response speeds) for neurodivergent people can help.
History of AI Application in Education: Autism Focus
Artificial Intelligence (AI) has a long and rich history in education dating back to the
1950’s and 1960’s (Doroudi, 2022). The current principles of AI were founded by cognitive
scientists who desired to build new models of human knowledge during the cognitive revolution.
Alan Newell and James Moore worked to develop one of the first AI tutoring systems called
Merlin in the 1970’s (Doroudi, 2022). Unfortunately, the system never materialized, and Merlin
was considered a failure by its creators, mostly due to its lack of usability and no impact on the
rest of the field.
However, this didn’t mark the end of educational applications of AI. A competitive and
moderately successful tutoring system surfaced in 1970 by Jaime R. Corbonell. After decades of
work and collaboration, researchers at Carnegie Mellon founded Carnegie Learning Inc.,
specializing in the development of cognitive tutors for algebra and other fields. This is one
example of broad academic progress in education, as several other examples of AI and tutoring
systems have appeared throughout the decades.
In the modern context, previous research has led to the creation of two parallel strands:
One, called AIED (AI in Education) with a focus on information-processing theories and the
other, called ICLS (Internataional Conference of Artificial Intelligence in Education) with a
focus on constructivist and situativist theories (Doroudi, 2022). Although they experience some
moderate cooperation, their approaches remain largely separate. In addition to academic
research, the field of AI in education continues to rapidly expand. Most countries in the world
have incorporated artificial intelligence in their education systems (Zhang, 2025). Practical
applications include personalized learning paths and intelligent marking systems for homework.
To accomplish this, AI systems interfere with cognitive processes by adapting the presentation of
material and cognitive load intensity. They also monitor cognitive trajectories in real-time and
reveal logical breaks of argumentative structures in writing (Zhang, 2025).
Current Issues
Problems Surrounding Autism’s Diagnosis, Research, and Ethical Concerns
Lack of research focus persists as a major problem facing current research into AI applications
for autistic cognitive deficits. The majority of existing research only focuses on social skill
improvements in the autistic population, not cognition. Even in domains studied and focused on
by researchers, such as robots used to improve social communication, have produced mixed
outcomes (Adako et al., 2025). Furthermore, research into AI interventions for autistic adults is
almost nonexistent.
More broadly, conflicting data remains problematic for understanding the cognitive
strengths and deficits of neurodivergence separate from implementation of artificial intelligence.
For example, where one study finds no link between autism and executive functioning, other
studies show a significant difference (Jertberg et al, 2025). Sample size calls some studies into
question as well. Multiple studies concerning cognitive deficits have less than 50 participants
(Jertberg et al., 2025). Autism research also has historically focused on difficulties autistic
individuals’ experience instead of discussing their strengths (Lampinen et al., 2025).
Research in Autism Spectrum Disorder has long-standing ethical issues (N’eman et al.,
2023). First, current research in autism focuses on reducing autistic traits and enhancing masking
capabilities of autistic individuals. Masking is defined as the process in which individuals with
autism conceal their autistic traits in social situations to conform to neurotypical social standards
(Venkatesan & Tolani, 2024). Some researchers go so far as to state that the most optimal
outcome for autistic individuals is a complete elimination of symptoms, despite concerns raised
by autistic advocacy groups. Academic literature concerning intervention strategies, such as
those supporting ABA (Applied Behavioral Analysis) and measurement instruments like the
Social Reciprocity Scale-2 explore intervention of autistic traits that are, in fact, harmless
(N’eman et al., 2023).
The issues expand beyond targeted intervention research. In one study, Macari et al.
wanted to understand differences in emotional reactivity between neurotypical and autistic
toddlers by exposing them to frightening masks, objects, and strangers (N’eman et al., 2023).
Despite the outcry by autism advocates, the authors went through with the study (Macari et al.,
2021).
Research in autism also has issues with underrepresented groups, such as people of color
and individuals from lower socioeconomic classes (Maye et al., 2021). National autism databases
overrepresent white individuals from middle- to high- income brackets. This bias has significant
effects beyond academia. Diagnosis and the prevalence of autism differ across race and ethnicity,
even if the study accounts for socioeconomic status differences among population samples. It
also cascades into public health, with poorer outcomes for physical and mental health regardless
of an individual’s stage of life (Maye et al., 2021). Research often requires reliable transportation
and taking time off work, as well as high quality internet, which many families may not have.
Non-English speakers are also frequently excluded from research studies. Female populations are
consistently underrepresented as well (D’Mello, 2022).
Finally, autism research faces a severe challenge purely deriving from how broad ASD as
a diagnosis is (Waterhouse, 2021). As the diagnostic criteria of ASD has expanded, so has the
heterogeneity of individuals across various domains, including language, intelligence, comorbid
diagnoses, and severity (Rabot et al, 2023). Meta-analyses indicate that effect sizes decreased by
up to 80% in studies comparing neurodivergent and neurotypical individuals. This could be due
to the widening of diagnostic criteria for ASD (Rabot et al., 2023). Additionally, symptoms of
autism, such as communication problems, depression, and even repetitive behaviors, overlap
with other psychiatric and neurological conditions (Bertelli et al., 2024).
AI Research for Autistic Interventions
With respect to autism, current research remains sparse and implications of artificial
intelligence specifically concerning the autistic population is even less common (Kotsi et al.,
2025). Finding data on postsecondary applications proves difficult, as most research on autism
focuses on children instead of adults (Johnson et al., 2024). However, challenges with cognitive
processes occur throughout the lifespan, therefore some of the initial literature on AI
interventions can apply to autistic individuals regardless of age. Utilizing AI to assist autistic
students, in an analysis of 13 empirical studies, targeted social skills, emotional recognition, and
social communication. Only three of the studies targeted cognitive challenges (Kotsi et al.,
2025). In fact, Kotsi et al. (2025) highlight that research on improving cognition in autism
students lags, and most existing research they analyzed concerned only social skills and
emotional domains. This represents a significant gap in the application of AI technology in
autistic cognition.
Challenges of Using Artificial Intelligence in Autistic Educational Interventions
As discussed previously, studies on AI applications in students with autism are not plentiful,
especially in the adult population. However, some studies do exist but pertain mostly to children.
Most interventions that use AI focus on social interaction and eye contact (Yang et al., 2024).
General studies performed using special education students have been published. Large
Language Models (LLMs) and AI show significant promise in offering personalized learning and
support (Voultsiou & Moussiades, 2025). AI can also detect patterns in behavior to customize
interventions. Immersive technologies, LLMs, and AI integration provide a true positive impact
on the learning environment for special education students (Voultsiou & Moussiades, 2025).
Despite these benefits, several core issues with AI in education in general (and therefore
extrapolated to educational interventions for autism students specifically) persist. AI has the
potential to over-optimize and in doing so can diminish learners’ internal motivation to explore
material. Additionally, AI tools can optimize the learning process too efficiently, which may lead
the tool to start optimizing for the sake of data instead of knowledge transfer for the student
(Zhang, 2025). AI also struggles with intercultural adaptation, in that most AI programs suffer
from Western centrism, which may compete with non-Western educational traditions. Students
who over-rely on AI tools for writing also show a more diminished capacity for rigor and
identifying counterexamples (Zhang, 2025).
Specifically for autistic students, Kotsi et al. (2025) list several challenges, including
privacy concerns, data security, informed consent, unequal access based on socioeconomic
factors, and others. Training data for AI also has inherent biases (Dhabliya et al., 2025), and
these can affect educational AI tools.
Conclusion
Summary
People who have Autism Spectrum Disorder have an array of cognitive deficits in
addition to social challenges. In large meta-analyses, persistent evidence concludes that these
challenges appear in cognitive flexibility, memory, attention, and executive function. However,
this only tells part of the story, and findings are mixed at best in relation to understanding the
entirety of the disorder. Autistic individuals also display some advantages regarding cognitive
bias and rational information processing. Furthermore, attention deficits could be attributed to
narrower yet more focused attention when compared to neurotypical population samples.
Autistic individuals also suffer more from comorbid psychiatric diagnoses which impact
daily functioning when compared to the general population. Many studies on autistic cognition
do not account for symptoms that might possibly relate to other disorders and not autism itself.
Additionally, some studies state that people with autism perform better on tasks with greater
cognitive load and enhanced visual-processing capabilities. Because most studies on autism
focus on deficits or pathology instead of direct cognition, strengths of neurodivergent individuals
is largely understated or unexplored altogether.
In research, several systemic and critical problems plague advancements in understanding
neurodivergence. Ethical issues stem from both current research practices and academic studies’
objectives more broadly. Current exploration into mitigating targeted behaviors for social
congruence instead of cognition remains a consistent problem, and many of these targeted
behaviors (such as eye contact or hand flapping) are harmless. Research also suffers from smaller
sample sizes, inconsistent assessment instruments, bias against individuals from lower
socioeconomic classes and racially marginalized groups, and lack of participation of the female
population. Additionally, resources primarily focus on autistic individuals that do not have
intellectual disabilities even though this group comprises 38% of the total autistic population
(Bertelli et al., 2024).
In a broader context, the changes in the DSM-V potentially blur the lines of proper
diagnosis (Rabot et al., 2023). This has rippling effects throughout research and intervention
practices. Concurrent diagnoses also remain a persistent problem, and their interactions with the
disorder remain understudied. Misdiagnosis also could occur more frequently as certain
phenotypic autistic traits can also signify other psychiatric conditions.
Personal Point of View
In an effort to communicate effectively, I will intentionally shift from third to first person
for this section, since it requests my direct perspective. I was diagnosed with ASD Level 1 earlier
this year, but throughout my 31 years on this planet, I always knew I differed substantially from
my peers. I also have PTSD due to childhood trauma, but this only explained half of the story. I
took the opportunity to use this project to understand my condition, my cognitive differences,
and the intersection between neurodivergence and artificial intelligence, which remains a passion
I have for myself and other autistic individuals.
Starting this project, I thought it would be easy to find data that matched what I wanted. I
explored the UNT library eagerly searching for answers to my academic and personal questions.
What I found told a completely different story. I instead uncovered an array of morbid truths
concerning the state of the field: limited sample sizes, intentional exclusion of the intellectually
disabled, discrimination, and ethical concerns that made me temporarily walk away from the
research. In my entire academic career, I never experienced the level of disillusionment I have
experienced with this project. I finally see the dark underbellies of academic research I knew
existed but had never experienced face-to-face.
The problems of diagnosis cannot be ignored, either. Researchers struggle to get empirical data
because we have lumped together a huge, diverse population into a spectrum. Academics and
professionals can debate the empiricism of this decision, but it affects research in the field
regardless of its validity. Subsequently, this directly affects autistic people as it rolls downhill
from academia to practice.
Specific data on interventions using artificial intelligence also lack focus on bridging the
understood cognitive deficits in memory, executive function, and other areas autistic people
struggle with. Researchers, parents, and educators seem to have an explicit and painstakingly
misapplied focus on social integration and social skills. While social skills and training are
important for people with autism, a much better use of our time, resources, and effort would be
helping neurodivergent people achieve better methods of cognition.
The fields of psychology, education, and artificial intelligence have incredible potential to help
autistic people in many ways, but the lack of data and focus stunts this potential. We know that
LLMs, AI, and virtual reality (VR) have wide applications for autistic people and individuals
with special needs more broadly (Voultsiou & Moussiades, 2025), but our misapplied
concentration on how to “make autistic people fit in more” degrades our ability to move forward.
In conclusion, while the potential for AI to assist individuals with autism could have
amazing positive consequences for them, our discrimination, failures of research practices, and
misapplied focus hinders progress in bettering neurodivergent people’s lives. We need a holistic
interdisciplinary approach that addresses ethical, diagnostic, and research issues to increase the
effectiveness of interventions. I hope that by moving forward with a PhD program, I will be able
to contribute to the field by bringing both lived and academic experience. This project didn’t
make me want to stop but instead invigorated me to continue my journey. For that, I am grateful.
References
Adako, O. P., Adeusi, O. C., & Alaba, P. A. (2025). Ai in autism education: A review of
collaborative and longitudinal approaches. Disability and Rehabilitation: Assistive
Technology, 1–26. https://doi.org/10.1080/17483107.2025.2579825
Bertelli, M. O., Boniotti, V., & Scior, K. (2024). Is it still autism? the increasing broadening of
the autism spectrum. Autism Research, 18(1), 37–43. https://doi.org/10.1002/aur.3282
Crasta, J. E., Green, O. J., Gavin, W. J., & Davies, P. L. (2023). The relationship between
attention, sensory processing, and social responsiveness among adults on the autism
spectrum. Journal of Autism and Developmental Disorders, 54(8), 2972–2986.
https://doi.org/10.1007/s10803-023-06019-1
Dhabliya, D., Dari, S. S., Dhablia, A., Akhila, N., Kachhoria, R., & Khetani, V. (2024).
Addressing bias in machine learning algorithms: Promoting fairness and ethical design.
E3S Web of Conferences, 491, 02040. https://doi.org/10.1051/e3sconf/202449102040
Doroudi, S. (2022). The intertwined histories of artificial intelligence and Education.
International Journal of Artificial Intelligence in Education, 33(4), 885–928.
https://doi.org/10.1007/s40593-022-00313-2
D’Mello, A., Frosch, I., Li, C., Cardinaux, A., & Gabrieli, J. (2022). Exclusion of Females in
Autism Research: Empirical Evidence for a “Leaky” Recruitment-to-Research Pipeline.
https://doi.org/10.31234/osf.io/g65vk
Guillery-Girard, B., Parienti, J.-J., Eustache, F., Guenole, F., Baleyte, J.-M., Gerardin, P., Ring,
M., Bowler, D., Briant, A. R., & Desaunay, P. (2020). Memory in Autism Spectrum
Disorder: A Meta-Analysis of Experimental Studies. Psychological Bulletin, 146(5), 377–
410. https://doi.org/10.1037/bul0000225.supp
Hollocks, M. J., McQuaid, G. A., Lee, N. R., & Wallace, G. L. (2025). Cognitive flexibility
mediates the associations between perceived stress, social camouflaging and mental health
challenges in autistic adults. Autism Research, 18(8), 1595–1607.
https://doi.org/10.1002/aur.70061
Hollocks, M. J., McQuaid, G. A., Yerys, B. E., Strang, J. F., Anthony, L. G., Kenworthy, L., Lee,
N. R., & Wallace, G. L. (2023). Measuring flexibility in autistic adults: Exploring the
factor structure of the Flexibility Scale Self Report. Autism Research, 16(11), 2208–2219.
https://doi.org/10.1002/aur.3025
Jertberg, R. M., Begeer, S., Geurts, H. M., Chakrabarti, B., & Van der Burg, E. (2025). Slow but
steady: Similarities and differences in executive functioning between autistic and non‐
autistic adults. Autism Research, 18(4), 802–819. https://doi.org/10.1002/aur.70015
Johnson, S., Davis, S., Dilcher, V., Everett, A., Lee, S., & Miller, L. (2024). Cognitive Strategies
for Executive Functioning Skills Among Autistic Adults & Adults With Intellectual
Disabilities: A Scoping Review. American Journal of Occupational Therapy, 78(S2).
Kotsi, S., Handrinou, S., Iatraki, G., & Soulis, S.-G. (2025). A review of artificial intelligence
interventions for students with autism spectrum disorder. Disabilities, 5(1), 7.
https://doi.org/10.3390/disabilities5010007
Lage, C., Smith, E. S., & Lawson, R. P. (2022). A Meta-Analysis of Cognitive Flexibility in
Autism Spectrum Disorder. https://doi.org/10.22541/au.166733713.33443910/v1
Lampinen, L. A., Singer, J., Wang, X., VanHook, B., Wilkinson, E., & Bal, V. H. (2025). Selfreported strengths and talents of autistic adults. Autism.
https://doi.org/10.1177/13623613251364361
Macari, S. L., Vernetti, A., & Chawarska, K. (2020). Attend less, Fear more: Elevated distress to
social threat in toddlers with autism spectrum disorder. Autism Research, 14(5), 1025–
1036. https://doi.org/10.1002/aur.2448
Maye, M., Boyd, B. A., Martínez-Pedraza, F., Halladay, A., Thurm, A., & Mandell, D. S. (2021).
Biases, barriers, and possible solutions: Steps towards addressing autism researchers
under-engagement with racially, ethnically, and socioeconomically diverse communities.
Journal of Autism and Developmental Disorders, 52(9), 4206–4211.
https://doi.org/10.1007/s10803-021-05250-y
Ne’eman, A., Richman, K. A., McCarthy, A. M., & Wilkenfeld, D. (2023). A passing problem:
Evaluating harm and benefit in autism research. Ethics & Human Research, 45(6), 2–
18. https://doi.org/10.1002/eahr.500188
Rabot, J., Rødgaard, E.-M., Joober, R., Dumas, G., Bzdok, D., Bernhardt, B., Jacquemont, S., &
Mottron, L. (2023). Genesis, modelling and methodological remedies to autism
heterogeneity. Neuroscience & Biobehavioral Reviews, 150, 105201.
https://doi.org/10.1016/j.neubiorev.2023.105201
Rozenkrantz, L., D’Mello, A. M., & Gabrieli, J. D. E. (2021). Enhanced rationality in autism
spectrum disorder. Trends in Cognitive Sciences, 25(8), 685–696.
https://doi.org/10.1016/j.tics.2021.05.004
Tse, V. W., Lei, J., Crabtree, J., Mandy, W., & Stott, J. (2021). Characteristics of older autistic
adults: A systematic review of literature. Review Journal of Autism and Developmental
Disorders, 9(2), 184–207. https://doi.org/10.1007/s40489-021-00238-x
Venkatesan, S., & Tolani, P. (2024). Concealed and compromised: Masking, autism and graphic
medicine. Journal of Graphic Novels and Comics, 1–18.
https://doi.org/10.1080/21504857.2024.2426557
Voultsiou, E., & Moussiades, L. (2025). A systematic review of AI, VR, and LLM applications in
special education: Opportunities, challenges, and future directions. Education and
Information Technologies, 30(13), 19141–19181. https://doi.org/10.1007/s10639-025-
13550-4
Waterhouse, L. (2021). The problem of heterogeneity in autism: Response to Mottron (2021)
“Aradical change in our autism research strategy is needed: Back to prototypes.” Autism
Research. https://doi.org/10.1002/aur.2584
Yang, Y., Chen, L., He, W., Sun, D., & Salas-Pilco, S. Z. (2024). Artificial Intelligence for
Enhancing Special Education for K-12: A decade of trends, themes, and Global Insights
(2013–2023). International Journal of Artificial Intelligence in Education, 35(3), 1129–
1177. https://doi.org/10.1007/s40593-024-00422-0
Zhang, R. (2025a). Application of AI in Education and its influence on students’ critical thinking.
SHS Web of Conferences, 222, 01010. https://doi.org/10.1051/shsconf/202522201010
Zhang, R. (2025b). Application of AI in Education and its influence on students’ critical thinking.
SHS Web of Conferences, 222, 01010. https://doi.org/10.1051/shsconf/202522201010
Add comment
Comments