AI is the solution to costly and ineffective dyslexia programs

Worse, some screeners do not even test students’ skills. For example, the Shaywitz DyslexiaScreen (Yale) is a survey that asks teachers to make judgments on how often they think a student displays problems in their language behaviors such as oral expression.⁷

Given the screening mandate, some publishers have fashioned screeners out of their reading assessment instruments designed for the typical population. This ignores the language processing differences between typical and exceptional brains.

Schools are wary about dyslexia screening for their own reasons:

• Schools do not have the resources to provide special services to more students if screening identifies additional at-risk readers (see Cost below).
• Schools have to pay for the screeners and teachers to get trained in administering the screening.
• Teachers and students have to take time out to do the screening, in addition to other assessments (state and quarterly/trimester reading and math assessments, plus additional testing for special ed students).
• Teachers have to interpret screening results manually, one student at a time, to differentiate instruction.

Multisensory intervention

Multisensory intervention is promoted by the U.S.-based International Dyslexia Association (IDA).⁸ The IDA was formerly the Orton Society, named after neuropathologist Samuel T. Orton. Orton’s work on dyslexia started in the 1920s and led to multisensory teaching techniques due to his (incorrect) hypothesis that dyslexia was vision-based. (Dyslexia was later proven to be language-based.) Later, Anna Gillingham incorporated spelling rules to this multisensory Orton-Gillingham (OG) approach. Students use multiple senses to try to remember the sound of a letter, e.g., by tracing and feeling letter shapes.⁹

In conjunction with multisensory intervention, the IDA promotes structured literacy(multisensory structured literacy – MSL). The latter involves systematic instruction, from sounds and spelling to other language components (parts of words, sentences, meaning, plus reading).

Popularity: OG/MSL is the most popular dyslexia/reading intervention in the U.S. It is mandated in some states and required in teacher training programs. Schools purchase OG-based commercial programs and/or pay for OG training and certification for their teachers.¹⁰ Prominent dyslexia researchers support MSL. Some are IDA award recipients and/or members of IDA’s Council of Advisors and Scientific Advisory Board. Some also served on dyslexia task forces or literacy-focused committees at the state and federal levels. The IDA publishes several academic journals on dyslexia and literacy.¹¹

Research evidence lacking: Despite OG/MSL’s popularity and backing, research evidence of its efficacy for students with dyslexia beyond 3^rd grade is lacking, despite almost a century of application. This is the conclusion of meta-analytic reviews of rigorous clinical trials. A 2014 review by Durham and Yale University professors echoes this finding (Cambridge), as did other 2006, 2021, and 2022 reviews.¹² The reviews find that many studies lack scientific rigor. The U.S. Department of Education’s What Works Clearinghouse for literacy does not contain any study that has a significant positive effect on broad reading achievement for this group.¹³ Positive effects only show in narrow areas such as alphabetic knowledge in younger students.

Evidence-based

Yet, proponents call their methods and screeners “evidence-based.” According to the DOE’s Institute of Education Sciences, this term can be used when there is a rationale, based on research findings, that the practice is likely to improve student outcomes, with ongoing efforts to examine such effects.¹⁴ That is, methods and tools called “evidence-based” do not necessarily have evidence of positive effects.

Feasibility of implementation

Cost

Special ed: U.S. schools collectively spend over $120 billion a year on special education.¹⁵ The largest category is learning disabilities, mainly dyslexia.¹⁶ Reading difficulties also occur across other disabilities such as autism and ADHD.¹⁷ In some states like Connecticut, the cost of educating a student in special ed is twice as much as general education ($40k+ v. $20k+ per year).¹⁸ Due to this high cost, less than half the students with reading difficulties across the country get special ed.¹⁹ Often, only half of the bottom third of chronically failing students get services.

Teacher training: Dyslexia/reading interventions are expensive because teachers work one-on-one or in small groups. Also, teachers have to complete extensive coursework over at least 2 years and 100+ hours of practicum to be certified to use OG.²⁰ Some schools pay $10k for each teacher to get OG training.²¹ Schools also pay for teacher training to use dyslexia screeners and commercial programs.

Long-term cost: As OG cannot correct the reading difficulty, the same students still need services throughout school.²² Thus, the total cost for each special ed student in states like Massachusetts may be over $200k ($20k x 10 years – grades 3-12).

Literacy initiatives: There have been many intensive, well-funded initiatives to address this problem, which is the biggest and costliest in education. The Bush administration’s 3-year, $3 billion initiative did not produce statistically significant impact on reading comprehension.²³ Ohio state’s 4-year Dyslexia Pilot Project on grades 1-3 did not reduce the risk of reading failure in 3^rd graders ($960,000).²⁴ Various grant-supported, university-run interventions similarly had negligible impact.³

Fiscal pressures: In some states, such as New York and Minnesota, special ed spending is increasing by double digits, sometimes more than 20 percent.²⁵ This expansion is likely to affect more states post-pandemic as a result of learning loss due to school shutdowns.¹ The impact on young children during their most critical years of language development is still unclear at this point.

The federal government underfunds special ed and leaves states with a $10 billion shortfall every year.²⁶ School districts are forced to make up this deficit with tax increases. As a result, residents on fixed income in some cities such as Columbus, Ohio, are forced out of their homes.²⁷

If the current path of covering half the students in need is already unsustainable, how can we cover all students in need?

Fidelity of implementation

Research shows that a 3^rd grader with dyslexia will likely read below grade level throughout school regardless of intervention.²⁸ When reading outcomes are poor, fidelity of implementation is often blamed: hence the constant call for more teachers to get trained and for them to get trained adequately. However, it is time to re-examine the method itself if no one has been able to get this group to read on grade level consistently for almost 100 years. Moreover, even high-fidelity interventions run by university professors have not succeeded.

Scalability

Workflow bottlenecks: Bottlenecks at every stage of the process, from dyslexia identification through intervention, have created the current situation. Presently, dyslexia is diagnosed through lengthy one-on-one neuropsychological evaluations by certified specialists that cost about $5,000-10,000 per pupil.²⁹ Schools cannot afford to evaluate all at-risk students in this way. And there are not enough specialists to evaluate everyone.

Thus, a brief universal screening was introduced (to suggest risk of dyslexia, not diagnose). But the disjuncture remains between identification and intervention, whether with screeners or evaluations. Teachers have to interpret the results student by student to differentiate instruction.³⁰ And differentiated instruction can only serve one or just a few students at a time.

Dyslexia schools: Mainly OG-based, dyslexia schools can only serve small groups due to this labor-intensive method. At over $70,000 a year per pupil at the Windward School, this model is not feasible for wide adoption.³¹ Despite their supporters’ testimonials, they lack research evidence of efficacy as noted above.

Dyslexia classification

There is no consensus among researchers on the definition of dyslexia.¹² It is not classified as a disorder in the American Psychiatric Association’s authoritative diagnostic guide, Diagnostic and Statistical Manual of Mental Disorders 5^th ed. (DSM-5).

Disjuncture due to historical accident: Dyslexia was initially mistaken to be an eye problem.³² Due to this historical accident, diagnosis still remains with the medical profession. However, since physicians do not know how to resolve language difficulties, treatment is passed on to teachers. Up to now, families still go to pediatricians to get a dyslexia diagnosis in order to qualify for intervention at school.

Dyslexia or reading difficulty? Some researchers equate dyslexia with reading difficulty, while others consider dyslexia as a subset of reading difficulties. Because of this confusion, some researchers suggest dispensing with the term “dyslexia” and instead focus on reading difficulties.¹² The latter can be identified with norm-referenced measures in state and standardized reading assessments. Thus, the 25^th-30^th percentile in these assessments have been used as the dyslexia threshold in research.³³

This confusion permeates the school system:

• Many educators think of dyslexia as a rare condition that is not commonly found among their struggling readers.
• Teachers avoid using the term in fear of liability issues – dyslexia as a disability is covered by the IDEA (Individuals with Disabilities Education Act), which mandates free, appropriate education for students with disabilities.
• Some quarters discriminate between students who have been diagnosed (through expensive neuropsychological evaluations) and economically disadvantaged students whose poor reading skills are assumed to be environmental instead.¹²

Artificial intelligence

Game-based AI systems can evaluate and correct language processing deficits autonomously. These processing deficits underlie conditions such as dyslexia. Users play the language-based evaluative and corrective games generated by AI. A user typically goes through thousands of games in the course of a year of intervention.

The micro data collected on their game responses yielded new, detailed information about the linguistic-cognitive processes of dyslexic brains. Some findings:

• Dyslexia can be defined operationally in terms of language processing inefficiencies. These inefficiencies can be measured precisely (Efficiency = Accuracy x Speed).
• Below a certain level of efficiency in key areas of literacy development, reading difficulty or dyslexia manifests on the surface.
• The linguistic system in the brain is highly fragile and requires 90-100 percent efficiency in core areas for a person to function effectively as a speaker, listener, reader and writer. It is not unusual to find some struggling readers operating at 20 percent efficiency in some areas when they first start.
• Language processing efficiencies can be improved with weekly intervention (corrective training). Improvements in brain processing remain latent for the first several months of intervention (i.e., not noticeable overtly). When the user reaches 90-100 percent efficiency in core areas consistently, then, within a 2-month window, academic improvements follow, e.g., in spelling tests and eventually in state and standardized reading assessments.
• The effect is not gradual but shows a sharp knee-of-the curve upward trajectory. This improvement in brain efficiency often affects the whole person, in terms of learning behaviors, attitude and motivation.

Early results: AI generates an individually customized evaluation and intervention for each user. All games are single use, built specifically for that one user at that moment. Users who complete their programs have been able to get to grade-level reading within 1-2 years. They often start at below the 25^th percentile in state or standardized reading assessments.

Integrating screening/evaluation and intervention: AI evaluation results form the blueprint for intervention. AI tracks a student’s errors to get to the source of processing difficulties. To respond intelligently in this way, the AI system comprises massive complex databases and custom-built, domain-specific algorithms. The user’s own database, which includes all their game responses, grows as they progress. The system’s autonomous decision-making on what game to build for a user next is based on continuous, real-time analysis of all information in their own database, cross-referenced with other relevant databases.

To be an effective evaluator-interventionist, AI can be built to switch language modalities flexibly. For example, if a student misspells words with the “th” sound, AI may ask them to catch this sound in heard words in the next game.

Multisensory: This AI does not make struggling readers learn what they cannot learn. It first determines whether they can actually process a particular part of language. For example, when they start, many struggling readers say that their teachers’ instruction on phonics “does not make sense” (letter-sound relations). Because these students cannot process speech sounds, the letter shapes hold no meaning to them, no matter how many sensory activities they experience.

Structured literacy: For the shortest intervention possible, AI can target specific inefficient processes in each student’s linguistic system. Moreover, AI does not merely consider the structural components of language but also the non-typical ways in which exceptional brains operate.

Cost: This program currently costs less than 10 percent of what states spend on special ed per pupil. In New York, it is 1 percent.

NY special ed = $20k/pupil/year x 10 years of intervention = $200k.³⁷

AI program = $1k/pupil/year x 2 years = $2k (to get to reading proficiency).

Teacher training: The program is plug-and-play and web-based. It does not require installations, downloads, upgrades or teacher training. In fact, children require minimal supervision, which can be from any adult. It can run for 2 years or longer automatically without human intervention.

Scalability: AI can provide evaluations and interventions to millions of students today.

Dyslexia or reading difficulty: Different authorities in the field carve out “dyslexia” differently on the spectrum of language processing difficulties (or inefficiencies). Instead of focusing on this labeling issue, AI prioritizes getting all readers who are performing below the 50^th percentile to passing. This means identifying each person’s processing inefficiencies and correcting them.

Why does AI work? And why does it work so fast?

The dyslexia problem required AI to overcome 3 major obstacles that human specialists could not:

• Complexity of the linguistic system in the brain and the wide individual variation in the population with language processing difficulties.
• Speed of language processing in the brain (often in hundreds of milliseconds in parallel).
• Capacity to document and cross-reference billions of datapoints per student and do the same for millions of users.

AI is effective and efficient because it only generates the data that it needs to read at that point. Yet every decision is based on all relevant data at its disposal. Through AI, we have found a way to read individual brains at the level of granularity needed to correct brain processes.

Coral PS Hoh is a clinical linguist with a PhD in Linguistics. For the past 30 years, she worked directly with the gifted, twice-exceptional and individuals with language disorders. Her publications span topics in syntax, communication, cognition and the acquisition of phonology, speech, reading and writing, as well as the linguistic behaviors of exceptional populations. She has refereed for journals in linguistics, math education and special education. Her work was funded by the National Science Foundation. She is the architect of Dysolve® AI and was granted patents by the U.S. and other countries for its autonomous AI technology. Dr. Hoh co-founded the AI firm, EduNational LLC, and currently serves as its CEO.

NOTES AND REFERENCES

1. National Center for Education Statistics. (2022). National achievement-level results. NAEP Report Card: Reading. U.S. Department of Education, Institute of Education Sciences. https://www.nationsreportcard.gov/reading/nation/achievement/?grade=4
2. National Institute of Child Health & Human Development, NIH, U.S. Dept of Health & Human Services. (2012). How many people are affected by/at risk for reading disorders? https://www.nichd.nih.gov/health/topics/reading/conditioninfo/pages/risk.aspx
3. Fuchs, D., Compton, D.L., Fuchs, L.S., Bryant, V.J., Hamlett, C.L., & Lambert, W. (2012). First-grade cognitive abilities as long-term predictors of reading comprehension and disability status. Journal of Learning Disabilities, 45(3), 217-231.
4. National Center on Improving Literacy. (2024). State of Dyslexia. https://improvingliteracy.org/state-of-dyslexia/
5. New York State Senate. (2024). Senate Bill S5232.
6. University of Oregon. (2023). DIBELS 8^th Ed. https://dibels.uoregon.edu/dibels8
   Illuminate Education. (2023). Fastbridge Dyslexia Screener. https://www.illuminateed.com/products/fastbridge/reading-assessment/dyslexia-screening/
   Lexercise. (2023). Free Dyslexia Test for Children. https://www.lexercise.com/tests/dyslexia-test
7. Pearson. (2023). Shaywitz DyslexiaScreen. https://www.pearsonassessments.com/store/usassessments/en/Store/Professional-Assessments/MTSS-RTI-PBIS-Benchmarking/Shaywitz-DyslexiaScreen/p/100001918.html
8. International Dyslexia Association. (2024). IDA moves beyond matrix. https://dyslexiaida.org/update-on-idas-matrix/
9. Cox, A.R. (1985). Alphabetic phonics: An organization and expansion of Orton-Gillingham. Annals of Dyslexia, 35, 187-198.
10. Shresta, P., Tracy, T., Mazal, M., Blakeney, A., Kennedy, N., & May, H. (2022). A cost analysis of Reading Recovery and alternative interventions under the i3 Scale-Up. Paper presented at the Annual Conference of the American Education Research Association (AERA). San Diego, CA, Apr 23.
    Orton-Gillingham Academy. (nd). Orton-Gillingham practitioner certification. https://www.ortonacademy.org/training-certification/individual-certification/
11. Perspectives on Language & Literacy, The Examiner, Annals of Dyslexia, Reading & Writing
12. Elliott, J. G., & Grigorenko, E. L. (2014). The dyslexia debate. New York: Cambridge.
    Ritchey, K.D., & Goeke, J.L. (2006). Orton-Gillingham and Orton-Gillingham-based reading instruction: A review of the literature. The Journal of Special Education, 40(3), 171-183.
    Stevens, E.A., Austin, C., Moore, C., Scammacca, N., Boucher, A.N., & Vaughn, S. (2021). Current state of the evidence: Examining the effects of Orton-Gillingham reading interventions for students with or at risk for word-level reading disabilities. Exceptional Children, 87(4), 397-417.
    Sayeski, K.L., & Hurford, D.P. (2022). A framework for examining reading-related education research and the curious case of Orton-Gillingham. Learning Disabilities, 27(2), 1-11.
13. What Works Clearinghouse. Search results: Literacy topic. https://ies.ed.gov/ncee/wwc/Search/Products?searchTerm=&Topic=3
14. Institute of Education Sciences. (2024). Building and using evidence in data science education – What is an evidence-based practice? Evidence Based Practice. https://ies.ed.gov/EvidenceBased/dataliteracy/ds-ebp.asp
15. Graves, K. (Nov 20, 2023). Funding falls short for students with disabilities. The Principal Advocate. National Association of Elementary School Principals (NAESP). https://www.naesp.org/blog/funding-falls-short-for-students-with-disabilities/
    National Center for Education Statistics. (2024). Current Expenditures for Public Elementary and Secondary Education: School Year 2003-04. https://nces.ed.gov/pubsearch/pubsinfo.asp?pubid=2006352
16. Office of Special Education Programs. (2023). 44th Annual Report to Congress on the Implementation of the IDEA. Washington, DC: U.S. Dept of Education.
17. Germano, E., Gagliano, A. & Curatolo, P. (2010). Comorbidity of ADHD and dyslexia. Developmental Neuropsychology, 35(5), 475-493.
18. EdBuild. (2024). FundEd: State policy analysis. A detailed look at each state’s funding policies. http://funded.edbuild.org/state
19. Compare the 20 percent with dyslexia to the 9 percent of student population in special education for all disabilities – see Annual Reports to Congress on the Implementation of the IDEA.
20. Orton Gillingham Online Academy. (2022). Practicum for Level 1 with OG certification. ortongillinghamonlinetutor.com
21. Personal communication, Principal, Mill Road Intermediate School, Red Hook Central Schools, NY
22. Donegan, R. E., Wanzek, J., & Al Otaiba, S. (2020). Effects of a reading intervention implemented at differing intensities for upper elementary students. Learning Disabilities Research & Practice, 35(2), 62-71.
23. Gamse, B. C., Jacob, R. T., Horst, M., Boulay, B., & Unlu, F. (2008). Reading First Impact Study Final Report Executive Summary (NCEE 2009-4039). Washington, DC: National Center for Education Evaluation and Regional Assistance, Institute of Education Sciences, U.S. Department of Education.
24. Morrison, J.Q., Collins, T.A., & Hawkins, R.O. (2016). Evaluation of the Dyslexia Pilot Project: Year 4. Cincinnati, OH: University of Cincinnati.
25. Wastvedt, S. (May 16, 2018). As schools struggle with costs, special education takes a toll. NPR News. https://www.mprnews.org/story/2018/05/16/specialeducation-takes-toll-as-schools-struggle
26. Lieberman, M. (Apr 20, 2023). Special education is getting more expensive, forcing schools to make cuts elsewhere. EducationWeek. https://www.edweek.org/leadership/special-education-is-getting-more-expensive-forcing-schools-to-make-cuts-elsewhere/2023/04
27. Cleary, I. (Mar 26, 2024). Columbus residents speak up about property taxes in the city. NBC 4i. https://www.nbc4i.com/news/local-news/columbus/columbus-residents-speak-up-about-property-taxes-in-the-city/
28. Fiester, L. (2010). Why reading by the end of third grade matters. A KIDS COUNT Special Report from the Annie E. Casey Foundation. Baltimore, MD: Annie E. Casey Foundation.
29. Columbia University Irving Medical Center. (2024). Neuropsychological evaluation service. https://www.columbiadoctors.org/childrens-health/pediatric-specialties/psychiatry-psychology/our-services/neuropsychological-evaluation-service
30. Liewinko, K.N. (Mar 25, 2024). Dyslexia breakthrough: AI powered. Assembling Inclusion podcast. https://www.youtube.com/@assemblinginclusion/podcasts
31. Windward School. (2024). Affording a Windward education. https://www.thewindwardschool.org/admissions/tuition-financial-aid
32. Shaywitz, S.E. (2003). Overcoming dyslexia. New York: Knopf.
33. Snowling, M.J. (2000). Dyslexia. Oxford, UK: Blackwell.
    Wanzek, J. & Roberts, G. (2012). Reading interventions with varying instructional emphases for fourth graders with reading difficulties. Learning Disability Quarterly, 35(2), 90-101.
34. Hoh, C.P.S., & Haruta, E.Y. (2018). Dyslexia dissolved: Successful cases with learning disabilities, ADHD and language disorders. https://www.amazon.com/Dyslexia-Dissolved-Successful-Disabilities-Disorders/dp/173278860X
35. https://www.dysolve.com/press/
36. Registry of Efficacy and Effectiveness Studies, Institute for Social Research, University of Michigan
    https://sreereg.icpsr.umich.edu/sreereg/subEntry/21840/pdf?action=view
37. Office of the New York State Comptroller. (2020). Special report update: Education revenues and expenditures with a highlight on special education for regions outside New York City. New York: Office of the New York State Comptroller.