Sage Journals: Discover world-class research

Abstract

Although reading in the digital age looks different than in the past (e.g., on a screen, shorter texts spread across multiple sources), reading continues to be a central part of everyday life. Research in reading comprehension shows that this type of modern reading requires more complex knowledge building inference processes that are difficult for many adolescents and adults. Constructive and personalized learning activities can support readers’ knowledge building. Researchers have been using artificial intelligence (AI) to make these types of activities more effective and accessible. Emerging directions and considerations also result from the introduction of generative AI. Increased collaboration across researchers, developers, educators, and policymakers would afford empirically supported research, development, and implementation that can keep pace with the quickly evolving technological landscape.

Keywords

reading comprehension learning from text artificial intelligence education personalization

The digital age has changed how people read and learn. Comprehension theory and the science of learning suggest how to support today's readers and how researchers implement personalized, constructive learning activities using AI-based systems.

Key Points

Many adolescents and adults need support for the reading comprehension challenges that encounter in their everyday lives.

Readers are knowledge builders who need opportunities to engage in constructive activities that are personalized to their skills, knowledge, and interests.

AI-based tools can facilitate constructive and personalized learning activities in ways that help more people become better readers.

The rapid development of AI demands that policymakers, researchers, developers, and educators engage in collaborative and accelerated research and development toward theory-driven technologies that are effective and equitable.

Although the digital age has transformed the types of texts that people read and how they read them, reading remains central to how people acquire knowledge. Unfortunately, many adolescents and adults are not ready for the reading comprehension tasks that they will encounter in their classrooms, at work, and in their everyday lives (e.g., ACT, 2018). Researchers from across the brain and behavioral sciences alongside their colleagues in education have identified what challenges these struggling readers face and how to best support them to have better comprehension during individual reading tasks and to become better readers over time. In addition, the rapid development of artificial intelligence (AI) and AI-based technologies stand to greatly advance society's ability to support people's reading comprehension. On one hand, AI offers unprecedented opportunities to engage learners in theory-informed activities that can improve comprehension and to do so in ways that meet the learner where they are, so that they can build their skills. On the other hand, AI presents several considerations that ought to be studied before such technologies are indiscriminately implemented.

For many people, the idea of AI still feels like science fiction, yet AI is everywhere: smart home devices, predictive text in a Google search, or crafting an email all use AI. More recently, powerful generative AI tools have come online and tools such as ChatGPT serve as a way for anyone to be able to leverage this state-of-the-art technology with no programming experience necessary. Such ubiquity invites policymakers, educators, and the public to deepen their understanding of best practices uncovered by the science of learning and the ways in which AI is being used to support learners. After a brief overview of modern reading comprehension, the review then examines three broad ways in which researchers are leveraging AI in this space—examining the current state-of-the-art, ongoing challenges, and opportunities for future research.

Reading Comprehension in the Digital Era: The Modern Reader

Magliano and colleagues (2017) used the idea of the “The Modern Reader” to explore how the digital age has changed the way that people consume information. The modern reader uses reading as a knowledge building activity toward some goal: a college student attempting to complete coursework; an employee trying to learn a new procedure; a citizen who needs to make a high-stakes medical or political decision that will impact their family. As these examples highlight, reading comprehension is complex and purposeful (Britt et al., 2022). Readers need not only read a text and be able to answer simple questions about it, but they need to navigate across multiple, oftentimes conflicting sources in order to synthesize, evaluate, and communicate about complex ideas (Barzilai & Strømsø, 2018) and need to do so in specialized content domains (Goldman, 2012; see also Snow, 2002).

Unfortunately, reports show that adolescent and adult readers are underprepared to engage in these types of knowledge building tasks. PIAAC results suggest that nearly half of U.S. adults struggle to read texts that are “dense or lengthy and include continuous, noncontinuous, mixed, or multiple pages of text” nor can they complete tasks that require one “to identify, interpret, or evaluate one or more pieces of information” or “to construct meaning across larger chunks of text” (OECD, 2013). The majority of graduating high school seniors are unable to read at a proficient level (ACT, 2018) and 60%-75% of first-year college students are underprepared to read for their college courses (Perin, 2020).

Research in the fields of text and discourse processing has addressed these concerns by attempting to better understand what processes, skills, and knowledge are brought to bear as readers engage in complex comprehension tasks. Comprehension theories suggest that reading comprehension involves readers constructing a mental model of the topic at hand (McNamara & Magliano, 2009). Researchers conceptualize this mental model as a connectionist network of nodes and links. For better comprehension (i.e., a more elaborated and coherent mental model), readers need to not only process the explicit words and ideas, but to generate inferences that help to create the links across ideas in the text and to connect the text content with their existing prior knowledge (McNamara & Magliano, 2009). Broadly, the research suggests that (a) aspects of the text, task, and reader influence the quantity and quality of the inferences that a reader generates and that (b) instruction and scaffolding can help learners to construct more inferences that support a higher quality mental model.

Research further suggests that modern, purposeful reading is heavily dependent on a learner's prior knowledge (Cervetti et al., 2020; McCarthy & McNamara, 2021; Kendeou et al., 2023). Wang and colleagues (2021) had students complete a traditional multiple-choice reading comprehension test and a scenario-based assessment designed to reflect modern reading. Both tests used texts about American football. While the traditional test performance was strongly predicted by basic reading skills and general academic knowledge, performance on the scenario-based assessment was most strongly predicted by learners’ knowledge of football. Thus, there is renewed interest in how to measure and leverage what a learner already knows to help them build their knowledge (Hattan et al., 2023; McCarthy & McNamara, 2021; Simonsmeier et al., 2022).

Helping the Modern Reader: Exploring the Potential of AI

As the tasks for the modern reader have become more complex, researchers and educators have needed to develop methods of instruction, scaffolding, and evaluation to match that complexity. One emerging area is the use of AI to study and support comprehension. AI refers to a range of technologies that allow computers to perform tasks that would typically require human cognitive abilities. More simply, AI analyzes old data to identify patterns that allow it to make predictions about new data. AI allows computers to automate repetitive or complex tasks in ways that make complex activities more scalable. One prominent subset of AI is natural language processing (NLP), which focuses on developing methods for computers to understand, interpret, and generate human language. NLP tools can be used to extract various aspects of language, from individual words to the text as a whole. In the last decade, significant advancements have been made in NLP, particularly with the introduction of large language models (LLMs) such as BERT (Bidirectional Encoder Representations from Transformers) and GPT (Generative Pre-training Transformer). LLMs are trained on massive datasets, which allows them to learn statistical relationships between words and ideas. LLMs are a substantial advancement over previous AI models. They can learn from unlabeled (i.e., not pre-scored by humans) data, which allows them to train on massive datasets without manual labeling. They can also adapt to new tasks without explicit training, which also allows them to generate new content (Yan et al., 2023).

Although the field is in relative infancy, significant strides have been made in employing AI in reading comprehension research. The review highlights three areas where AI has been used to examine ways of studying and supporting the modern reader:

Implementing constructive learning activities at scale

Providing tailored, personalized instruction and feedback, and

Facilitating data collection and analysis that expands understanding of comprehension

Constructive Learning Activities

Much of what is known about support for reading comprehension comes from studies of skilled and less skilled readers. Less skilled readers tend to maintain a relatively narrow focus, largely restating information that is in the text. On the other hand, skilled readers engage in more inferencing as they read. Not only do they bridge ideas from sentence to sentence, but they also connect what they are currently reading to the larger global purpose of the text, and they are more likely to use their prior knowledge to elaborate upon what is in the text (e.g., Wolfe & Goldman, 2005). Critically, experimental studies show that encouraging less skilled readers to make inferences and to draw upon their prior knowledge improves their comprehension (McNamara, 2007). These findings further emphasize the role of knowledge building and are consistent with a broader base of research in the science of learning (see How People Learn II, NASEM, 2018), demonstrating that activities in support of active and constructive¹ processing are better for learning (Chi & Wylie, 2014; Dunlosky et al., 2013; Fiorella & Mayer, 2016).

Constructive Comprehension Tasks. Constructive activities can be broadly categorized into two ways: the focal task and supplementary strategies. A constructive task is often open-ended, such as reading texts to write an essay. Essays are a common way to both invite and evaluate learning. Asking readers to write encourages them to take what they have read and to transform and organize the information in ways that construct a more coherent mental model (Graham & Hebert, 2011; Wiley & Voss, 1999). Unfortunately, writing is a barrier for many students, and they often do not get sufficient instruction and feedback to improve their writing, due to the time-intensive nature of essay writing (Graham & Hebert, 2011). Automated writing evaluators (AWEs) make it possible for students to have more opportunities to practice writing and to get feedback on how to improve. AWEs can provide near instantaneous evaluation that is on par with human judgments (Shermis et al., 2016). Rather than needing to wait for their teacher to return their assignment, students can engage in multiple rounds of practice and feedback. Research also demonstrates that combining AWEs with teacher feedback can be a powerful approach. Wilson and Czik (2016) found that supplementing teacher feedback with AWEs allowed the teachers to devote their resources to giving higher order feedback that was focused on content and structure (see also Baffour, 2023).

Constructive Strategies. We use constructive strategies to refer to an activity that is usually supplementary to the focal task. Constructive comprehension strategies such as self-explanation, generating a concept map, writing a summary, or drawing a diagram are effective tools for better comprehension (Dunlosky et al., 2013; Fiorella & Mayer, 2016). AI-supported tools stand to make these complex constructive strategies easier to implement in modern learning contexts. For example, SMART (Student Mental Model Analyzer for Reading and Teaching) helps college students with their content reading. Many college courses require students to read prerequisite materials before class. However, these texts are often challenging, and students may not have anyone to help them work through the content at home. SMART is a web-based tool that students can access from anywhere. First, SMART prompts students to write a summary of their assigned chapter. Then, SMART uses NLP and network analysis to generate a concept map of the student's summary. The student's map is compared to an ideal or “benchmark.” Students receive a summative score and formative feedback that helps them revise their summary to emphasize not only the number of ideas, but the connections between those ideas to help them construct a more coherent mental model (Kim & McCarthy, 2021). Thus, SMART ensures that the students are not just coming to class having read the assignment, but that they have understood the content well enough to engage in class activities that can further their knowledge building.

Research has demonstrated the utility of open-ended construction of written content, which typically would require an instructor to interpret the writing and provide feedback. AI technologies allow these learning tasks to be completed more efficiently so that more students have more opportunities to engage in meaningful knowledge building.

Personalized Instruction

A learner approaches a given reading task with a unique blend of strengths and limitations and, with respect to prior knowledge, different experiences. Each of these influences how and what the person learns. Given this variability, attempting to reach the “average learner” often fails to meet any one individual (Connor & Morrison, 2016). Thus for individualization and personalization of instruction, these approaches go beyond asking “what works?” to asking, “what works, for whom, and under what conditions?” Personalized instruction emphasizes not only abilities, but also the need to consider students’ interests and the relevance of the content to their experiences (Walkington & Bernacki, 2018). These efforts also highlight a movement away from deficit-based remediation, toward asset-based instruction. Not only are asset-based approaches more equitable, but they also align with best practices in knowledge building as drawing from their own knowledge and lived experiences allows students to engage in the constructive processes that support learning (McCarthy & McNamara, 2022). Work in AI has been developing methods to better evaluate differences in learners and their learning process and to deliver a variety of supports (e.g., feedback) to address those differences.

Evaluating the Learner and Their Learning Process. Constructive learning activities are not only more effective, they also provide richer information about student learning. Traditional assessments are relatively limited to testing what a student knows rather than the comprehension and reasoning processes about how they know. Such tests are also limited in that they evaluate decontextualized, isolated skills rather than coordination of context-specific factors and processes (Lee, 2023). Automating the use of complex reading comprehension tasks affords a greater understanding of what goes into modern reading in more authentic contexts.

Stealth assessment is a promising new approach to evaluating student learning that uses AI to analyze learners’ behavioral data (Shute & Rahimi, 2022). Stealth assessment offers a rich and nuanced understanding of students’ existing skills and knowledge, as well as their learning processes, without subjecting them to numerous standardized tests. Students are not aware that they are being assessed, which is less stressful and allows them to engage in an authentic learning task. Developing research demonstrates use of stealth assessment to evaluate students’ proficiencies, such as their general vocabulary knowledge or their performance on a general reading comprehension test (McNamara et al., 2023).

Critically, constructive learning activities not only beneficial learning, but they can be mined for insights into the learning process. For example, readers’ self-explanations can identify learners’ attentional focus, constructed inferences, and knowledge sources. As mentioned, comprehension strategies, such as creating bridging and elaborative inferences, indicate good mental model building. NLP algorithms can reliably detect these comprehension strategies (e.g., Magliano et al., 2016), and recent additions of LLMs increase the accuracy and generalizability of these detectors (Nicula et al., 2023). Leveraging technology also allows including student affect and motivation in the learner model. For example, researchers have used behavioral data (keystrokes, eye-tracking); NLP-based sentiment analysis of students’ responses can detect confusion and boredom (e.g., Allen et al., 2016). Modeling and detecting students’ learning helps provide increasingly tailored feedback.

Providing tailored feedback at multiple levels. Although feedback is important, the quantity, quality, and timing of feedback affect learning and achievement (Wisniewski et al., 2020). Thus, abundant ongoing work exploring how different types of feedback can be implemented at scale and how they might impact learning in authentic contexts. These levels can be broadly categorized as: feedback messages about the current task, tailored strategy prompts and scaffolds, and task adaptivity.

Providing Feedback Messages . First, AI-based systems can provide feedback messages about students’ performance on a particular task. Corrective feedback (i.e., correct/incorrect) is relatively easy to provide, but is less effective for complex learning tasks than higher information feedback that offers suggestions and strategies on how to improve (Wisniewski et al., 2020). AI tools that go beyond corrective feedback, either through offering constructive feedback directly to learners or by recommending feedback messages for instructors to use, have effects on immediate performance and long-term learning (McCarthy et al., 2022).

Strategy Adaptivity. A second-way researchers are exploring personalization is through identifying when to prompt and scaffold different comprehension strategies. While constructive strategies are generally useful, they are not one-size-fits-all. Some students need scaffolds to effectively use constructive strategies (McNamara & Kendeou, 2022). Conversely, over-scaffolding students who already understand the content can hinder their learning (Gerjets et al., 2006). In addition, ongoing work is examining not only which strategies work best for which learners, but also how different combinations of strategies (e.g., explaining then retrieval) might differentially impact learning (e.g., Roelle et al., 2023). This level of customization is largely impossible for a single teacher to provide, whereas such efforts would be relatively simple for an AI model to generate. Thus, this is a growing area of exploration and experimentation.

Task Adaptivity . Finally, adaptivity can involve changing the task itself. Traditionally, all students move through modules in the same order. Students who are struggling may become discouraged, and students who are excelling may become bored. Instructional adaptivity addresses this concern by allowing students to progress through content at their own pace and in a sequence that best aligns with their strengths and weaknesses. For example, intelligent textbooks use stealth and formative assessment to construct an understanding of what a learner knows and use that data to direct a student to a particular module that best suits their current needs for knowledge building. The types of feedback these tools can employ and data about their efficacy are still emerging, but recent instructional adaptivity yields more efficient learning and students generally find the recommendations to be helpful and encouraging (see Jiang et al., 2022).

Effective adaptivity requires understanding features of the task. NLP can evaluate texts in a variety of ways, such as identifying key content or evaluating text complexity. Indeed, AI has greatly improved the ability to identify what features of text are involved in comprehension. Although traditional readability metrics are commonly used in classrooms, they are not particularly good at predicting reading comprehension (Begeny & Greene, 2014; Shanahan, 2020). Early readability metrics like Flesch-Kincaid and Dale-Chall were designed to approximate reading fluency rather than comprehension and were constrained to low-level features of text (e.g., number of syllables, sentence length) that were relatively easy to compute. Although such features affect reading, these formulas do not take into account discourse-level features such as the connectedness between words and ideas (e.g., cohesion) that support mental model building. The advent of sophisticated NLP tools that can capture linguistic features such as semantic overlap has allowed for the creation of new readability metrics that can provide a more accurate and contextualized evaluation of text complexity that afford more effective adaptivity (Choi & Crossley, 2022; Crossley et al., 2017, 2023).

Generating Personalized Content. One current limitation in the ability to adapt to individual learners is that many AI-based tools rely on a library of pre-existing materials (e.g., texts, questions, prompts). Generative AI has the potential to allow for the creation of new content tailored to each student's individual needs and interests. Imagine a student completing a complex reading and writing task in which the texts are simplified on-the-fly and matched to the interests and extant knowledge. The system prompts a comprehension strategy when needed and provides adaptive support to help the student use the strategy. Such adaptivity would yield a powerful, theoretically supported tool. However, AI generation is still in its early stages, and using generative tools without a deep understanding of how they work and careful evaluation can lead to further problems. When ChatGPT first launched, many people were alarmed to see it providing made-up facts and citations, but such so-called hallucinations highlight that generative tools are designed to produce “human-like language,” not necessarily facts (O’Brien, 2023). In addition, AI is already prone to algorithmic bias, and generative AI seems to be raising the stakes further. Recent headlines highlight how generative AI can exacerbate inequities in both obvious (Silva, 2022) and more insidious ways (CBS News, 2023; Small, 2023). Given these growing concerns, legislators are developing policies (whitehouse.gov, 2023) and researchers are raising issues of Fairness, Accountability, Transparency, Ethics (FATE) in AI and technology (Memarian & Doleck, 2023). Because generative AI is relatively new, much research remains to be done to understand the costs and benefits of AI technologies and research; policy will need to be frequently revisited and revised as technology progresses.

Insights Across Contexts and Learners

In addition to implementation, AI is also being used for discovery.

Using Big Data to Understand Individuals and Contexts. Much of the robust body of research on the science of learning and reading comprehension comes from lab-based experimental studies. Although rigorous, this work uses a limited set of text materials with small, homogeneous samples and often fails to replicate across studies or generalize to ecological settings (Bryan et al., 2021; McNamara et al., 2022). Leveraging the scalability of AI may help to better capture the complexity across texts, tasks, and readers.

The utility of big data is perhaps most obvious in the case of individual differences. Small sample sizes mean that researchers are limited to examining a few individual differences at a time, which ignores critical interactions and intersectional identities (McNamara et al., 2022). Larger data sets treat an individual's skills and experiences as valuable parts of their learning, rather than noise, which affords the development of better algorithms that are less prone to biases and support more truly personalized learning.

AI can also help investigate more contexts. Research and development in reading comprehension and educational technologies have largely focused on K-12 and higher education learners. There is growing demand for instruction and tools to address adult education and re/upskilling (World Economic Forum, 2023). While there will likely be some generalizable factors and processes, it is critical to explore the unique strengths and needs of adult learners (McCarthy & McNamara, 2021). More research is needed to explore whether these approaches generalize across populations and domains and what modifications can be made to better align AI-supported research and implementation into different contexts.

Advances in LLMs have allowed for more efficient analysis of big data. Previously, if a researcher wanted to build an algorithm to predict a rubric score or detect a behavior, they would need to provide the computer with a large set of pre-labeled (i.e., human scored) data. By contrast, LLMs can teach themselves with relatively limited human involvement, making it feasible to analyze data of this scale.

New Methods of Analysis. AI not only makes faster work of analyses, but it also provides new ways of examining data. For example, Allen and colleagues (2016) tested the assumption that constructing inferences was central to comprehension using NLP. Borrowing from the logic that texts that are more cohesive yield better comprehension, they demonstrated that greater cohesion in a student's explanation was evidence of a coherent mental model as shown by comprehension test performance. These results are novel evidence of an assumed theoretical relation. NLP is becoming increasingly accessible—researchers can easily implement ChatGPT into their analysis pipeline through an API and there are powerful, but easy-to-use NLP tools for both Python and R (e.g., natural language toolkit, or NLTK) as well as downloadable and web-based tools that require no programming skills at all (e.g., Choi & Crossley, 2022; Crossley et al., 2019). AI advances gain opportunities to identify new patterns and to interrogate them in ways that can shed light on understanding of the complex processes involved in modern reading comprehension.

Rapid, Collaborative Research. Implementing AI-based technologies into learning environments suggests a need to reconsider how scientists do research and development. Lab findings do not always translate to the classroom, and many AI-supported technologies often fail to deliver. In addition to the inherent noise of a real-world environment, tools often fail to make an impact because researchers and developers rarely get input from end users (McCarthy & McNamara, 2021; Yan et al., 2023). As a result, technologies may end up underutilized because they do not fit the needs of an authentic learning context. Such concerns highlight the need for (a) co-design in which researchers, developers, educators, and users work together throughout the design and evaluation process (Roschelle et al., 2006), as well as (b) supplementing lab-based studies and randomized control trials with more agile, theory-driven experimentation, and iterative refinements (McCarthy & McNamara, 2021; McNamara et al., 2022). As learning and learning environments become more complex, there is a greater need to develop and incentivize partnerships that draw upon the expertise and experience of all those involved.

Implications for Policy

The ongoing work described above highlights several policy implications. Policymakers, developers, and educators should recognize that AI is an emerging area, but there are ample research-backed approaches to reading comprehension that are showing promise. Educators and developers can learn more about how engaging readers in constructive activities that are personalized to their skills, knowledge, and interests can support learning and how AI can support these efforts. Researchers can leverage big data and AI approaches to analysis to better identify what personalizable factors are most important to evaluate and respond to. Policymakers should further incentivize co-design and collaboration across researchers, developers, and educators so that AI-based reading comprehension supports are designed in ways that meet the needs of the modern reader in authentic learning contexts.

All stakeholders should recognize that AI can support equity through greater access to personalized instruction, but that AI can also increase inequities (e.g., perpetuating biases). Thus, researchers, developers, educators, and policymakers all must continually critically evaluate AI-based tools to ensure help rather than harm.

Conclusions

Rapid advances in technology and AI are changing the ways people interact with information, but they also afford opportunities to address the challenges they present. Research shows that readers are knowledge builders who need opportunities to engage in constructive activities that are personalized to their skills, knowledge, and interests. AI offers opportunities to deliver these types of activities and to collect and analyze learner data that can enhance the efficacy and equity of such activities. Understanding and supporting the needs of the modern reader requires consideration of the modern context. AI is part of that context. Although generative AI technologies are only a few years old, they are already revolutionizing the way that people interact with information, and new conceptions of the modern reader will likely follow. This rapid shift highlights that the speed of this important work will require more collaborative and iterative research and development.

Footnotes

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

ORCID iD

Kathryn S. McCarthy

Notes

References

ACT. (2018). The condition of college and career readiness. Retreived from http://209.106.83.28/schools/_DISTRICT_/ACT/ACT%20National%20Reports/National-CCCR-2018.pdf.

Allen

L. K.

Jacovina

M. E.

McNamara

D. S.

(2016). Cohesive features of deep text comprehension processes. In Trueswell

Papafragou

Mirman

(Eds.), the 38th annual meeting of the cognitive science society: Recognizing and representing events, CogSci 2016 (pp. 2681–2686). The Cognitive Science Society.

Allen

L. K.

Mills

Jacovina

M. E.

Crossley

D'Mello

McNamara

D. S.

(2016, April). Investigating boredom and engagement during writing using multiple sources of information: the essay, the writer, and keystrokes. In Proceedings of the Sixth International Conference on Learning Analytics & Knowledge (pp. 114–123). Association for Computing Machinery.

Baffour, P. (2023). AI can grade a student essay as well as a human. But it cannot replace a teacher. Retrieved from https://www.the74million.org/article/ai-can-grade-a-student-essay-as-well-as-a-human-but-it-cannot-replace-a-teacher/

Barzilai

Strømsø

H. I.

(2018). Individual differences in multiple document comprehension. In Braasch

J. L. G.

Bråten

McCrudden

M. T.

(Eds.). Handbook of multiple source use (1st ed., pp. 99–116). Routledge.

Begeny

J. C.

Greene

D. J.

(2014). Can readability formulas be used to successfully gauge difficulty of reading materials? Psychology in the Schools, 51(2), 198–215. https://doi.org/10.1002/pits.21740

Britt

M. A.

Durik

Rouet

J. F.

(2022). Reading contexts, goals, and decisions: Text comprehension as a situated activity. Discourse Processes, 59(5-6), 361–378. https://doi.org/10.1080/0163853X.2022.2068345

Bryan

C. J.

Tipton

Yeager

D. S.

(2021). Behavioural science is unlikely to change the world without a heterogeneity revolution. Nature Human Behaviour, 5(8), 980–989. https://doi.org/10.1038/s41562-021-01143-3

CBS Interactive. (2023). Stanford study indicates AI chatbots used by health providers are perpetuating racism. CBS News. Retreived from https://www.cbsnews.com/sanfrancisco/news/ai-chatbots-are-supposed-to-improve-health-care-but-research-says-some-are-perpetuating-racism-2/

10.

Cervetti

G. N.

Pearson

P. D.

Palincsar

A. S.

Afflerbach

Kendeou

Biancarosa

Berman

A. I.

(2020). How the reading for understanding initiative’s research complicates the simple view of reading invoked in the science of reading. Reading Research Quarterly, 55, S161–S172. https://doi.org/10.1002/rrq.343

11.

Chi

M. T.

Wylie

(2014). The ICAP framework: Linking cognitive engagement to active learning outcomes. Educational Psychologist, 49(4), 219–243. https://doi.org/10.1080/00461520.2014.965823

12.

Choi

J. S.

Crossley

S. A.

(2022). Advances in Readability Research: A New Readability Web App for English. In 2022 International Conference on Advanced Learning Technologies (ICALT) (pp. 1–5). IEEE.

13.

Connor

C. M.

Morrison

F. J.

(2016). Individualizing student instruction in reading: Implications for policy and practice. Policy Insights from the Behavioral and Brain Sciences, 3(1), 54–61. https://doi.org/10.1177/2372732215624931

14.

Crossley

Heintz

Choi

J. S.

Batchelor

Karimi

Malatinszky

(2023). A large-scaled corpus for assessing text readability. Behavior Research Methods, 55(2), 491–507. https://doi.org/10.3758/s13428-022-01802-x

15.

Crossley

S. A.

Kyle

Dascalu

(2019). The tool for the automatic analysis of cohesion 2.0: Integrating semantic similarity and text overlap. Behavioral Research Methods, 51(1), 14–27. https://doi.org/10.3758/s13428-018-1142-4

16.

Crossley

S. A.

Skalicky

Dascalu

McNamara

D. S.

Kyle

(2017). Predicting text comprehension, processing, and familiarity in adult readers: New approaches to readability formulas. Discourse Processes, 54(5-6), 340–359. https://doi.org/10.1080/0163853X.2017.1296264

17.

Dunlosky

Rawson

K. A.

Marsh

E. J.

Nathan

M. J.

Willingham

D. T.

(2013). Improving students’ learning with effective learning techniques: Promising directions from cognitive and educational psychology. Psychological Science in the Public Interest, 14(1), 4–58. https://doi.org/10.1177/1529100612453266

18.

Fiorella

Mayer

R. E.

(2016). Eight ways to promote generative learning. Educational Psychology Review, 28, 717–741. https://doi.org/10.1007/s10648-015-9348-9

19.

Gerjets

Scheiter

Catrambone

(2006). Can learning from molar and modular worked examples be enhanced by providing instructional explanations and prompting self-explanations? Learning and Instruction, 16(2), 104–121. https://doi.org/10.1016/j.learninstruc.2006.02.007

20.

Goldman

S. R.

(2012). Adolescent literacy: Learning and understanding content. The Future of Children, 22, 89–116. https://doi.org/10.1353/foc.2012.0011

21.

Graham

Hebert

(2011). Writing to read: A meta-analysis of the impact of writing and writing instruction on reading. Harvard Educational Review, 81(4), 710–744. https://doi.org/10.17763/haer.81.4.t2k0m13756113566

22.

Hattan

Alexander

P. A.

Lupo

S. M.

(2023). Leveraging what students know to make sense of texts: What the research says about prior knowledge activation. Review of Educational Research, 00346543221148478. https://doi.org/10.3102/00346543221148478

23.

Jiang

(2022). Recent advances in intelligent textbooks for better learning. In Niemi

Pea

R. D.

(Eds.), AI in learning: Designing the future (pp. 247–261). Springer.

24.

Kendeou

McMaster

K. L.

McNamara

D. S.

Wilke

B. C.

(2023). Literacy. In Schutz

P. A.

Muis

K. R.

(Eds.), Handbook of educational psychology (pp. 553–575). Routledge.

25.

Kim

M. K.

McCarthy

K. S.

(2021). Improving summary writing through formative feedback in a technology-enhanced learning environment. Journal of Computer Assisted Learning, 37(3), 684–704. https://doi.org/10.1111/jcal.12516

26.

Lee

C. D.

(2023). We ask students what they understand, not how they understand: Making reasoning comprehension processes visible and explicit. The Reading Teacher. https://doi.org/10.1002/trtr.2255

27.

Magliano

J. P.

McCrudden

M. T.

Rouet

J. F.

Sabatini

(2017). The modern reader: Should changes to how we read affect research and theory?. In Schober

Rapp

D. N.

Britt

M. A.

(Eds.), The Routledge handbook of discourse processses. Routledge.

28.

Magliano

J. P.

Ray

Millis

K. K.

(2016). The reading strategy assessment tool: a computer-based approach for evaluating comprehension processes during reading. In Crossley

S. A.

& McNamara

D. S.

(Eds.), Adaptive educational technologies for literacy instruction (pp. 282–287). Routledge.

29.

McCarthy

K. S.

Crossley

S. A.

Meyers

Boser

Zampieri

(2022). Toward more effective and equitable learning: Identifying barriers and solutions for the future of online education. Technology, Mind, & Behavior, 3(1). https://doi.org/10.1037/tmb0000063

30.

McCarthy

K. S.

McNamara

D. S.

(2021). The multidimensional knowledge in text comprehension framework. Educational Psychologist, 56(3), 196–214. https://doi.org/10.1080/00461520.2021.1872379

31.

McCarthy

K. S.

McNamara

D. S.

(2022). Knowledge: A fundamental asset. In Tierney

Rizvi

Ercikan

(Eds.), International Encyclopedia of Education (4th Ed.). Elsevier.

32.

McNamara

D. S.

(Ed). (2007). Reading comprehension strategies: Theories, interventions, and technologies. Psychology Press.

33.

McNamara

D. S.

Arner

Butterfuss

Fang

Watanabe

Newton

Roscoe

R. D.

(2023). iSTART: Adaptive Comprehension Strategy Training and Stealth Literacy Assessment. International Journal of Human–Computer Interaction, 39(11), 2239–2252. https://doi.org/10.1080/10447318.2022.2114143

34.

McNamara

D. S.

Arner

Butterfuss

Mallick

D. B.

Lan

A. S.

Roscoe

R. D.

Baraniuk

R. G.

(2022). Situating AI (and big data) in the learning sciences: Moving toward large-scale learning sciences. In Artificial Intelligence in STEM Education (pp. 289–308). CRC Press.

35.

McNamara

D. S.

Kendeou

(2022). The early automated writing evaluation (eAWE) framework. Assessment in Education: Principles, Policy & Practice, 29(2), 150–182. https://doi.org/10.1080/0969594X.2022.2037509

36.

McNamara

D. S.

Magliano

(2009). Toward a comprehensive model of comprehension. Psychology of Learning and Motivation, 51, 297–384. https://doi.org/10.1016/S0079-7421(09)51009-2

37.

Memarian

Doleck

(2023). Fairness, Accountability, Transparency, and Ethics (FATE) in Artificial Intelligence (AI), and higher education: A systematic review. Computers and Education: Artificial Intelligence, 5, 100152. https://doi.org/10.1016/j.caeai.2023.100152

38.

National Academies of Sciences, Engineering, and Medicine. (2018). How people learn II: Learners, contexts, and cultures. National Academies Press.

39.

Nicula

Dascalu

Arner

Balyan

McNamara

D. S.

(2023). Automated Assessment of Comprehension Strategies from Self-Explanations using LLMs. Information, 14(10), 567. https://doi.org/10.3390/info14100567

40.

O’Brien

(2023). Chatbots sometimes make things up. Is AI’s hallucination problem fixable? AP News. Retrieved from https://apnews.com/article/artificial-intelligence-hallucination-chatbots-chatgpt-falsehoods-ac4672c5b06e6f91050aa46ee731bcf4.

41.

OECD. (2013). The survey of adult skills: Reader’s companion. OECD.

42.

Perin

D. (

2020). The Wiley Handbook of Adult Literacy. John Wiley & Sons.

43.

Roelle

Endres

Abel

Obergassel

Nückles

Renkl

(2023). Happy together? On the relationship between research on retrieval practice and generative learning using the case of follow-up learning tasks. Educational Psychology Review, 35(4), 102. https://doi.org/10.1007/s10648-023-09810-9

44.

Roschelle

Penuel

Shechtman

(2006). Co-design of innovations with teachers: Definition and dynamics. In Barab

S. A.

Hay

K. E.

Hickey

D. T.

(Eds.), ICLS ‘06: Proceedings of the 7th International Conference on Learning Sciences (pp. 606–612). International Society of the Learning Sciences. https://repository.isls.org//handle/1/3563

45.

Shanahan

(2020). Limiting children to books they can already read: Why it reduces their opportunity to learn. American Educator, 44(2), 13–17, 39.

46.

Shermis

Burstein

Elliot

Miel

Foltz

(2016). Automated writing evaluation: A growing body of knowledge. In MacArthur

Graham

Fitzgerald

(Eds.), Handbook of writing research (pp. 395–409). Guilford Press.

47.

Shute

V. J.

Rahimi

(2022). Stealth assessment. Rapid Community Report Series. Digital Promise and the International Society of the Learning Sciences. https://repository.isls.org//handle/1/7671

48.

Silva

(2022). It took just one weekend for Meta’s new AI chatbot to become racist. Mashable. Retreived from https://mashable.com/article/meta-facebook-ai-chatbot-racism-donald-trump

49.

Simonsmeier

B. A.

Flaig

Deiglmayr

Schalk

Schneider

(2022). Domain-specific prior knowledge and learning: A meta-analysis. Educational Psychologist, 57(1), 31–54. https://doi.org/10.1080/00461520.2021.1939700

50.

Small

(2023). Black artists say A.I. shows bias, with algorithms erasing their history. The New York Times. Retreived from https://www.nytimes.com/2023/07/04/arts/design/black-artists-bias-ai.html

51.

Snow

(2002). Reading for understanding: Toward an R&D program in reading comprehension. RAND Corporation.

52.

Walkington

Bernacki

M. L.

(2018). Personalization of instruction: Design dimensions and implications for cognition. The Journal of Experimental Education, 86(1), 50–68. https://doi.org/10.1080/00220973.2017.1380590

53.

Wang

O'Reilly

Sabatini

McCarthy

K. S.

McNamara

D. S.

(2021). A tale of two tests: The role of topic and general academic knowledge in traditional versus contemporary scenario-based reading. Learning & Instruction, 73, 101462. https://doi.org/10.1016/j.learninstruc.2021.101462

54.

White House. (2023). Fact sheet: President Biden issues executive order on safe, secure, and trustworthy artificial intelligence. Retreived from https://www.whitehouse.gov/briefing-room/statements-releases/2023/10/30/fact-sheet-president-biden-issues-executive-order-on-safe-secure-and-trustworthy-artificial-intelligence

55.

Wiley

Voss

J. F.

(1999). Constructing arguments from multiple sources: Tasks that promote understanding and not just memory for text. Journal of Educational Psychology, 91(2), 301–311. https://doi.org/10.1037/0022-0663.91.2.301

56.

Wilson

Czik

(2016). Automated essay evaluation software in English Language Arts classrooms: Effects on teacher feedback, student motivation, and writing quality. Computers & Education, 100, 94–109. https://doi.org/10.1016/j.compedu.2016.05.004

57.

Wisniewski

Zierer

Hattie

(2020). The power of feedback revisited: A meta-analysis of educational feedback research. Frontiers in Psychology, 10, 3087. https://doi.org/10.3389/fpsyg.2019.03087

58.

Wolfe

M. B.

Goldman

S. R.

(2005). Relations between adolescents’ text processing and reasoning. Cognition and Instruction, 23(4), 467–502. https://doi.org/10.1207/s1532690xci2304_2

59.

World Economic Forum. (2023). The Future of Jobs Report 2023. Retreived from https://www.weforum.org/publications/the-future-of-jobs-report-2023/

60.

Yan

Sha

Zhao

Martinez-Maldonado

Chen

. & Gašević

(2023). Practical and ethical challenges of large language models in education: A systematic literature review. British Journal of Educational Technology. https://doi.org/10.1111/bjet.13370

Reading Comprehension and Constructive Learning: Policy Considerations in the Age of Artificial Intelligence

Abstract

Keywords

Tweet

Key Points

Reading Comprehension in the Digital Era: The Modern Reader

Helping the Modern Reader: Exploring the Potential of AI

Constructive Learning Activities

Personalized Instruction

Insights Across Contexts and Learners

Implications for Policy

Conclusions

Footnotes

Declaration of Conflicting Interests

Funding

ORCID iD

Notes

References