With so much changing so fast, what new developments in assessment at the classroom and school level bear watching? In the final post in this four-part series, Adelaida Kim and Thomas Hatch provide examples of “micro-innovations” in assessment that leverage AI and new technologies to support the development of specific skills and abilities in different subjects and levels. Part one provided an overview of some uses of AI in both large-scale standardized tests and classroom-based assessments. Part two described some of the new platforms, apps, and tools that teachers can use to create, analyze, and score assessments, particularly those that support more student-centered learning. Part three took a deeper look at and compared the strengths and weaknesses of the assessment capabilities of selected teacher-directed AI platforms, EdTech tools, and AI “assistants” with multimodal capabilities. For related stories on AI and education, see: Can AI “ignite the mind and heart”? Stability & change in the education system in China (Part 3); Scanning the global headlines for recent news on AI, schools, and education; AI, Cellphones, Literacy, Students’ Mental Health, Political Turmoil and More: Scanning the Headlines for the Top Education Stories for 2025.
+++++++++++++++
Fueled by AI and new technologies, innovations in assessment are emerging and quickly finding their way into classrooms, educational platforms, and large-scale tests. These include a wide range of digital tools that purport to help educators improve assessments across subjects and levels (see, for example, NWEA’s 75 digital learning tools and apps that teachers can use to support formative assessment and instruction in the classroom). Beyond these generic tools, however, a number of “micro-innovations” are demonstrating how new assessments can support the development of specific skills and knowledge in different subjects and at different levels. The effectiveness of most of these new assessment tools and approaches has not yet been established, but a review of recent articles from major education news outlets provides a sense of what’s already out there. The book Artificial Intelligence and Education in the Global South also provides a glimpse of AI’s role in assessments across different subjects and contexts in developing countries.
A game-based approach for assessing linguistic skills among students with dyslexia
Dysolve is an AI-based educational platform developed to support learners with dyslexia and related language-processing differences through adaptive, game-based activities rather than traditional static assessments or interventions. The platform provides users with customized interactive verbal games; as students play, the program analyzes responses in real time and generates new games that focus on specific cognitive and linguistic skills that need further development. To increase novelty, games are only used once and then disappear. Preliminary findings indicate potential improvements in reading outcomes for some elementary-aged learners,
Read-alouds and speech recognition for assessing reading abilities
Conventional early-grade reading assessments have relied on training educators to administer one-on-one tests, but this individualized approach makes large-scale testing expensive and demanding. Newer AI-based approaches, combined with developments in speech recognition technologies, however, can be administered to many students at once while still gathering detailed information about phonemic awareness, decoding ability, and reading fluency. For example, SoapBox Labs has developed speech-recognition technology designed to interpret children’s voices and convert oral reading or spoken responses into data that can inform assessment. This approach seeks to embed assessment within routine classroom activities, such as having early elementary students read aloud while automated systems transcribe speech, analyze fluency or comprehension indicators, and generate information that may guide instructional decisions. Following its acquisition by Curriculum Associates, this technology has been incorporated into existing educational programs, including i-Ready, reflecting broader efforts to integrate voice-based analytics into literacy instruction.
Among a slew of related platforms and tools, Amira fuses voice-based AI into reading activities with a chatbot or “reading assistant.” Amira’s reading assistant begins by observing a student read for six to ten minutes and gauges the child’s ability to name letters, sound out words and combine them into phrases and sentences. From that initial screening, Amira can gather information to spot signs of dyslexia, assess phonics mastery, estimate a student’s vocabulary size, and get a handle on decoding skills and comprehension levels. Then, the program reports findings to a teacher to support differentiated instruction, estimating the literacy levels and needs of individual students.
Along with the Rapid Online Assessment of Reading (ROAR) and several other edtech-enabled approaches, Amira has been approved by states such as California and New Jersey to facilitate early screening of students for difficulties in learning to read. Some initial evidence suggests these digital tools can improve reading outcomes for some students, but with some limits. According to Ran Liu, chief AI scientist at Amira, “We’ve learned that around 25-30-minute sessions are where we see a ceiling effect.” Although Amira can also be used with Spanish-speaking students, all of these tools need to be adapted for other languages, and critical concerns include how well they will work for students with accents or speech-related challenges.
“[T]hese digital tools can improve reading outcomes for some students, but with some limits. According to Ran Liu, chief AI scientist at Amira, ‘We’ve learned that around 25-30-minute sessions are where we see a ceiling effect.’”
Automated Early Literacy Assessment in Low-Resource Languages
Although many of the most well-known AI tools focus almost exclusively on English, training AI models on locally collected speech data also enables assessments in a wide range of languages, including those that are often underrepresented in educational technology systems. To that end, the EGRA-AI project in South Africa explores whether speech-recognition models can automatically assess children’s early reading abilities in African languages such as Sepedi and isiXhosa. The system records children reading letters or words aloud and uses machine learning models to determine whether their pronunciation is correct.
AI-Supported Feedback on Student Writing in contexts with limited internet access
In addition to AI-enabled platforms like Classtime and Brisk that provide automated feedback on student performance across many subjects and levels, some tools are being designed specifically to meet the needs of educators working in areas with limited digital infrastructure and large class sizes. For example, in the AIED Unplugged initiative in Brazil, teachers photograph students’ handwritten essays and upload the images to an AI system that evaluates them using a predefined rubric. The system converts handwriting using optical character recognition and provides rubric-aligned feedback on organization, grammar, and argumentation. The system is designed for teachers to review the feedback before returning it to students.
A game-based approach to assessment in physics
Although many digital tools focus on foundational skills like reading, NWEA (Northwest Evaluation Association) has been working with the game maker Filament Games to develop a digital assessment that examines middle school students’ understanding of Newton’s Second Law of Motion. The assessment uses a collaborative 3D simulation in which pairs of students adjust variables, such as mass and force, to synchronize virtual vehicles, enabling the system to capture evidence of scientific reasoning during problem-solving. The assessment – Distance Dash – is a game-based, digital assessment delivered through the Roblox platform. In the game, “students pick a skateboard, a bike, a grocery cart, or an automobile, load each with different items, then collaboratively fine-tune the forces placed on them. The whole time, the game covertly measures several objectives, including whether students understand the principles of acceleration and how to apply optimal force.” The project is part of broader efforts to explore game-based environments as potential spaces for classroom assessment, while also raising questions about how such platforms shape collaboration, engagement, and the measurement of student learning.
AI-Supported Preparation for High-Stakes Examinations
In Liberia, AI-driven chatbots are being piloted to help secondary school students prepare specifically for the West African Examinations Council (WAEC) exams. These chatbots simulate exam-style questions, provide explanations of correct answers, and offer targeted practice in areas where students struggle. Although they are primarily designed as learning tools, such systems implicitly rely on continuous formative assessment: the chatbot must infer a student’s level of understanding in order to generate appropriate questions and feedback.
AI-Based Measurement of Collaborative Problem-Solving
AI is also being used to measure competencies that are difficult to assess through traditional tests. In the ACTNext “Crisis in Space” pilot, students work together in a multiplayer game environment to solve a simulated emergency scenario. AI systems analyze students’ dialogue and interactions—such as turn-taking, information sharing, and problem-solving strategies—to assess collaborative skills. Rather than evaluating a single correct answer, the system analyzes patterns in communication and teamwork. This approach reflects broader efforts in educational measurement to assess 21st-century competencies, including collaboration and collective problem-solving, using AI-driven analysis of behavioral data
Oral assessments in higher education.
With students’ use of AI tools raising concerns about cheating, many educators are looking for alternatives to essays, research reports, and other written assessments. Those concerns, in turn, have fueled a renewed interest in oral assessments. For instance, Panos Ipeirotis, a professor at New York University’s business school, noticed that many students in his data science class were unable to discuss or defend their own written work when called upon. As Ipeirotis put it, “If you cannot defend your own work live, then the written artifact is not measuring what you think it is measuring.”
“If you cannot defend your own work live, then the written artifact is not measuring what you think it is measuring.” – Panos Ipeirotis
In response, he and a colleague built an “AI examiner” using conversational speech technology that probed students’ thinking about their capstone projects and that could question students about one of the cases discussed in class. With a class of over 30 students, a single instructor could not conduct such individualized oral exams, but Ipeirotis’ AI agent assessed 36 students for about 25 minutes each over nine days. Ipeirotis even had ChatGPT, Claude, and Gemini score students’ responses and asked them to review the scores and generate a final grade, with Claude designated as the “chair” of the panel to synthesize the decisions. Student reactions were mixed — most found the format more stressful than traditional written tests, but many also acknowledged it was a more authentic measure of their understanding
Large-scale developments in assessment that might eventually be used at the classroom and school level
Efforts to develop digital and multimodal assessment tasks for large-scale assessments may also provide models for how educators might assess capacities not normally measured in conventional assessments. Notably, the Program for International Student Assessment (PISA) has introduced a Creative Thinking assessment to examine the creative capacities of 15-year-old students in more than 60 countries. This assessment differs from conventional standardized measures by incorporating interactive digital tasks, including opportunities for students to produce drawings and respond to open-ended prompts with multiple possible solutions rather than a single correct answer. In addition to documenting creative expression, the assessment reflects broader efforts to capture dimensions of student learning that extend beyond strictly cognitive outcomes, such as engagement, flexibility in problem solving, and responsiveness to novel tasks. These kinds of tasks provide a model for other multi-modal assessments of competencies that may eventually find their way into tools and technologies available to educators.
Similarly, with the concerns about the authorship of college applicants’ personal essays, tech-enabled video interviews may become a more common part of the college application process. Even before the AI boom, companies like InitialView were developing platforms that allowed students to record video interviews they could send with their applications to a number of US colleges. Particularly popular with international applicants, the platform had, by 2014, attracted over 17,000 applicants from China. In the last few years, the company has developed a related application, VIVA, in which students upload a research paper or project, and an AI agent then generates a series of questions about their project that students answer during a recorded video interview. Originally offered by Caltech to accompany research papers that some applicants submitted as application supplements, the same video-based approach could be used to support oral defenses and classroom presentations in both K-12 and higher education.
Questions for the future
These examples illustrate some of the ways AI and other digital tools may help educators conduct assessments tailored to specific skills, competencies, and subjects at different levels and in more- and less-developed contexts. At the same time, these examples demonstrate how quickly things are changing and how difficult it is to obtain detailed information about the effectiveness and impact of most AI-related developments. In this context, many educators and students may find themselves using these AI and technology tools without really knowing:
- Which of these tools is most likely to improve learning?
- For which students?
- Under what conditions?
- At what cost to professional judgment, student agency, and curricular coherence?
Under these conditions, AI developers and researchers have a responsibility to work with educators to keep these critical questions at the forefront of their collective work.
International Education News 