Chapter 3, Part 4 — AI & Assessment

Chapter 3, Part 4 — Applying AI to Education

AI & Assessment

“To know what you know and what you do not know, that is true knowledge.” —Confucius

An important third consideration of AI and the Education Triad is assessment and the power of AI to transform how learning progress is measured. While assessment is highly integrated with both teaching and learning, the process of assessment itself is one of the areas most likely to be transformed by AI in the short term.

Implications of AI in Assessment for Teaching and Learning

Utilizing AI as an assessment tool in education could arguably prove to be one of the most powerful drivers of accelerating education around the world due to its ability to integrate into more tech-based curricula, provide more timely feedback for students, and enhance the teachers’ understanding of their students’ progress more efficiently. Four of the important ways AI is already making strides in assessment—and will likely continue to make a difference—are: A) providing real-time progress monitoring (more teacher-focused); B) the opportunity for instant feedback to guide learning (more student-focused); C) improving standardized assessments; and D) supporting progress toward educational equity, especially for less advantaged student populations. All four advancements and opportunities, facilitated by AI, are explored below.

Real-time Progress Monitoring for Teachers

When an AI assessment tool called Alef was tested in the United Arab Emirates public school system, a study found that the immediate feedback and personalized instruction provided by Alef were helpful for both teachers and students (Roll et al., 2023). The study showed Alef helped students better understand their own progress. Teachers re-evaluated their students’ understanding, which allowed them to tailor their teaching style and content accordingly. This individual style of assessment is increasingly important as people come to understand “that conclusions based on statistical means can be misleading or exclusionary for learners who do not conform to ‘average’ or majority demographics” (Roll et al., 2023). So having these individual assessments can play a crucial role in capturing every student’s understanding and needs, even students who don’t fall into predetermined categories of understanding.

The goal of the Alef system and other AI assessment tools like it is to promote more “precision education.” In another study done at a Taiwan elementary school, an assessment tool called TALP (Technology-Assisted Learning Program) was used to identify students who were at risk of falling behind in the early stages of their learning development. This early detection is crucial for teachers to intervene and provide targeted support, thereby improving student engagement and learning outcomes. This study highlighted that TALP is particularly effective for students with lower academic performance, indicating the potential for tools like these to help bridge educational gaps for students falling behind (Liu, 2022).

In addition to capturing educational gaps, another goal of many of these AI assessment tools is to keep students within their zone of proximal development, a theory initially proposed by Soviet psychologist Lev Vygotsky (1896–1934) and further developed by Murray and Arroyo (2002). Vasiliy Kolchenko, associate professor at New York City College of Technology, points out in his article Can Modern AI Replace Teachers? that AI-driven educational programs are akin to computer games, designed to keep players engaged and interested, by not making the challenges too hard or too easy, but keeping the students in their zones of proximal development (Kolchenko, 2018). If the game is too easy, it can make students feel bored or disinterested; if the game is too difficult, it can cause frustration. Because of the ability to accurately assess the ability of the individual students, AI can be used as a valuable tool to give teachers updated feedback on the students’ understanding, which allows the teachers to keep the students within the zone of proximal development.

The Power of “Instant Feedback” for Students

The ability to give instant feedback should not be underestimated when looking at how AI will affect learning. Studies have shown that feedback is more effective when it is closer to the time that the work is completed (Kumar, 2023; Yang et al., 2022). Because of this, and AI’s ability to synthesize information and identify mistakes in real-time, instructors can use AI technology to provide feedback instantly, rather than after the student has reached the wrong answer (U.S. Department of Education, 2023). This works especially well in STEM courses where mistakes are more clear-cut. However, as AI improves, providing instant feedback for other subjects could prove helpful as well. In the beginning, the feedback could be automatically generated and sent to the teacher to make sure that the feedback is accurate before they send it to the student, so they are kept in the loop. But after confidence is built in the feedback systems, the feedback given to students could be almost instantaneous (U.S. Department of Education, 2023).

In a case study, Dr. A.I. Case, an adjunct faculty member affiliated with the faculty of education at two publicly funded universities, began experimenting with AI-assisted services (Kumar, 2023). He wanted to adopt this specific service to provide prompt feedback, aiming to significantly enhance the learning process. Dr. Case noted the educational benefits of timely feedback and recognized that factors like his own fatigue and the timing of assignment reviews could affect the quality and timeliness of feedback. He also believed that the consistency of AI-provided feedback could mitigate issues related to personal constraints such as time pressures and lack of rest, ultimately supporting a more stable and reliable feedback system for students.

Furthermore, for multi-step problems, hints and instant corrections can help students stay on track (U.S. Department of Education, 2023; Yang et al., 2022). For example, on a multi-step math equation, AI could provide feedback or offer hints when needed every step of the way. This allows students to get help along the way and fully understand each problem before moving to the next. Without a teacher constantly present, it’s easy to imagine many students getting a step wrong at the beginning of the sequence and struggling to finish the problem because of missing a critical concept. Supported by hints and instant feedback, AI therefore theoretically has the highly positive potential to help students master every step on their way to finishing the problem and achieving better mastery of the material in general.

More Holistic Feedback on Standardized Testing

Current standardized tests lean on using multiple-choice questions as the primary form of assessment because they are the easiest and cheapest to score. This strategy works well for measuring a test taker’s ability to find a correct answer among other choices, but it doesn’t gather much information about what the test takers actually know, how well they can synthesize information, or their ability to think critically. In a world where facts and information are increasingly available, the ability to think critically is also important (Iyer, 2019). By incorporating AI into the grading process of standardized tests that include a broader range of questions beyond multiple choice, these tests can potentially present a more comprehensive way to assess student knowledge.

Because AI can maintain consistent grading parameters, it has the potential to support more complex free-response questions that don’t have a simple a, b, c, or d answer. When training the AI models to grade the test responses, the AI models could be trained on expert graders and on a consensus of what is important. This would potentially remove the need to employ thousands of teachers grading the free-response sections according to their interpretation of the syllabus. Instead, assessments would incorporate more free-response sections graded by AI trained using exemplars from highly regarded educators.

Perils of AI & Assessment: Algorithmic Discrimination and Stunting Creativity

Algorithmic Discrimination

In 2022, the White House released a landmark statement regarding algorithmic discrimination, saying, “Designers, developers, and deployers of automated systems should take proactive and continuous measures to protect individuals and communities from algorithmic discrimination and to use and design systems in an equitable way” (The White House, 2022).

Algorithmic discrimination occurs when computer algorithms systematically and unfairly disadvantage certain groups of people based on characteristics like race, gender, or age, often as a result of biases in their design, data, or implementation. This issue is hard to identify and is most often built into systems by accident as it reflects the biases that exist in the company and in society. One example was Amazon’s machine learning recruiting algorithm that developed a bias against women. The algorithm was programmed to look at past applicants to judge current applicants’ suitability for the roles they applied for. In this case, because Amazon was hiring for a tech role where men were more frequently hired, the algorithm learned from this and graded female applicants lower than the male applicants, even ruling out applicants from all women’s colleges entirely (Iriondo, 2018). It is therefore extremely important for consumers to understand this issue and be vigilant in identifying incidences or potential incidences of algorithmic discrimination.

One recent example of algorithmic discrimination rearing its head in education was on the assessment side. In 2020, the COVID-19 pandemic forced England’s Office of Qualifications and Examinations Regulation to cancel its nationwide exams used to calculate final grades and college placement. Instead, they chose to replace it with an algorithm that would automatically determine these grades. When the final grades and scores were released, it was discovered that students from disadvantaged backgrounds were more likely to be downgraded. Private schools received more than twice as many A grades compared with students at comprehensive schools (Duncan, 2020).

When creating and training programs, these systems are built on a foundation of human knowledge, which is embedded with human and systemic biases that affect the AI outputs. Currently, because most of the tech world today is pioneered by white males, Black and Hispanic groups are sometimes underrepresented in training data and should be among the groups carefully monitored to ensure fair representation (Schwartz et al., 2022).

Another example where algorithmic discrimination was flagged was with a test monitoring program from a tech firm called Proctorio. The firm used AI to flag any test takers who might be cheating during the test. While the program was not highly accurate overall, with its best recognition rate at 75%, it failed to recognize Black faces more than half the time (Clark, 2021). This demonstrates that when implementing new technologies across educational settings, both users and creators must be vigilant about potential algorithmic biases hidden within these systems.

Recognizing and addressing embedded bias early and proactively is essential to ensure new AI technologies support equitable education for all, and do not inadvertently reinforce inequality. This conscious, informed oversight will be vital in ensuring that educational AI acts as a bridge that narrows equity gaps, rather than as a wedge that further expands them.

Conclusion

The integration of AI into education, as currently bounded, presents both exciting opportunities and critical risks. AI can personalize learning, ease administrative burdens, and potentially enable more equitable access to education. But it also introduces new concerns, including deepening the digital divide and weakening the human interactions crucial to critical thinking and social development.

Algorithmic discrimination highlights how emerging technologies can unintentionally perpetuate inequality. As a result, careful oversight, transparency, and inclusive design will be essential. There are still more questions than answers as educators, policymakers, and technologists wrestle with the changes AI brings. While the near-term future of AI in education seems clear and incremental, the long-term future remains uncertain, and perhaps more transformative than we can currently imagine.

The question of how we shape that future, one that prioritizes human flourishing as well as technological advancement, is the focus of Chapter 4.

References

Clark, R. (2021). Students of color are getting flagged to their teachers because testing software can’t see them. The Markup. https://themarkup.org

Duncan, P. (2020). Who won and who lost: When A-levels meet the algorithm. The Guardian. https://www.theguardian.com

Iriondo, R. (2018). Amazon scraps secret AI recruiting engine that showed bias against women. VentureBeat. https://venturebeat.com

Kumar, S. (2022). Faculty members’ use of artificial intelligence to grade student papers: A case of implications. Journal of Educational Technology Development and Exchange, 15(1), 1–13.

Kolchenko, V. (2018). Can modern AI replace teachers? Not so fast! Artificial intelligence and adaptive learning: Personalized education in the AI age. International Journal of Smart Education and Urban Society, 9(1), 1–10.

Liu, Y. (2021). A case study of the adaptive learning platform in a Taiwanese elementary school. Asian Journal of Education and Social Studies, 14(3), 1–9.

Murray, T., & Arroyo, I. (2002). Toward measuring and maintaining the zone of proximal development in adaptive instructional systems. International Journal of Artificial Intelligence in Education, 13(1), 1–28.

Roll, I., Wylie, R., & Brill, E. (2020). Artificial intelligence in education. Nature Human Behaviour, 4, 529–530. https://doi.org/10.1038/s41562-020-0861-6

Schwartz, A., Dobbe, R., Scheuerman, M. K., Hanna, A., Denton, E., Fried, G., Green, B., Kasy, M., & Hu, L. (2022). Towards a standard for identifying and managing bias in artificial intelligence. Communications of the ACM, 65(11), 58–66.

The White House. (2022). Blueprint for an AI Bill of Rights: Making automated systems work for the American people. https://www.whitehouse.gov/ostp/ai-bill-of-rights

U.S. Department of Education. (2023). Artificial Intelligence and the Future of Teaching and Learning: Insights and Recommendations. Office of Educational Technology. https://tech.ed.gov/ai/

Yang, J., Yu, Z., & Jin, Y. (2021). Artificial intelligence in intelligent tutoring robots: A systematic review and design guidelines. Educational Technology Research and Development, 69, 327–356. https://doi.org/10.1007/s11423-021-09960-y