Adaptive Learning Apps
A Tutor for Every Student
Can We Solve the 2-Sigma Problem in a Different Way?
Although the research on tutoring suggests that most students have the potential to reach high levels of achievement, Bloom’s paper was called the 2-sigma problem because he recognized that it’s simply impractical, due to costs, to tutor every student one-on-one.
So how can we replicate the profound benefits of one-to-one tutoring and mastery-based instruction for all students?
While we can't tutor everyone, technology certainly can! Adaptive learning apps are designed to operate like personalized tutors. Good-quality learning apps adapt to each student’s level, creating a personalized learning path by honing in on weaknesses and ensuring concepts are mastered before the student graduates to new material.
The learning conditions studied by Bloom were difficult to implement in 1984, but technology such as adaptive learning apps has enabled all students to have access to personalized learning.
Are Adaptive Apps as Effective as Human Tutors?
Definitely! Numerous studies have shown that adaptive learning apps (also called Intelligent Tutoring Systems, or ITS), are just as effective as human tutors at teaching students.
For instance, VanLehn (2011)’s meta-analysis of human and computer tutoring studies compared the impact of human 1:1 tutoring to ITS (called “step-based” learning in the study). He found that the impact of human tutoring on student test scores was 0.79σ while the impact of the ITS was 0.76σ, so the effectiveness of computer tutoring is nearly the same as human tutors.
Figure 2 shows the impacts of 3 different types of computer learning systems VanLehn studied, compared to human tutoring and no tutoring.
Adapted from "The Relative Effectiveness of Human Tutoring, Intelligent Tutoring Systems, and Other Tutoring Systems" by K. VanLehn, 2011, Educational Psychologist, Volume 46 (4), 197-221.
In a larger meta-analysis, Ma et al. (2014) found that there was no significant difference between learning from ITS and learning from individualized human tutoring or small-group instruction. The researchers looked at 107 study results involving over 14 thousand participants, so this meta-analysis was really thorough!
Later, Kulik & Fletcher (2015) published another meta-analysis of 50 studies that looked at the effect of computer tutoring systems compared to normal classroom instruction. They found that the adaptive learning apps raised student test scores by 0.66σ, comparable to studies on human tutoring.
How Do Learning Apps Adapt to Each Student?
To act as virtual tutors, learning apps need to adapt to the knowledge levels and performance of each student, just as human tutors do. Adaptive learning apps can:
- Give guidance and feedback to students
- Scaffold student learning
- Personalize learning paths
So how do they do this?
Although the design of each app is different, in general, a typical adaptive learning app (Figure 3) has four basic components, or modules (Alkhatlan and Kalita, 2018):
- An expert knowledge module, which stores knowledge of the subject taught
- A student module, which stores the current knowledge state of each student
- A tutoring module, which makes decisions about how to teach
- A user interface module, which interacts directly with the student
Giving Guidance and Feedback
The knowledge module stores all of the educational content: the questions, explanations, and responses. It evaluates the student’s performance by comparing the student’s answers to the correct answers encoded in the module.
The tutoring module, on the other hand, determines how to teach the student, deciding how to present the teaching material to the student and provide feedback to the student.
When a student completes a problem, there is an interaction between the knowledge module and the tutoring module. For example, if the student inputs an incorrect answer, the knowledge module will signal to the tutoring module that an incorrect answer was received. The tutoring module will then decide how to provide feedback to the student: should they be presented with a hint, a partial answer, or a fully guided solution?
Figure 3 shows how feedback flows through the modules of an adaptive learning app.
Adapted from "Intelligent Tutoring Systems: A Comprehensive Historical Survey with Recent Developments" by Ali Alkhatlan and Jugal K. Kalita, 2018, 1:1, page 5.
Scaffolding Student Learning
Scaffolding is “a kind of guided prompting that pushes the student further along the same line of thinking” (Chi et al., 2001). It is an interactive process between the student and tutor (human or app!) that allows the student to gradually take steps to perform the skill entirely by themselves.
Adaptive apps scaffold students’ learning by deciding how to present the teaching material to the student. This may be through increasingly difficult problems on the same topic, more specific hints as to how to solve a problem, or partially worked through explanations to problem solutions.
The tutoring module is the driver of scaffolding in an adaptive app, acting as an intermediary between the knowledge module (the subject material and actual questions) and the student module (which tracks what the student knows and is ready to learn).
Personalizing Learning Paths
Although feedback and scaffolding are forms of adaptivity and individualization, they are not the same as individualized task selection- this is where the real personalization of learning occurs.
Most adaptive apps create a personalized learning path for their students. The student module records the initial knowledge state of the student (usually through a placement test) and then updates that knowledge state as the student moves along their personalized path and masters the material.
The tutoring module personalizes learning paths by gathering information from the knowledge module on the students’ behavior (for example how many questions the student answered correctly on a test) and from the student module on the student’s knowledge state (for example what concepts the student has already mastered). It then determines which activity to give each student next, just as a human tutor would tell a student to watch a video, practice a problem or take a test.
Apps can act as digital tutors, which solves half of Bloom’s 2-sigma problem for us. The other part of the solution is using this tutoring technology in combination with mastery learning.
- Alkhatlan, A. and J. Kalita. (2019). Intelligent Tutoring Systems: A Comprehensive Historical Survey with Recent Developments. International Journal of Computer Applications 181(43):1-20
- Chi, M., S. Siler, H. Jeong, T. Yamauchi, & R. Hausmann. (2001). Learning from human tutoring. Cognitive Science. 25. 471-533.
- Kulik, J. & J.D. Fletcher. (2015). Effectiveness of Intelligent Tutoring Systems: A Meta-Analytic Review. Review of Educational Research 86.
- Ma, W., O. O. Adesope, J.C. Nesbit, & Q. Liu. (2014). Intelligent tutoring systems and learning outcomes: A meta-analysis. Journal of Educational Psychology, 106(4), 901–918.
- VanLehn, K.. (2011). The Relative Effectiveness of Human Tutoring, Intelligent Tutoring Systems, and Other Tutoring Systems. Educational Psychologist. 46:4, 197-221.