AI in Education
Artificial Intelligence in Education (AIED): What It Is and What It Actually Does
AIED, or Artificial Intelligence in Education, is the research and practice field that builds software which adapts to how each learner is doing. This article explains what falls inside AIED, the four system types that actually work in classrooms, what the evidence says, and how Indian institutions are deploying it right now.
Admin ·
AIED, or Artificial Intelligence in Education, is the research and practice field that builds software which adapts to how each learner is doing. This article explains what falls inside AIED, the four system types that actually work in classrooms, what the evidence says, and how Indian institutions are deploying it right now.
What AIED means
AIED stands for Artificial Intelligence in Education. It is a named academic field, not a marketing phrase: the International AIED Society has run a peer-reviewed conference every year since 1989, and the International Journal of Artificial Intelligence in Education has been publishing since 1990. So when a paper or a job posting says AIED, it means the study and construction of educational software that models a learner and changes its behaviour based on that model.
That last part is the whole distinction. A video lecture is not AIED. A quiz that gives the same 20 questions to every student is not AIED. A system that notices you keep dropping the negative sign when moving terms across an equals sign, then serves you three targeted problems on exactly that, is AIED.
The field sits on three legs: computer science, cognitive psychology, and education research. That mix explains why AIED papers argue about pedagogy as much as about model accuracy. A recommender that raises engagement while teaching the wrong thing is a failure in this field, which is not how most consumer AI is judged.
The four systems that actually do the work
Almost every real AIED deployment is one of four things, or a combination. Knowing which one you are looking at cuts through vendor language fast.
Intelligent tutoring systems are the oldest and most studied. They hold a model of the subject, a model of the learner, and a set of teaching rules, then work through problems with a student step by step. Carnegie Learning's algebra tutor, built on decades of work by John Anderson's group at CMU, is the canonical example and is still in American schools.
Adaptive learning platforms are lighter. They sequence content and questions based on performance, without simulating the reasoning inside a single problem. Most Indian edtech that claims personalisation is this. Automated assessment covers scoring: multiple choice is trivial, short answers are workable, and essay scoring is genuinely hard but has been in production since the Educational Testing Service began using e-rater alongside human graders on the GRE in 1999. Learning analytics is the fourth, and it does not teach at all. It watches. Dropout risk flags, attendance patterns, which chapter the whole class stalled on.
Generative AI cuts across all four now. A chatbot tutor is an ITS with a language model instead of hand-written rules, which makes it far cheaper to build and far less predictable about what it teaches.
| System type | What it does | Where you meet it |
|---|---|---|
| Intelligent tutoring system | Guides a student through each step of a problem, diagnoses the specific error | Maths and physics practice, coding tutors |
| Adaptive learning platform | Picks the next lesson or question based on past performance | Most personalised-learning apps, spaced revision tools |
| Automated assessment | Scores answers, from MCQs to essays and code | Board-scale grading pilots, entrance test practice, coding platforms |
| Learning analytics | Predicts dropout and flags struggling students to teachers | University dashboards, LMS reports, government school MIS |
What the evidence supports, and what it does not
The strongest finding in AIED is old and boring: step-level tutoring works. Kurt VanLehn's 2011 review compared human tutors, step-based computer tutors, and plain answer-based systems, and found step-based tutors landed close to human tutoring and well ahead of no tutoring. That is a real effect, replicated over years, mostly in well-defined subjects like algebra, stoichiometry and programming.
The weakest claims are the broad ones. There is no good evidence that an AI system teaches essay writing, history argument or spoken fluency as well as a decent teacher. Learning styles matching, still sold as personalisation, has no research support at all. And engagement metrics are not learning: a student can spend 40 minutes in an app and learn nothing measurable.
One more thing worth being blunt about. AIED helps most where there is no teacher attention available, and India has plenty of that. It helps least where a good teacher is already doing the job. If what adaptive learning actually means is the part you want to check, that is where the mechanics get specific.
How AIED is being used in India
India is running some of the largest AIED deployments in the world by user count, mostly outside formal classrooms. WhatsApp is the delivery channel more often than any app.
The Rajasthan and Delhi government school systems have used adaptive practice through partners like Convegenius and Rocket Learning, sending short daily activities to parents' phones. NCERT and CBSE have moved on AI as a subject too: CBSE has offered Artificial Intelligence as a skill subject from Class 9 since 2019, and by 2023-24 more than 10,000 affiliated schools had signed up for it. NPTEL and SWAYAM carry AI courses for teachers. On the private side, Physics Wallah, Unacademy and Vedantu all run doubt-solving assistants, and most of them are retrieval systems over their own question banks rather than raw language models.
What is thinner is assessment. Indian boards still grade by hand, and automated essay scoring in Hindi, Tamil or Marathi is barely built because the labelled data does not exist. That gap is a real opportunity for anyone with a linguistics background and some engineering.
NEP 2020 mentions AI in education directly and pushed the creation of the National Educational Technology Forum, so the policy cover exists. Funding and teacher training are the constraints, not permission.
If you want to work in AIED
The field hires from two directions and most people underrate one of them. Engineers get in through recommender systems, NLP and knowledge tracing. Teachers, curriculum designers and psychometricians get in through content and item design, and they are scarcer. A company with a working model and no one who can write 4,000 good calibrated questions is stuck.
A practical way in: pick one narrow skill, say quadratic equations for Class 10, and build a system that diagnoses the five most common errors and responds to each. Small scope, real evaluation, something to show. That single project teaches more about AIED than any course, because you will hit the actual hard problem, which is not the model. It is writing the pedagogy down precisely enough for a machine to follow it.
The teachers who learn to specify what good teaching looks like, step by step, will be the ones building this. Not the ones who learn to prompt.
- Learn Bayesian knowledge tracing and item response theory before any deep learning method, since both are still what production systems run on
- Read two years of IJAIED or the AIED conference proceedings to see what has already failed
- Build one narrow diagnostic tutor end to end and test it on ten real students
- If you teach, document your own error taxonomy for one chapter, that document is the scarce asset
FAQs
1. Is AIED the same as edtech?
No. Edtech covers all technology used in education, including video lectures, LMS platforms and school ERP software. AIED is the narrower subset where the software builds a model of the learner and changes what it does based on that model.
2. Will AI replace teachers in India?
No, and the arithmetic explains why. India has roughly one teacher for every 26 students in elementary school, and AI systems are strongest at drill, feedback and diagnosis, not at motivation, discipline or classroom judgement. The realistic shift is teachers spending less time on grading and more on the students the system flags.
3. Which AI tools can a school actually start with?
Start with grading and administrative load, not teaching. Automated MCQ scoring, attendance and dropout dashboards inside your existing LMS, and question generation for practice sheets give measurable time savings within a term, while tutoring systems need curriculum alignment work before they help.
4. What qualifications do you need for AIED research?
Most AIED researchers hold a postgraduate degree in computer science, cognitive science, educational psychology or learning sciences. In India, IIT Bombay, IIIT Hyderabad and TISS have relevant groups, and a strong publication at the AIED conference matters more for hiring than the degree label.
5. Does AIED work for languages other than English?
Partly. Adaptive sequencing and MCQ scoring are language-agnostic, so they work fine in Hindi or Kannada. Anything involving free text, essay scoring or conversational tutoring drops sharply in quality for Indian languages because the labelled training data is scarce.
6. How do you measure whether an AIED system is actually teaching?
Compare pre-test and post-test scores on the same skill against a control group that did not use the system, and check retention a few weeks later. Time spent, streaks and completion rates tell you about engagement, not learning, and the two often move in opposite directions.
Related reading
- What Is Adaptive Learning in Artificial Intelligence?
- Intelligent Tutoring Systems: How AI Actually Teaches One Student at a Time
- A Neural Approach to Automated Essay Scoring: What the Paper Says and How to Build One
- AI in Learning Management Systems: What It Actually Does
- how can ai be used in education