Learning modes
Eight ways to teach. Each one written down.
A learning mode is the teaching method the tutor uses. Each of the eight has written rules for how the tutor opens, questions, responds and gives feedback, and each draws on published research.
Why tutoring
One-to-one tutoring works. The challenge is doing it at scale.
In 1984 Benjamin Bloom reported that pupils tutored one-to-one did far better than pupils taught in a class, and asked how to get close to that for every learner. He called it the 2 sigma problem. Later research found a smaller but still large effect.
These are studies of tutoring and teaching methods in general, mostly with human teachers. They did not test Nested Learning.
Bloom (1984), The 2 Sigma Problem: The Search for Methods of Group Instruction as Effective as One-to-One Tutoring. Educational Researcher 13(6), 4–16.
Pupils tutored one-to-one, with regular checks and corrective feedback, scored about two standard deviations above pupils taught in a conventional class of about 30. The average tutored pupil did better than 98% of the class.
What it does not show: Small, short studies with school pupils. Later reviews found a smaller effect for human tutoring (VanLehn, 2011). Not a study of AI or of Nested Learning.
Read the source: The 2 Sigma Problem: The Search for Methods of Group Instruction as Effective as One-to-One TutoringVanLehn (2011), The relative effectiveness of human tutoring, intelligent tutoring systems, and other tutoring systems. Educational Psychologist 46(4), 197–221.
Human tutoring raised scores by about 0.79 standard deviations over no tutoring, well below two. Computer tutors that give feedback on each step of a problem came close, at about 0.76.
What it does not show: Mostly science, maths and technology, and older than today's AI. Not a study of Nested Learning.
Read the source: The relative effectiveness of human tutoring, intelligent tutoring systems, and other tutoring systemsWhich mode?
Choose by what you want learners to do
Pick the goal; the table gives the mode. Teachers choose the mode for their organisation and can change it at any time.
| If you want learners to… | Use |
|---|---|
| Test and sharpen an argument | Socratic Partner |
| Learn something new, step by step | Rosenshine |
| Rehearse a viva or oral assessment | Oral Examiner |
| Reflect on an experience or placement | Gibbs Reflection |
| Practise decisions in a realistic scenario | Simulator |
| Plan study and build confidence | Academic Coach |
| Get feedback on a draft | Academic Feedback |
| Let the tutor adapt to each learner | Adaptive Learning |
An interactive chooser, with a sample exchange for your age group and subject, is planned.
The eight modes
What each mode does, and the research behind it
Every mode teaches from your rulebook and is built not to write assessed work. Learners can type or speak in every mode, and switch read-aloud on or off.
Socratic Partner
Questions only. The learner does the explaining.
The tutor asks one question at a time, finds the claim or assumption in each answer, and asks a narrower follow-up that makes the learner justify it. It does not explain, summarise or give the answer.
Use it for: Seminar and essay preparation, testing an argument, critical thinking.
Learner: I think social media makes teenagers less happy.
Tutor: What would you need to see to tell whether social media causes it, rather than unhappy teenagers using it more?
Sample exchange. Illustration with synthetic data. Research basis
Chi and Wylie (2014), The ICAP framework: linking cognitive engagement to active learning outcomes. Educational Psychologist 49(4), 219–243.
Learning rises as learners move from listening, to doing, to building their own explanations, to dialogue in which each answer builds on the other's reasoning.
What it does not show: A framework built from earlier studies; some later experiments have not found the same order.
Read the source: The ICAP framework: linking cognitive engagement to active learning outcomesRosenshine
Explicit teaching in small steps, with frequent checks.
The tutor asks what the learner already knows, models only the smallest useful step, asks them to apply it, then checks understanding before moving on. It corrects one misconception at a time.
Use it for: Learning something new, building foundations, revising a topic step by step.
Learner: Can you teach me how to balance chemical equations?
Tutor: One small step first. In H₂ + O₂ → H₂O, count the oxygen atoms on each side. What do you get?
Sample exchange. Illustration with synthetic data. Research basis
Rosenshine (2012), Principles of Instruction: Research-based strategies that all teachers should know. American Educator 36(1), 12–19.
Ten principles, including a short review at the start, new material in small steps, many questions, worked examples, guided practice and frequent checks for understanding.
What it does not show: A summary rather than a new experiment; some principles are broadly stated.
Read the source: Principles of Instruction: Research-based strategies that all teachers should knowOral Examiner
A formal viva: one question, a clear judgement, a follow-up.
The tutor takes control as an examiner, asks one question at a time and judges each answer as strong, partial or weak against the course guidance. Every five answers it gives a progress review with strengths, weaknesses and a development target.
Use it for: Rehearsing a viva, interview or end-point assessment; checking readiness.
Learner: Formative assessment is when you test students during the course.
Tutor: Examiner judgement: partial. Correction target: you have described timing, not purpose. Follow-up question: how does a teacher use the evidence from a formative check?
Sample exchange. Illustration with synthetic data. Research basis
Roediger and Karpicke (2006), Test-enhanced learning: taking memory tests improves long-term retention. Psychological Science 17(3), 249–255.
Students who were tested on a passage remembered more of it two days and a week later than students who spent the same time restudying it, though restudying did better on a test five minutes later.
What it does not show: Short prose passages in a laboratory, with no feedback; not a study of oral examinations or AI.
Read the source: Test-enhanced learning: taking memory tests improves long-term retentionGibbs Reflection
Reflection one phase at a time, ending in an action plan.
The tutor takes the learner through description, feelings, evaluation, analysis, conclusion and action plan, naming each phase and keeping it short. The learner supplies the experience, the interpretation and the plan.
Use it for: Reflective writing, placements, workplace learning logs.
Learner: My first team meeting as a supervisor didn't go well.
Tutor: Description first: what happened, in the order it happened?
Sample exchange. Illustration with synthetic data. Research basis
Gibbs (1988), Learning by Doing: A guide to teaching and learning methods. Further Education Unit, Oxford Polytechnic.
Sets out a six-stage reflective cycle: description, feelings, evaluation, analysis, conclusion and action plan. It is widely used in health, social care and teacher education.
What it does not show: A practical model, not a tested intervention: it structures reflection rather than proving gains.
Simulator
Safe practice for high-stakes moments.
The tutor sets a realistic scenario from the course materials and asks for a full attempt: what the learner would say or do, and why. Only then does it give feedback: a strength, the main gap against the criteria, and an encouraging close.
Use it for: Practising workplace, customer, clinical or compliance decisions.
Tutor: Scenario: a customer's order arrived damaged, and they ask for a refund outside the 14-day window. Your turn: what do you say, and why?
Learner: I'd apologise, check the policy for damaged goods, and offer a replacement first.
Sample exchange. Illustration with synthetic data. Research basis
Cook, Hatala, Brydges and colleagues (2011), Technology-enhanced simulation for health professions education: a systematic review and meta-analysis. JAMA 306(9), 978–988.
Compared with no training, simulation training was linked to large gains in knowledge, skills and behaviour, and moderate gains in patient outcomes.
What it does not show: Health professions only, and compared with no training rather than other teaching; not AI role-play.
Read the source: Technology-enhanced simulation for health professions education: a systematic review and meta-analysisAcademic Coach
Goals, confidence and study strategy, not content.
The tutor helps the learner set a goal, look at where they are, weigh options and commit to a next step, building on what they said earlier in the session. Replies are three sentences at most.
Use it for: Study planning, motivation, confidence, managing workload.
Learner: I keep leaving my assignments until the last minute.
Tutor: That pattern is common, and it can change. Think of the last time it happened: what got in the way in the week before?
Sample exchange. Illustration with synthetic data. Research basis
Hattie, Biggs and Purdie (1996), Effects of learning skills interventions on student learning: a meta-analysis. Review of Educational Research 66(2), 99–136.
Study-skills support worked best when it was taught within the subject and helped students plan, monitor and reflect on their own learning, rather than as general skills on their own.
What it does not show: Covers study-skills programmes in general; not a study of coaching by AI.
Read the source: Effects of learning skills interventions on student learning: a meta-analysisAcademic Feedback
Feedback on the learner's own draft, against the criteria.
When a learner shares their writing, the tutor gives criterion-referenced feedback first, then asks at most three questions for the next revision. It gives feedback; it does not rewrite the work.
Use it for: Feedback on drafts, reports and coursework against marking criteria.
Learner: Can you check my conclusion against the criteria?
Tutor: Your conclusion restates your findings clearly. The criteria also ask you to evaluate limitations. Which limitation of your method would a marker expect to see here?
Sample exchange. Illustration with synthetic data. Research basis
Evans (2013), Making sense of assessment feedback in higher education. Review of Educational Research 83(1), 70–120.
Feedback helps most when students understand the criteria, can act on what they are told, and take an active part in using it.
What it does not show: A review of higher-education research; not a study of AI feedback.
Read the source: Making sense of assessment feedback in higher educationAdaptive Learning
Chooses the teaching approach that fits the learner, turn by turn.
On every turn the tutor reads how the learner is doing (overwhelmed, confused, stuck, bored or curious) and switches approach to suit: smaller steps, a concrete example, a stretch task or a hint. It draws diagrams and sets worksheets from the course materials, then marks the answers.
Use it for: Independent study, mixed-ability groups, or when you are unsure which mode to pick.
- Diagrams and worksheets drawn from your course materials
- Worksheet answers marked by the tutor
- Voice controls: speak instead of typing, read-aloud on or off
Learner: There's so much in this week's reading. I don't know where to start.
Tutor: Let's take one step. Which of the three models in the reading did your lecturer spend longest on?
Sample exchange. Illustration with synthetic data. Research basis
Wood, Bruner and Ross (1976), The role of tutoring in problem solving. Journal of Child Psychology and Psychiatry 17(2), 89–100.
Introduced 'scaffolding': the tutor keeps the learner on task, simplifies it, points out what matters and demonstrates, then withdraws support as the learner succeeds.
What it does not show: Young children on a block-building task; it describes tutoring rather than measuring gains.
Read the source: The role of tutoring in problem solvingVanLehn (2011), The relative effectiveness of human tutoring, intelligent tutoring systems, and other tutoring systems. Educational Psychologist 46(4), 197–221.
Human tutoring raised scores by about 0.79 standard deviations over no tutoring, well below two. Computer tutors that give feedback on each step of a problem came close, at about 0.76.
What it does not show: Mostly science, maths and technology, and older than today's AI. Not a study of Nested Learning.
Read the source: The relative effectiveness of human tutoring, intelligent tutoring systems, and other tutoring systems
Coming soon
Try a mode yourself
A version you can try in your browser, with a choice of mode and sample course materials, is being built. Until then, book a call and we will run any mode on material like yours.
Questions
Questions about learning modes
What are the AI tutor's learning modes?
A learning mode is the teaching method the tutor uses. There are eight: Socratic Partner, Rosenshine, Oral Examiner, Gibbs Reflection, Simulator, Academic Coach, Academic Feedback and Adaptive Learning. Each has written rules for how the tutor opens, questions, responds and gives feedback.
Which learning mode should I choose?
Choose by what you want learners to do. Socratic Partner for testing an argument, Rosenshine for learning something new step by step, Oral Examiner for rehearsing a viva or assessment, Gibbs Reflection for reflective writing, Simulator for practising decisions in realistic scenarios, Academic Coach for study planning and confidence, Academic Feedback for feedback on a draft, and Adaptive Learning when you want the tutor to choose its approach turn by turn.
What is the 2 sigma problem?
It is Benjamin Bloom's 1984 finding that pupils tutored one-to-one scored about two standard deviations above pupils taught in a conventional class, and his challenge to find methods that come close at scale. Later reviews found a smaller effect for human tutoring, about 0.79 standard deviations (VanLehn, 2011). Neither study tested Nested Learning.
What does Adaptive Learning do?
It reads how the learner is doing on every turn, such as overwhelmed, confused, bored or curious, and switches to the teaching approach that suits that state. It draws diagrams and sets worksheets from the course materials, then marks the answers.
Who chooses the learning mode?
Teachers do. A teacher chooses the learning mode, and the rulebook it teaches from, for every learner in their organisation. Nested Learning switches on the modes each organisation wants when it is set up.
Is there evidence that these learning modes work?
Each mode is built on published research about a teaching method, such as retrieval practice for Oral Examiner and scaffolding for Adaptive Learning. That research studied the methods, mostly with human teachers, not Nested Learning. We do not yet publish outcome figures for Nested Learning itself.