Why Great Educators Can't Be Automated

1. Clear steps don't help if you haven't taught the idea first

In my own classes, I have seen this happen time and again: you can break a difficult maths or science question into clear, numbered steps, but if the student hasn't been taught the core idea first, they can't even follow step one.

Giving a student a five-step checklist does not help if they do not understand what each instruction actually means. If a student does not understand why we differentiate, telling them to 'find the derivative' is not a helpful clue; it is just another phrase to be confused by.

This is where automated apps fall short. An AI gives a student a problem and offers automated hints as they go. But when a student is genuinely stuck, they do not need more hints on step three. They need an experienced educator to stop, take a step back, and walk through three things in order:

• Explain the idea simply: making sense of the concept before asking the student to solve anything.
• Think through it aloud: modelling the approach on a real example, so the student sees why each choice is made.
• Practise the steps together: only once the underlying idea is secure do guided steps become genuinely useful.

As cognitive scientists Paul Kirschner, John Sweller and Richard Clark demonstrated, novices cannot discover complex methods through interactive hints alone; their working memory becomes quickly overwhelmed. An automated app can give you steps to follow, but it takes an experienced educator to build the understanding so you can actually follow them.

2. The 'Socratic AI' trap and the illusion of understanding

When an AI withholds the answer and offers step-by-step questions, students quickly learn to play the system. Instead of stopping to think deeply, a student who is struggling will type 'I don't know', ask for another hint, or make random guesses until the computer narrows the question down to an obvious choice.

In education research, this is known as 'funneling' (Bauersfeld, 1980). A sequence of bite-sized questions guides the student through each tiny part of a problem. The student answers each mini-question correctly and feels like they get it. But on an exam, without the computer feeding them clues, they are completely stuck. They were guided through the calculation, but they never had to figure out how to start on their own.

This is not just theory; it has been tested directly in large classroom studies:

• In a 2024 study led by researchers at Wharton and Stanford (Bastani et al., published in the Proceedings of the National Academy of Sciences), high school maths students who used standard AI to study did well during practice, but scored 17% worse on unassisted tests than students who had never used AI at all.
• Even when the researchers used a guided, step-by-step AI designed to withhold answers, it still produced no real improvement on independent tests compared to ordinary practice.

The researchers found that automated hints create an illusion of competence: because the computer makes each step feel easy, students think they understand the material, but because they never had to struggle through it themselves, the knowledge does not stick.

3. Why relational trust matters

In 2026, educational AI models can speak in real time, withhold answers, and hold a turn-by-turn dialogue. On the surface, it sounds like tutoring. But classroom reality reveals where simulated conversation falls flat: learning difficult things is not just an information exchange; it is an emotional and social experience.

When a problem gets genuinely tough, a student will casually close an app, guess flippantly, or click for hints to escape the discomfort. An algorithm cannot hold a student accountable, because there is no social or personal stake.

An experienced educator brings what software architecture fundamentally cannot provide:

• Reading the unspoken signals: a tutor watching a live digital whiteboard does not just listen to words. They watch the digital pen hesitate for twenty seconds, begin a calculation, and cross it out. They recognise self-doubt, second-guessing and fatigue long before a student types a word.
• Relational accountability: when a trusted mentor who has worked with a student for months says, 'I know this feels overwhelming right now, but you have the foundation to solve this. Take a breath, look at the diagram, and tell me what you see', that is not just dialogue. It is human reassurance that keeps the student in the struggle instead of giving up.
• Knowing the student over time: an educator remembers that a student rushes their algebra when they panic about the clock, or that their quick 'I get it' is actually masking a gap from three weeks ago.

A 2024 study by Stanford and Brown researchers (Tutor CoPilot) looked at over 700 tutors and 1,000 students. It found that AI on its own did not improve student outcomes. But when experienced human tutors used technology to support their live teaching, student mastery rose by 4 to 9 percentage points. Technology can simulate conversation, but it cannot care about the outcome, build genuine trust, or hold a student through the discomfort of difficult learning.

4. When lack of confidence masquerades as a content gap

In senior secondary studies, I often mentor students whose biggest hurdle is not a lack of content knowledge, but a lack of self-confidence under exam pressure.

One student recently came to a tutoring session visibly distressed after attempting a trial exam in Chemistry. Midway through the paper, they began panicking, convinced they were forgetting key concepts. Overwhelmed, they stopped the trial exam altogether and walked into our session saying: 'I don't know anything anymore. I've forgotten everything.'

An automated marking tool or an AI app would have looked at that incomplete paper and diagnosed it as a catastrophic content failure. But when I sat down with the student and we worked through those exact questions together, the real breakdown became clear:

• The distractor trap: the student had fixated on a couple of unfamiliar, technical words in the question prompt that they had never encountered before.
• The panic spiral: seeing those unfamiliar words triggered an immediate emotional response; the student assumed the question was testing advanced knowledge they had missed, leading to a complete cognitive shutdown.
• The reality: those technical terms were just contextual background. When we read the question calmly, the underlying chemistry was something the student already knew how to do, and those unfamiliar words were not even needed to answer the question.

Once we unpacked the question together, stripped away the intimidating vocabulary, and revealed the familiar chemistry underneath, the student's confidence returned, and they solved the questions with ease.

Students in middle and senior secondary school rarely struggle because of raw ability alone. Far more often, the breakdown happens in executive functioning: knowing how to structure revision, managing time during multi-step exam questions, and staying calm under pressure. Evidence for Learning's Teaching and Learning Toolkit rates metacognition and self-regulation approaches as high impact, showing that teaching students how to plan, monitor and evaluate their own thinking leads to substantial academic gains.

Psychologist Albert Bandura showed that genuine confidence is not built through easy praise or automated reassurance; it is built through earned success: facing a difficult challenge, working through the confusion, and solving it through your own effort. An algorithm cannot talk a student down from an exam panic, deconstruct the emotional trigger of an unfamiliar word, or restore the self-belief needed to sit back down and finish the paper.

5. Why families value continuity over automation

For families preparing for high-stakes milestones — such as Victorian Selective Entry, JMSS, UCAT, or senior VCE and HSC exams — the cost of superficial understanding is high. The illusion of competence created by automated hints often collapses only when SAC scores or trial exam results come in.

Educators with genuine classroom and curriculum experience bring something technology cannot replicate: a cumulative picture of how your child thinks, what their strengths are, and how they react when a problem gets tough. Continuity compounds this: a tutor who stays with the same group remembers which topic was shaky two months ago, making sure it is properly understood rather than papered over.

None of this means that experience alone guarantees results. The evidence points to what experienced educators do: diagnosing root causes, guiding genuine thinking, teaching self-regulation, and tracking progress over time. That is why EduPrism begins with an unassisted diagnostic assessment, keeps the same specialist with each small group, and monitors real progress in every single session.