AI in Classrooms Right Now: The Good, the Messy, the Urgent

In Part 1 of this series, we asked the hard questions about what AI is and what ethical challenges it poses for Australian schools. Now let’s get specific. Because AI in education is no longer theoretical — it is happening in classrooms right now, in ways that are extraordinary, messy, promising, and at times deeply confronting.

This post looks at what AI is actually doing in schools today: the genuine benefits being documented in practice, the scenarios that are challenging everything we thought we knew about assessment and academic integrity, and the question that sits at the centre of it all — when AI can design the task, complete the task, and grade the task, what exactly is the student doing?

What AI Is Actually Doing in Australian Schools

Australia is moving faster on AI in schools than many realise. Research conducted in 2024 found that 78.2% of secondary schools were already using AI in some form. The nature of that use varies enormously — from a teacher using AI to generate a first draft of a rubric, to a school deploying a government-built AI tutoring assistant for every student.

The most significant state-level examples are worth knowing about:

Can AI Help Teachers Reach More Kids?

This is the question that genuinely excites many educators — and the evidence, while still early, is encouraging in specific areas.

AI tutoring tools can provide a form of one-on-one support that no teacher can replicate at scale. In a class of 28 students, a teacher can interact meaningfully with each child for perhaps a few minutes per lesson. An AI tutoring tool can engage every student simultaneously, adapt to their current understanding in real time, and provide immediate, individualised feedback. For students with English as an additional language, for students with learning difficulties, for students in remote schools with limited specialist support, these capabilities represent a genuine opportunity to reduce disadvantage.

AI can also free up teacher time for the work that only humans can do. When AI handles routine tasks — generating first drafts of lesson materials, creating differentiated worksheet versions, producing formative quiz questions, or providing initial written feedback on drafts — teachers can redirect their attention to relationship-building, targeted small-group instruction, and the nuanced professional judgement that is at the heart of great teaching. Our earlier posts on formative assessment and designing a performance of worth are directly relevant here — AI can support these processes, but it cannot replace the teacher’s judgement in designing tasks that are worth students’ genuine effort.

Most high school teachers who used AI in their work in the 2024–25 school year reported that these tools saved time and improved the quality of their work. That finding matters — teacher workload is one of the most significant drivers of attrition from the profession, and anything that genuinely reduces administrative burden without compromising student outcomes is worth taking seriously.

The Scenario Nobody Wants to Talk About

Here it is. The scenario that is already happening in classrooms around the world, including in Australia. A teacher uses AI to design an assessment task. A student uses AI to complete that task. The teacher uses AI to provide feedback on the submission. And somewhere in that loop, the student has done very little actual thinking.

This is not a hypothetical. Research examining the impact of generative AI on assessment has found that assessments were easily compromised using generative AI, and that many markers found it challenging to identify which student work was human-authored, which was AI-modified, and which was AI-authored. In higher grade bands, students using AI can generate work that is not only undetectable but that achieves grades their own unaided writing would not have reached. For teachers, this creates an unsettling dynamic where the marking process becomes contaminated by uncertainty — second-guessing authorship, with grade disparities emerging between markers who suspect AI use and those who do not.

In one telling study at a UK business school, markers showed unusually large grade disparities between first and second markers — with some appearing to have consciously or subconsciously marked down work they suspected was AI-generated. This is not a defensible assessment practice, but it reflects a very human response to a genuine problem of attribution in an AI-saturated environment.

The Cognitive Offloading Problem

There is a subtler risk beneath the academic integrity conversation that deserves more attention: cognitive offloading. Research is finding that students who rely on AI to complete higher-order thinking tasks may produce stronger immediate answers but develop weaker long-term learning and less independent capability. When novices lean on AI they often skip the hard metacognitive work — the struggle, the confusion, the revision — that builds genuine expertise.

A national survey of over 1,200 teens and young adults found that 41% believe AI will have both positive and negative impacts on their lives in the next decade. Those expecting negative impacts worry primarily about job displacement and AI dominance — but the educational risk of cognitive offloading is arguably more immediate. If students are using AI to avoid the hard thinking that builds neural pathways and conceptual understanding, the short-term grade gain may come at a long-term developmental cost.

This is precisely why the design of NSW’s EduChat is significant. By refusing to provide direct answers and responding instead with guiding questions, it attempts to position AI as a Socratic tutor rather than a task-completion engine. Whether that design holds up when students are stressed, time-poor, and looking for shortcuts remains to be seen.

What About Academic Integrity? The Research Is Mixed

The academic integrity question is genuinely complicated. Here is what the research actually shows:

Surveys of secondary students have found that approximately 19% of teens who are aware of ChatGPT have used it to help with schoolwork. The same research found that students consider using AI to research new topics broadly acceptable (69%), but consider using it to write an essay far less so (20%). There appears to be a student-held intuition about where AI assistance crosses a line — though that intuition is inconsistently applied under pressure to perform.

Interestingly, a study based on confidential student surveys found that cheating rates did not significantly increase in the early period of AI adoption — though a majority of teachers believe AI will negatively impact academic integrity over the next three years. The gap between what students are doing and what teachers fear they are doing is itself an important data point.

What is clear is that the solution is not solely technological. Detection tools using AI to detect AI-generated text are in an arms race with the generative tools themselves and will never be definitively reliable. The answer, as multiple researchers have concluded, is partly cultural — building a genuine school ethos around academic integrity, helping students understand the value of their own thinking, and redesigning assessments so that AI completion provides no meaningful advantage.

That last point brings us to the most important pedagogical response available to teachers right now: task design. Our post on the “so what” of learning is directly relevant here — tasks with a genuine real-world purpose, requiring personal voice, local knowledge, and specific evidence of process, are inherently more AI-resistant than generic essay prompts. The performance of worth is making a comeback.

Will AI Hurt or Help the Traditional Model of Schooling?

Both. The traditional model of schooling — teacher-delivered content, individual written assessment, grades as the primary currency of learning — is genuinely vulnerable to AI disruption. If the goal of education is to produce a correct written response, AI can do that faster and often better than a student can. That should prompt a serious rethink of what we’re assessing and why, not a defensive reaction against the technology.

But the elements of schooling that AI cannot replicate — human relationships, mentorship, the socialisation of learning, the development of character, curiosity, and resilience — become more important, not less, in an AI-saturated world. The teacher who knows a student’s story, who notices when something is wrong, who believes in a child’s potential before the child believes it themselves — no algorithm does that. And that is where the teaching profession should be investing its energy and its advocacy.

In Part 3 of this series, we’ll look forward: what could an AI-informed education system look like if we designed it well? What would we keep, what would we redesign, and what would we fight to protect? Stay tuned.