Turnitin,
which processes around 200 million student submissions annually, reported
that AI-written content was detected in over 22 million pieces of student
work in the first year after it deployed AI detection tools. That figure is
likely conservative: it reflects only what detection systems identified, not
the substantially larger volume of AI-assisted work that escaped detection or
that was submitted in institutions without access to detection tools. The
question those numbers raise is not primarily about academic integrity,
though it is partly that. It is about what students are actually developing
when the cognitive work of constructing an argument, selecting evidence, and
reasoning through a problem is delegated to a language model that produces
the finished output before the understanding has had a chance to form. The
efficiency of AI-assisted homework is not in dispute. The educational value
of the process that AI has replaced is.

A 2023 report from the US
Department of Education on AI in education warned that without
clear institutional guidelines, students may rely too heavily on generative
tools in ways that undermine skill development. The concern is grounded in
established cognitive science. The processes that homework is designed to
activate, retrieval practice, elaboration, application of new knowledge to
unfamiliar contexts, are precisely the processes that AI assistance
short-circuits when it provides the output before the student has engaged
with the difficulty. A student who submits an AI-generated analysis of a
historical event has not developed the analytical capacity that struggling
with the problem was intended to build. The submission is completed. The
learning objective is not.

What the Evidence Shows About AI and Learning

The relationship between AI assistance and learning outcomes is
more nuanced than either enthusiasts or critics acknowledge. Research
published in Frontiers
in Education found that AI can genuinely enhance learning when it
is paired with structured critical thinking exercises that require students
to evaluate, critique, and extend AI-generated outputs rather than simply
submit them. In this mode, the AI functions as a first draft that the student
must interrogate and improve rather than a finished product that the student
delivers unchanged. The distinction matters enormously for whether the tool
builds or erodes the competencies that education exists to develop.

The challenge is that this productive mode of AI use requires
deliberate pedagogical design that most existing homework assignments have
not been constructed to encourage. A standard essay question offers no
structural incentive to engage critically with an AI-generated response
rather than submitting it unchanged. The redesign of assessment to make AI a
tool that accelerates learning rather than substitutes for it is
straightforward to describe and institutionally demanding to implement,
requiring curriculum revision, teacher professional development, and coherent
institutional policies that most schools and universities are still
developing. The gap between knowing what good AI-integrated pedagogy looks
like and being able to deliver it at scale is the central educational
challenge of this moment.

The Policy Response

Universities in the United Kingdom and the United States are
experimenting with a range of approaches to AI use in assessed work, from
blanket prohibition through mandatory disclosure to active integration of AI
tools into assessment design. Blanket prohibition faces obvious enforcement
challenges: AI-detection tools have high false-positive rates that create
procedural risk when students are penalised on the basis of detection alone,
and several universities have faced legal challenges from students whose
original work was incorrectly flagged as AI-generated. Several major institutions
have moved toward frameworks that require students to document their use of
AI tools and demonstrate their own understanding of the submitted work,
shifting accountability from technological detection to educational
transparency.

The US Department of Education’s report emphasised developing
clear guidelines in consultation with teachers rather than imposing them,
recognising that educators closest to the learning environment have the most
relevant knowledge of how AI tools are affecting their students in practice.
Schools that have developed workable policies have generally treated AI as a
fact of students’ working lives that must be addressed educationally rather
than eliminated technologically. As our analysis of how
AI affects different populations unequally found, the
distributional effects of AI tools in education also matter: students with
reliable access to AI assistance and the digital literacy to use it
effectively gain advantages that those without that access do not, which
means AI in education can amplify existing inequality as readily as it
challenges it.

The Assessment Question Underneath

The deeper educational question raised by AI homework tools
concerns what assessment is actually measuring. If the goal is graduates who
can synthesise information, construct arguments, and apply knowledge to
unfamiliar problems, then assessment formats that AI can complete without
those processes occurring are not measuring the right things. The response to
AI is, in this light, an opportunity to reconsider what homework is for and
what genuine evidence of learning looks like in practice.

Oral examinations, project-based assessments, in-class writing,
and evaluated real-world problem-solving are harder to delegate to AI and are
often more valid measures of the competencies that education is meant to
develop. They are also more resource-intensive to administer at scale, which
is why written homework became the dominant assessment mode in large
educational institutions in the first place. The tension between what AI
makes difficult to assess and what institutions can practically deliver
cannot be resolved by detection tools or prohibition alone. It requires
deliberate decisions about educational priorities that individual teachers,
schools, and systems need to make now, rather than continuing to defer while
the technology develops faster than the institutional response. AI has moved
the timeline forward for conversations that were already
overdue.

The Teacher’s Position

The educators who are most directly affected by AI’s arrival in
homework are also the least well-supported in managing it. Teachers who want
to redesign their assessment approaches to address AI need professional
development time, curriculum flexibility, and institutional permission to
depart from established formats. They are operating in environments shaped by
examination requirements, parental expectations, and institutional
accountability metrics that were designed for a pre-AI context and have not
been updated. The mismatch between what the evidence says effective AI-era
pedagogy looks like and what institutional structures permit is the central
obstacle to effective educational response, and it is not a problem that
technology can solve.

The teachers who have developed workable approaches to AI in their
classrooms have largely done so through individual initiative and informal
professional networks rather than through systematic institutional support.
Scaling those approaches requires the kind of sustained curriculum investment
and professional development that education systems in both the UK and the
United States have consistently struggled to fund consistently. As our
analysis of how
AI affects different populations unequally found, the people most
directly affected by AI-driven change in institutions are often those with
the least power to shape how those institutions respond. For teachers
navigating AI in their classrooms, that observation has immediate practical
significance. Our coverage of how
AI reshapes information environments and critical thinking is
directly relevant to understanding what media literacy education needs to
address. The institutional response needs to support them as the
professionals closest to the problem, not simply impose guidelines developed
without their input.

About the Author

Stuart Kerr is the Technology Correspondent for LiveAIWire. He
writes about artificial intelligence, emerging technology, and the forces
reshaping work, business, and society.