The idea of using AI to support learning is not new. Early intelligent tutoring systems appeared in research labs decades ago, aiming to mimic the attention and feedback of a human tutor inside software. Over time these systems were tested in schools and universities, producing a body of evidence that they can raise achievement compared with traditional whole class teaching and generic computer-based exercises.
Decades of intelligent tutoring research show AI can quietly lift achievement beyond traditional teaching
Meta analyses of dozens of controlled evaluations typically find effect sizes around two-thirds of a standard deviation for intelligent tutoring systems compared with conventional instruction, which is roughly the difference between a student at the fiftieth percentile and one at the seventy-fifth percentile in test performance. AI’s role in job transformation is similarly noted in how it can enhance human capabilities in various fields.
For years though, these tools were confined to specific subjects such as mathematics and physics, and they usually required dedicated software, careful implementation, and institutional support. The launch of large language models accessible through simple chat interfaces created a different dynamic. Suddenly students could summon a writing coach, coding assistant, or math explainer from any device with an internet connection.
Research reviews covering generative AI in education note a surge of empirical studies beginning in 2023, driven strongly by the appearance of tools like ChatGPT and similar systems. At the same time, institutional surveys report rapid growth in the use of generative AI across education organizations, with one international study estimating that use rose from 45 percent of institutions in 2023 to 86 percent in 2024, the highest rate reported among major sectors.
How students are using generative AI right now
Student behavior has responded quickly. Surveys of American middle school, high school, and college students in 2025 show that reported use of AI for homework climbed from just under half of students to nearly two-thirds in the space of a few months.
In the same surveys, more than 70 percent of students said they had used at least one type of AI tool for school-related activities, going beyond homework help to tasks such as studying for tests and preparing presentations.
Broader international snapshots reinforce the picture of fast normalization, though with wide variation. A recent synthesis of evidence on generative AI use in education reports figures ranging from 14 percent to 67 percent of students having used generative AI for schoolwork or study, depending on country, age group, and survey method.
Some studies find that students in technically oriented programs, such as computer science and engineering, are more likely to be regular users than peers in humanities courses, which aligns with classroom reports of heavy use of AI coding assistants and debugging help.
The attitudes behind this use are complicated. Research on teenagers in the United Kingdom, for example, finds that only about one quarter regularly seek out AI tools for homework, and that many use them only when teachers explicitly recommend them.
Qualitative work in the same study shows young people divided about whether AI support feels like legitimate help or something closer to cheating. That ambivalence is increasingly echoed in university focus groups and teacher interviews worldwide, where students describe extensive AI use as both routine and morally uncomfortable.
What the evidence actually says about learning gains
Beneath the headlines there is a substantial body of rigorous work on AI-supported instruction, especially outside the newest generative systems. Meta analyses of intelligent tutoring systems consistently show that these tools outperform business as usual instruction when they are well aligned with curriculum and assessments.
One review of fifty controlled evaluations found a median effect size of 0.66 standard deviations for intelligent tutoring systems compared with conventional teaching, confirming robust positive impacts on test scores. Other syntheses report moderate gains for college students as well, with intelligent tutoring outperforming most other forms of computer-based instruction and textbook practice, though still lagging behind individual human tutoring.
More recent work examines conversational AI tutors, adaptive practice platforms, and large language model-based assistants. A review of computer-assisted and AI tutoring interventions reports that structured practice systems tend to yield small to moderate gains, while intelligent tutoring and conversational AI tutors often deliver effect sizes in the range of 0.4 to 0.8 standard deviations when compared with no tutoring at all.
These results suggest that the best AI tutors are approaching the effectiveness of small group human tutoring in certain domains, especially in mathematics and structured skills for students in the upper elementary and middle school years.
In low resource contexts, adaptive math platforms and AI-supported practice tools have been deployed to large numbers of students with promising short-term results. Studies in such settings often report notable improvements in math achievement within months when algorithmic personalization is integrated into regular lessons, particularly for learners who were previously far behind grade level expectations.
Syntheses of generative AI applications in STEM education also tend to find moderately positive effects on both achievement and engagement, though the evidence base is still building and uneven across subjects.
The key nuance from these studies is that gains are largest when AI tools are tightly coupled to clear learning objectives and assessments, and when they supplement rather than replace teacher guidance. When systems are simply bolted onto existing courses without alignment, the impact is smaller and sometimes negligible.
The risk of learning less by outsourcing more
At the same time, the very flexibility that makes generative AI attractive to students can undermine deep learning when used without structure. Survey and interview studies increasingly document patterns of students asking AI tools to complete significant portions of their work, from drafting essays to solving multistep math problems, and then submitting the output with minimal reflection or modification.
In these cases, learners may experience smoother workflows and less anxiety, but the cognitive processes of planning, argumentation, and problem-solving are effectively offloaded to the system.
Reviews of AI-supported writing instruction capture this tension clearly. Students often show improvements in grammar, formatting, and surface style when using AI writing assistants, yet evidence for gains in higher-order skills such as thesis development, critical reasoning, and originality is far less conclusive.
Teachers report that some students become adept at crafting prompts and iterating AI output, but not necessarily more skilled at constructing arguments or integrating evidence.
Concerns about dependence show up in attitudinal data as well. Studies of generative AI in education find groups of students who describe themselves as reliant on AI for both understanding content and producing assignments, and this self-described dependence is associated with lower self-reported effort and higher tolerance for practices that blur into academic dishonesty.
Research on homework use in the United States raises similar alarms, with educators worried that rising AI reliance is eroding opportunities for students to struggle productively and build critical thinking skills.
Systematic reviews of generative AI in pedagogical practice emphasize that while benefits for efficiency and access are evident, there are ongoing unresolved questions about how these tools affect creativity, metacognition, and the development of an authentic academic voice.
The technology can accelerate understanding when used as a scaffold, but it can also make it easier to bypass understanding entirely.
Implications for schools, businesses, and society
For education systems the implications are double-edged. On one hand, AI tutors and generative helpers offer a scalable way to provide individual feedback and practice at a cost that is orders of magnitude lower than human tutoring.
Analyses of AI tutoring services highlight that subscription prices in the range of a few dollars per month compare favorably with human tutoring rates that run tens of dollars per hour, making AI one of the only realistic options for universal tutoring access.
In an era of widening achievement gaps and teacher burnout, this potential is hard to ignore.
On the other hand, institutions must grapple with academic integrity, equity, and the redefinition of essential skills. Large surveys of education organizations show that generative AI is now widely adopted in back office processes and instructional support, and that many leaders expect a positive financial return on investment, with one study estimating a more than threefold ROI for every dollar spent on generative AI solutions.
Yet those same reports and related educator surveys describe mixed feelings among teachers. Many value AI for reducing administrative load and helping personalize learning, but a smaller share currently see strong evidence of improved learning outcomes, and concerns about overreliance and bias are common.
For businesses and employers, the rise of AI-assisted education raises questions about how to interpret grades and credentials in a world where assignments may be heavily mediated by algorithms. As AI tools become standard in workplaces, the expectation may shift toward knowing how to collaborate effectively with AI while still demonstrating independent reasoning and problem-solving.
That in turn should influence what schools prioritize, steering curricula toward skills that are complementary to AI such as critical thinking, ethical judgment, and the ability to design and evaluate systems rather than simply consume their output.
At a societal level, the stakes include trust in educational institutions and confidence in the meaning of expertise. If AI-supported shortcuts become the norm and assessments fail to distinguish between superficial performance and genuine understanding, then degrees and certifications risk losing their signaling power.
Conversely, if schools succeed in integrating AI transparently with clear norms, students can graduate with both stronger conceptual foundations and practical experience in using powerful tools responsibly.
How to integrate AI without hollowing out learning
The research points toward a few guiding principles for educators and policymakers, even though many details remain unsettled. Complementing this classroom-focused research, analyses of official AI policy documents from Swedish universities have introduced dynamic alignment frameworks that link strategic, pedagogical, ethical and legal, operational, and adaptive considerations to guide institutional responses to generative tools.
First, AI tools are most effective when designed and deployed as part of a coherent instructional strategy. This means aligning AI activities with explicit learning goals, using systems to provide formative feedback and extra practice rather than to generate final answers, and regularly checking whether students are actually gaining the intended knowledge and skills.
Second, trust and clarity are essential. Students need transparent guidance about what counts as acceptable AI support and what crosses into academic misconduct. That guidance should be grounded in both institutional values and the realities of modern work, where AI assistance is increasingly normal.
Surveys showing that many teenagers only use AI for homework when teachers suggest it indicate that educator framing has real influence on student choices.
Third, assessment practices must evolve. If coursework can be easily completed by AI systems, then it is reasonable to expect those tasks to be redesigned. Oral examinations, project-based work with process documentation, in-class problem solving, and iterative feedback cycles are all ways to make it harder to outsource thinking completely and easier to observe actual learning.
At the same time, simply banning AI is unlikely to work, given widespread access and the legitimate benefits it can offer.
Finally, there is a need for continued, careful research on generative AI in classrooms. Current meta-analyses are richer for intelligent tutoring systems and structured practice tools than for large language model-based assistants, and existing studies often focus on short-term outcomes in specific subjects.
Longer-term work that tracks how AI use shapes motivation, creativity, and transfer of learning across contexts will be crucial. So will studies that pay close attention to equity, examining whether AI-supported instruction narrows or widens gaps between different student groups.
Looking ahead
Generative AI in education is still in its early chapters, but the plot has already advanced far enough to show both promise and peril. There is robust evidence that well-designed AI tutors and adaptive systems can raise achievement, especially in structured domains and for students who lack access to human support.
At the same time, there is growing documentation of students leaning on AI to do the thinking for them, alongside unresolved questions about how this will affect creativity, critical reasoning, and the meaning of academic integrity.
The next few years will likely determine whether education systems treat generative AI as a partner in learning or as an unregulated shortcut. The most hopeful path involves using AI to extend the reach of good teaching while deliberately protecting space for slow thinking, struggle, and original work.
That path will demand thoughtful policies, redesigned assessments, and steady communication with students and families about what learning should look like in an AI-saturated world. It is not a simple route, but it is the one most aligned with helping learners gain both powerful tools and enduring understanding.
Conclusion
Artificial intelligence is no longer a side experiment in education. It has become part of the everyday fabric of classrooms, homework and exams, at a pace that is faster than almost any previous technology shift in schooling. The central question is no longer whether students will use AI but whether they will learn with it or quietly let it think in their place.
A Short History Of Technology Changing How Students Learn
Education has lived through waves of cognitive outsourcing before. Calculators moved arithmetic offloaded to machines. Search engines changed how students gather information, making recall less central than navigation and evaluation. Early learning management systems digitized coursework but did little to change the deeper habits of thinking.
What makes this moment different is that AI systems are now generating explanations, arguments and code, not just storing or retrieving content. Generative models can produce essays, problem sets and project plans that look like student work. That shifts the boundary between support and substitution in ways that traditional plagiarism rules and test designs were never designed to handle.
What AI In Classrooms Looks Like In 2026
Across school systems and universities, generative AI has moved from pilot projects to deliberate integration in teaching and learning environments. Many institutions are building AI directly into virtual campuses, learning platforms and course tools so that students and instructors access assistance inside the same environment where lessons and assessments live.
The scale of student adoption is already global. Recent surveys show that more than eight in ten students now report using AI in some form for their studies, with usage in universities jumping from roughly two thirds of students to well over ninety percent in a single academic year. In practical terms, this means that for many assignments, a substantial share of a class will have at least consulted an AI system before turning in work.
Teachers are using these tools as well. One major dataset from the 2024 to 2025 school year found that a large majority of both teachers and students had experimented with AI for instruction or study support. Educators report using AI for drafting lesson materials, generating quizzes and simplifying complex explanations, especially in environments where workload and burnout remain serious concerns.
Despite this rapid uptake, policy and governance are lagging behind. Roughly half of middle and high schools around the world have some form of AI policy, yet only a small minority of teachers say those policies are clear and actionable. At the higher education level, many universities still operate without a formal AI policy even as more than half of their communities believe the institution is not fully prepared to manage AI responsibly.
Students themselves are aware of the tension. Global survey data indicates that around sixty percent of students worry classmates might misuse AI for unfair advantage, with concern rising to more than seventy percent in North America. That anxiety coexists with very high daily usage, creating a climate where norms and expectations remain unsettled.
Economically, AI in education has become a substantial market. Estimates put the value of the global AI education sector at more than seven billion dollars in the mid 2020s, with projections that it could grow several fold by the end of this decade and expand even further into the 2030s. That growth reflects demand not only from schools but also from corporate training, language learning and professional upskilling platforms.
Under the surface of those numbers, several structural trends are visible. The first is a shift away from generic consumer AI tools toward platforms built specifically for education, with features such as class level administration, curricular alignment and assessment support. A second is the rise of AI agents and assistants that can act as tutors, study companions or grading aides, often operating inside established learning systems. A third is the increasing use of simulation based learning in STEM fields, where AI helps create dynamic scenarios for labs, data analysis and problem solving practice.
Delegation Without Understanding: The Risk Educators Are Worried About
When scientists warn that AI could change education forever, they are not only pointing to new features or bigger budgets. The deeper concern is cognitive. If students learn to treat AI as a shortcut for any challenging task, they risk trading understanding for convenience.
There is already evidence that many students reach first for AI when confronted with reading dense texts, writing essays or solving complex problems. Studies show that the most common uses of generative AI among middle and high school students include research support, essay editing and brainstorming. At the university level, usage specifically for assessments has surged, moving from roughly half of students to nearly nine in ten within a year. Those patterns suggest that AI is directly touching the core of graded work, not just peripheral study aids.
On its own, using AI for brainstorming or editing is not harmful. In fact, it can be excellent practice if students compare revisions and critique suggestions. The risk appears when systems are treated as black boxes. If a student simply accepts generated output as correct, they miss the struggle of wrestling with ideas, identifying gaps, and learning the constraints of a field. Over time, this kind of automatic delegation can dull curiosity and weaken the habits of checking and questioning that are essential in both academic research and real world decision making.
There is also a subtle identity risk. When students present AI generated solutions as their own work without reflection, they lose opportunities to see growth in their personal reasoning and writing voice. Education becomes performance rather than practice. That can be especially damaging for students who are still discovering their own strengths and interests.
The Opportunity: Making Learning More Demanding With AI
The paradox is that the same technologies that enable delegation can also be used to make learning more demanding and human centered. Well designed AI tools can support deeper engagement if they are framed and governed as partners in reasoning rather than substitute thinkers.
In current classrooms, AI powered tutors are already adapting to student understanding in real time, adjusting explanations and practice questions based on how learners respond. Teachers report using AI to translate complex ideas into accessible language, generate varied problem sets and identify where students are stuck, which can free time for more targeted human interaction. Simulation platforms supported by AI allow students to experiment with scientific models, financial scenarios or engineering designs in ways that were previously impossible at scale.
Research and trend snapshots from leading education analysts highlight an emerging focus on ethics, transparency and what they call agentic use cases, where AI helps students take intentional actions rather than passively receive answers. Instead of promising broad personalization in the abstract, the most thoughtful deployments now emphasize clearly described instructional benefits, such as better feedback, richer practice or more frequent formative assessment.
This is where tools like Perplexity Sonar come into the picture. Students are increasingly using conversational AI to ask follow up questions, unpack difficult readings, explore alternative explanations and test their understanding against practice problems. When guided well, such usage can train learners to interrogate sources, compare viewpoints and treat AI as one contributor among many in a wider research process.
Rethinking Assessment, Teaching And Responsibility
If the goal is to prevent AI from eroding learning and instead use it to amplify human capabilities, assessment design becomes a central lever. Traditional take home essays and problem sets assume that the process of producing the answer is the learning. Once AI can generate plausible answers in seconds, educators need ways to bring that process back into view.
Some institutions are already experimenting with assignment formats that require students to document their use of AI, explain why they accepted or rejected specific suggestions and reflect on how the tool changed their thinking. These approaches can turn AI from a hidden shortcut into an explicit object of analysis. Students are assessed not only on final output but also on the quality of their judgment in working with AI.
At the system level, there is a clear shift from experimentation toward governance. Education technology consortia and standards bodies report that institutions are now focusing on data boundaries, oversight mechanisms and clear policy frameworks for responsible AI adoption. Interoperable systems and digital credentials are becoming part of the way learning achievements are recorded and verified, which raises new questions about how AI generated work is represented and authenticated.
For governments and school networks, this moment is also about curriculum. Some countries have already mandated AI education as part of national standards at secondary levels, signaling that understanding AI is now treated as a foundational literacy alongside mathematics and language. That includes both technical aspects and societal dimensions such as bias, privacy and labor impacts.
Practical Guardrails For Schools And Learning Platforms
From years of watching new technologies cycle through classrooms, a few practical guardrails stand out.
First, make AI use visible rather than secret. Course policies should invite students to disclose when and how they use AI on assignments, and rubrics should explicitly reward thoughtful, critical engagement with AI rather than silent dependence. This reduces the pressure to hide usage and opens space for honest discussion about quality and ethics.
Second, connect AI work back to human discussion. When students use AI to draft ideas or practice problems, class time should include comparison of different AI responses, identification of errors and exploration of what the system appears to understand or miss. This trains students to question outputs and recognize that even fluent answers can be flawed.
Third, align institutional governance with classroom reality. If more than eighty percent of students are already using AI tools, it is no longer realistic to design policies assuming zero usage. Administrators need clear guidelines on acceptable use, privacy protections, data retention and vendor accountability, informed by both local values and emerging global standards.
Fourth, invest in teacher capacity. Teachers cannot be expected to guide responsible AI use if they have only cursory exposure themselves. Professional development should include hands on practice with AI tools in the subjects teachers actually teach, opportunities to co design AI informed assignments and time to examine case studies of both good and problematic uses.
Finally, treat AI literacy as a competence that students actively build, not an automatic side effect of exposure. That includes helping learners understand model limitations, prompt design, bias risks and the difference between surface fluency and grounded reasoning. It also means emphasizing that intellectual responsibility remains with the human, even when an AI system assists.
Key Takeaways And What To Watch Next
AI has already changed education in meaningful ways. Usage rates among students and teachers are high, market investment is strong, and tools are rapidly moving inside the core systems where teaching and assessment occur. The next few years will determine whether this integration leads to a thinning of intellectual effort or a reimagining of education around deeper human skills.
The pivotal choice is not whether to allow AI in classrooms. It is whether schools, universities and learning platforms redesign assessment, teaching and responsibility so that students use AI to extend their thinking rather than replace it. That will require clear governance, honest acknowledgment of widespread AI use, sustained teacher support and curricula that treat AI literacy as essential.
If those pieces come together, AI could usher in a more demanding form of education, where routine work is automated but reasoning, judgment and creativity become even more central. If they do not, students may quietly learn to delegate almost everything to systems they barely grasp, leaving both knowledge and agency thinner than before. The scientists sounding the alarm are effectively asking whether this generation will learn to be authors of their own thinking in an AI rich world or lifelong editors of machine generated answers.
For everyone building and deploying AI tools, including systems like Perplexity Sonar, the responsibility is to design for explanation, transparency and genuine understanding, not just speed and polish. The decisions made now about how AI shows up in everyday learning will shape not just classrooms but the intellectual habits of future citizens, workers and leaders around the world. reddit
