AI Use Leads UC Berkeley Math Professor to See Little Point in Grading Homework ( Image with AI)

AI Use Leads UC Berkeley Math Professor to See Little Point in Grading Homework

Artificial intelligence is rapidly changing the way people learn, work, and solve complex problems. From generating written assignments to writing computer programs and answering difficult mathematical questions, modern AI tools are becoming increasingly capable. However, their growing presence in education is creating a serious challenge for teachers and universities: how can educators determine whether students genuinely understand their coursework when AI can produce answers that appear correct?

This question has become particularly urgent at the University of California, Berkeley, where renowned mathematician Richard Borcherds has reconsidered the value of traditional homework grading. According to a report published by The Daily Californian on October 23, 2026, Borcherds decided that grading mathematics homework for correctness no longer made sense under the circumstances he was observing in his classes.

After seeing students use AI tools to obtain consistently high scores, the professor concluded that paying graders to evaluate the correctness of take-home assignments had become difficult to justify. Instead, he now awards credit based on whether students submit their homework.

Although Borcherds acknowledged that this arrangement is not a particularly good system, he explained that he had not found a satisfactory alternative. His decision highlights a growing problem for modern education: assignments designed to measure individual knowledge can lose their value when students can use AI to produce convincing solutions.

The issue also raises broader questions about the future of mathematics, the purpose of university education, academic integrity, and the skills students will need in an increasingly automated world.

Who Is Richard Borcherds?

Richard Borcherds is a distinguished mathematician and professor at the University of California, Berkeley. He is internationally recognized for his contributions to mathematics and received the Fields Medal in 1998, one of the most prestigious awards in the mathematical sciences.

His work has contributed to advanced areas of mathematics, making his concerns about artificial intelligence particularly noteworthy. When an accomplished mathematician questions whether traditional homework assignments remain useful, the discussion extends beyond an individual classroom.

Borcherds's experience illustrates how AI is affecting even subjects that many people once considered especially resistant to automation.

Mathematics requires more than memorizing facts. It involves logical reasoning, abstract thinking, pattern recognition, problem-solving, and the ability to construct rigorous proofs. These skills have traditionally been regarded as demanding forms of intellectual work.

However, modern AI systems can increasingly solve many mathematical problems, explain established concepts, generate calculations, and assist with formal reasoning. Their capabilities are changing expectations about what students can accomplish independently.

Borcherds has expressed concern that if AI continues improving at its current pace, machines could eventually perform mathematical tasks better than humans across a much wider range of activities.

His concerns do not establish that human mathematicians will become unnecessary. Instead, they highlight the uncertainty facing educators as the relationship between human intelligence and machine-assisted problem-solving continues to evolve.

Why the Professor Questioned Traditional Homework Grading

Homework has long been a central component of mathematics education. Students receive problems to solve independently, submit their answers, and receive marks based on accuracy, reasoning, and the quality of their solutions.

The system is intended to accomplish several goals simultaneously.

First, homework gives students opportunities to practise mathematical techniques. Second, it helps teachers identify misunderstandings. Third, grades provide an indication of how well students have mastered particular concepts.

Traditional homework also encourages independent thinking. Students must work through unfamiliar problems, identify relevant mathematical principles, and determine how to reach a solution.

Artificial intelligence complicates this process.

A student can now enter a mathematical problem into an AI assistant and receive a detailed solution in seconds. Depending on the problem and the system being used, the response may include calculations, explanations, alternative methods, and even a formal proof.

This creates a difficult situation for instructors. A completed assignment may look excellent even when the student has not independently understood the underlying mathematics.

According to The Daily Californian, Borcherds observed students using AI to achieve consistently high scores. As a result, he questioned whether grading assignments for correctness continued to provide meaningful information about students' abilities.

If an assignment measures the quality of an answer produced with AI rather than the student's own understanding, its usefulness as an assessment tool becomes questionable.

Borcherds consequently shifted toward awarding credit for completing and submitting homework rather than evaluating whether each answer was correct.

The change represents a significant departure from conventional mathematics assessment. It also demonstrates how quickly AI can force educators to reconsider long-established academic practices.

How Artificial Intelligence Is Changing Mathematics Education

Artificial intelligence has introduced new possibilities for mathematics education, but it has also created problems that schools and universities are still learning to address.

In earlier generations, students generally relied on textbooks, classroom instruction, calculators, reference materials, and help from teachers or classmates. Although these resources could make learning easier, students still had to perform much of the work required to develop a solution.

AI assistants can now take on a much larger portion of that work.

A student struggling with algebra might ask an AI system to solve an equation. Someone studying calculus could request an explanation of differentiation or integration. A university student working on advanced mathematics might ask for help understanding a complicated proof.

These tools can provide immediate assistance, making educational support available outside normal classroom hours.

However, the same capabilities can also allow students to bypass the learning process.

For example, a student may submit a correct calculus solution without understanding why a particular theorem applies. Another student may produce a sophisticated proof by copying an AI-generated response without being able to explain its logical structure.

In both cases, the submitted work can create a misleading impression of mastery.

The central challenge is therefore not simply whether AI can solve mathematics problems. It is whether students are developing the knowledge and reasoning skills that mathematics assignments are intended to measure.

Universities must determine how to preserve those learning outcomes while acknowledging that AI tools are increasingly accessible.

This challenge is likely to affect not only mathematics but also programming, writing, engineering, economics, statistics, and other subjects in which AI can generate work that previously required substantial individual effort.

Why Grading Homework for Correctness May Become Less Reliable

Grades are often treated as evidence of academic achievement. Students use them to evaluate their progress, universities use them to make academic decisions, and employers sometimes consider them when assessing applicants.

However, grades are only useful when the assessment process measures the abilities it is intended to evaluate.

If students can routinely use AI to produce correct answers, take-home homework scores may become less reliable as indicators of individual understanding.

Consider two students completing the same difficult assignment.

The first student works independently, spends several hours reviewing course materials, makes mistakes, and eventually develops a correct solution.

The second student submits the problem to an AI assistant and receives a polished answer within seconds.

If both students receive the same score, the grade does not reveal the difference in effort or understanding.

This does not mean that every student using AI is avoiding learning. Some students use these systems to clarify difficult concepts, compare solution methods, or identify errors in their own work.

Nevertheless, when an instructor cannot reliably distinguish independent reasoning from generated answers, the meaning of a homework score becomes less clear.

The problem becomes particularly serious when homework contributes substantially to a student's final grade.

A student who relies heavily on AI may receive high marks without developing the skills needed for a closed-book examination, a research project, or a professional task requiring independent judgment.

Meanwhile, students who follow the assignment rules and complete the work themselves may face greater difficulty achieving similar scores.

This creates concerns about fairness, motivation, and the credibility of academic evaluation.

Borcherds's decision reflects one response to this problem: if correctness scores cannot reliably measure independent achievement, an instructor may decide that traditional grading is no longer worth the effort.

However, this approach also has limitations because completion alone cannot demonstrate whether students understand the material.

A more sustainable solution will likely require universities to rethink how they design assignments and measure learning.

The Problem With AI Detection Tools

One possible response to AI-generated homework is to use software designed to identify content produced by artificial intelligence. However, this approach has its own limitations.

UC Berkeley has examined AI-detection technologies, including tools associated with Turnitin and Rumi. According to the university's statements reported by The Daily Californian, the tools tested were not considered suitable for standardization across the campus.

This finding highlights an important distinction between identifying suspicious work and proving that a student has violated academic rules.

AI-detection systems examine characteristics of submitted material to estimate whether it may have been generated by a machine. However, their results are not always reliable enough to establish who actually produced a particular assignment.

A student might write an answer independently and still have it incorrectly flagged by a detector. Conversely, a student could substantially edit an AI-generated response, making its origin more difficult to identify.

These limitations create serious concerns for universities.

If institutions rely too heavily on unreliable detection systems, students who complete their assignments honestly could face unfair accusations. Teachers could also spend considerable time investigating questionable results without reaching a clear conclusion.

Mathematics introduces additional complications. A correct equation, a standard proof technique, or a familiar mathematical explanation may look similar regardless of whether a student or an AI system produced it.

Consequently, determining authorship from the final answer alone can be difficult.

A more effective strategy may involve examining students' understanding directly. Teachers can ask learners to explain their reasoning, solve related problems independently, or discuss how they approached a particular question.

Such methods provide evidence of understanding that a written submission alone may not offer.

AI detection may remain one component of academic integrity procedures, but it cannot replace thoughtful assessment design, transparent classroom policies, and fair investigations.

UC Berkeley's Approach to AI Use in Education

The situation at Berkeley reflects a broader challenge facing universities: how to establish consistent expectations when AI tools are developing faster than traditional educational policies.

According to The Daily Californian, the Berkeley Division of the Academic Senate asks faculty members to explain course expectations regarding AI use in their syllabi.

However, the rules can vary considerably depending on the subject, course level, and instructor.

This flexibility recognizes that AI may be useful in some educational activities while being inappropriate in others.

For example, an introductory course might prohibit AI-generated answers during independent assessments because students need to develop foundational skills. An advanced computer science course, by contrast, might allow students to use AI tools to examine programming techniques, provided they understand and explain the resulting code.

A research-oriented mathematics course could permit AI-assisted exploration while requiring students to verify every mathematical claim and disclose how the technology contributed to their work.

There is no single policy that automatically resolves every problem.

Berkeley has also developed resources through its AI Hub to help instructors understand available tools, training materials, and policy options. Nevertheless, the final decision about classroom AI use remains largely with individual faculty members.

This approach makes it possible to adapt rules to different subjects, but it can also create uncertainty for students taking multiple courses.

A student may be allowed to use AI for brainstorming in one class but prohibited from using it for similar work in another.

Clear communication is therefore essential. Students need to understand not only whether AI is permitted but also which activities are acceptable, when disclosure is required, and what consequences may follow from unauthorized use.

Without consistent communication, even well-intentioned students may misunderstand their responsibilities.

Is AI Use Always Academic Cheating?

The debate surrounding AI in education often treats AI use and academic dishonesty as if they were the same thing. In reality, the situation is more complicated.

Artificial intelligence can support learning when students use it to understand difficult ideas, receive feedback, or practise unfamiliar techniques.

For example, a student struggling with calculus might ask an AI assistant to explain the chain rule using a simple example. The student could then attempt several problems independently and use the tool to understand mistakes.

Similarly, someone studying linear algebra might request an intuitive explanation of matrix multiplication before returning to the textbook.

In these situations, AI can function as an additional educational resource.

The problem arises when students use AI in ways that violate their instructors' rules or undermine the purpose of an assignment.

Submitting a generated solution as evidence of independent mathematical ability can be misleading when the assignment specifically requires individual work.

It can also deprive students of the practice needed to develop genuine expertise.

The distinction depends on the course policy, the assignment's objectives, and how the technology is used.

Students should therefore avoid assuming that every instructor permits AI simply because the technology is widely available. They should also avoid assuming that every use of AI is prohibited.

The most responsible approach is to follow the rules for each course, disclose AI assistance when required, and ensure that submitted work reflects the student's actual understanding.

Universities, meanwhile, need to communicate expectations clearly enough that students can comply with them.

The Difference Between Getting an Answer and Learning Mathematics

One of the most important questions raised by AI-generated homework concerns the difference between obtaining a correct answer and developing mathematical understanding.

These two outcomes are related, but they are not identical.

Mathematics education is intended to help students develop the ability to recognize patterns, construct logical arguments, select appropriate methods, and apply concepts to unfamiliar situations.

A student who understands a mathematical principle can often adapt it to a new problem. Someone who merely copies a solution may struggle when the question changes slightly.

Consider a student learning quadratic equations.

An AI system might immediately provide the roots of an equation and explain the calculation. However, the student must still understand what the roots represent, why the quadratic formula works, and when alternative methods are appropriate.

Without this understanding, the student may have difficulty applying the same concept during an examination or in a real-world problem.

The same distinction becomes even more important in advanced mathematics, where solutions may depend on multiple theorems, carefully justified assumptions, and lengthy chains of reasoning.

A polished proof does not necessarily demonstrate that its recipient understands why the proof is valid.

This is why educators must consider the purpose of each assignment before deciding how AI should be used.

If the objective is to practise a technique, students may benefit from solving problems independently before consulting an AI assistant.

If the objective is to evaluate the quality of an AI-generated proof, using the technology may be entirely appropriate.

The educational value depends on whether the activity develops the intended skills.

Why Traditional Homework May Need to Change

Borcherds's experience raises questions about the future of take-home assignments.

Traditional homework assumes that students will complete problems outside class and submit their own work for evaluation. Teachers then use those submissions to estimate how well students understand the material.

AI challenges that assumption because it can perform many of the tasks required to complete assignments.

Rather than abandoning homework altogether, universities could redesign assignments so that students must demonstrate their reasoning and decision-making.

One option is to require students to explain why they selected a particular method. Instead of submitting only a final answer, they could describe the assumptions behind their approach and discuss alternative solutions.

Another possibility is to use personalized questions that vary between students. Although AI may still help solve these problems, personalization can make it harder to rely on a single copied answer.

Teachers could also ask students to submit drafts, intermediate calculations, or brief reflections explaining what they found difficult.

These approaches are not foolproof, but they can provide more information about the learning process.

A particularly useful strategy is to combine take-home assignments with short, supervised assessments.

Students might complete challenging homework with access to approved resources, then solve a related problem independently during class. The supervised task helps establish whether they understand the underlying concepts.

This arrangement recognizes that independent practice and formal assessment serve different purposes.

Homework helps students learn through repetition and experimentation. Examinations and oral discussions can provide additional evidence of individual understanding.

The future of assessment may therefore involve a combination of traditional exercises, AI-assisted activities, in-person tests, and explanations of student reasoning.

Could Oral Examinations Become More Important?

Oral examinations offer one possible response to the challenges of AI-generated work.

During an oral assessment, an instructor asks students to explain a concept, solve a problem, or justify a mathematical argument. The instructor can then ask follow-up questions based on the student's answers.

This format makes it possible to examine understanding beyond the submitted solution.

For example, a student might present a proof of a theorem. The instructor could ask why a particular assumption is necessary, whether the argument works under different conditions, or how the result connects to another mathematical concept.

A student who understands the material should generally be better equipped to respond meaningfully.

Oral examinations also allow teachers to identify misconceptions that might remain hidden in written assignments.

However, they have limitations.

Conducting individual oral examinations requires substantial time, especially in large university classes. Students may also experience anxiety when answering questions under pressure, meaning performance can be influenced by communication skills and nervousness as well as subject knowledge.

Consequently, oral examinations are unlikely to replace every form of assessment.

Short oral explanations, classroom discussions, individual problem-solving sessions, and occasional interviews may offer more practical alternatives.

The goal should not be to eliminate every opportunity for AI assistance. It should be to create multiple ways for students to demonstrate genuine learning.

The Impact on Honest Students

One of the most troubling aspects of AI-generated homework is the potential effect on students who follow the rules.

Imagine a class in which some students independently solve difficult problems while others rely heavily on AI-generated answers.

If grades depend primarily on correctness, students using AI may obtain high marks with less effort. Those working independently may receive lower scores despite spending more time developing their skills.

This can create a perception that academic honesty puts students at a disadvantage.

Over time, such perceptions may weaken the incentive to practise independently.

Students may begin to wonder why they should spend hours solving a problem when an AI system can generate a correct answer in seconds.

However, grades obtained through AI assistance do not necessarily translate into genuine competence.

Students who avoid learning foundational concepts may encounter difficulties when they need to solve unfamiliar problems without technological support.

The challenge for educators is to ensure that assessment systems reward the abilities students are expected to develop.

Fairness also requires clear rules. Students should know whether AI is allowed, what forms of assistance are acceptable, and how their work will be evaluated.

If instructors prohibit certain tools, they should communicate those restrictions before students begin their assignments.

If AI is permitted, assessment criteria should distinguish between appropriate assistance and the skills students must demonstrate independently.

A transparent system can reduce confusion and encourage students to focus on learning rather than merely maximizing marks.

What Does AI Mean for the Future of Mathematics?

Borcherds has expressed uncertainty about whether he should encourage students to pursue mathematics as a career.

His concern reflects the possibility that increasingly capable AI systems could automate tasks traditionally performed by mathematicians.

This is an important question, but it requires careful consideration.

Mathematics includes a wide range of activities. Some involve routine calculations, symbolic manipulation, or checking established results. Others require identifying promising research questions, developing new concepts, connecting different areas of knowledge, and deciding which problems deserve attention.

AI systems may become increasingly capable in many of these areas. However, progress in automated problem-solving does not automatically mean that every mathematical profession will disappear.

The future will depend on how AI capabilities develop and how researchers, universities, businesses, and public institutions adopt them.

Even if AI systems become highly effective at producing mathematical solutions, people may still need to evaluate whether those solutions are correct, relevant, useful, and based on appropriate assumptions.

Human mathematicians may also contribute by choosing research directions, interpreting results, developing educational approaches, and applying mathematical knowledge to practical problems.

At the same time, it would be unrealistic to assume that mathematics careers will remain unchanged.

Students entering the field may need to learn how to work effectively with AI, verify machine-generated reasoning, and combine mathematical expertise with computing and other disciplines.

The most useful preparation may involve developing strong foundational knowledge while becoming familiar with the strengths and limitations of modern AI tools.

Rather than choosing a career based on the assumption that AI will have no effect, students should prepare for a profession in which the division of work between humans and machines may continue to change.

Can AI Replace Human Mathematical Creativity?

The question of whether AI can replace human mathematical creativity remains open.

Modern systems can produce solutions, identify patterns, and assist with research tasks that once required extensive human effort. These capabilities challenge traditional assumptions about the relationship between computation and intelligence.

However, evaluating mathematical creativity involves more than determining whether a system can produce a correct answer.

Researchers may need to consider whether a result is genuinely new, how it changes existing understanding, and whether it opens useful directions for further investigation.

AI may contribute to these processes by generating hypotheses, testing possibilities, or identifying connections that researchers have overlooked.

Human mathematicians can then examine those suggestions, determine their significance, and decide how they fit into the broader development of the field.

It is also important to distinguish between a technological prediction and an established fact.

Borcherds's warning that AI could outperform human mathematicians across all tasks within a few years expresses his concern about the pace of development. It should not be interpreted as proof that such an outcome is inevitable.

AI progress can be rapid, but its future trajectory remains uncertain.

The capabilities of a system also depend on the particular task, available information, computational resources, and the reliability of its reasoning.

Mathematics will almost certainly continue to change as AI improves, but the precise nature of that transformation is difficult to predict.

Why Universities May Be Slow to Adapt

Borcherds also discussed the possibility that academic institutions could take years to respond fully to AI developments.

Universities operate through established systems of teaching, assessment, accreditation, research, and administration. Changing those systems can require agreement among faculty members, administrators, academic committees, and other stakeholders.

Even when new technology creates an obvious challenge, institutions may need time to evaluate alternatives and implement revised procedures.

For example, replacing traditional examinations with oral assessments could require additional staff, changes to course schedules, and new methods of evaluating students consistently.

Introducing AI-assisted assignments may require faculty training, revised academic integrity policies, and guidance on data privacy.

Universities must also consider accessibility. Not every student has equal access to paid AI subscriptions, reliable internet connections, or suitable devices.

A policy that assumes every student can use the same tools may unintentionally create disadvantages.

These practical concerns help explain why educational change can be slower than technological development.

Nevertheless, institutions cannot rely indefinitely on assessment methods that no longer measure learning effectively.

The challenge is to adapt thoughtfully rather than reacting to every new AI capability with either unrestricted acceptance or a complete prohibition.

A Balanced Approach to AI in Education

The controversy surrounding homework grading suggests that universities need a balanced approach to artificial intelligence.

Completely banning AI may be difficult to enforce in every context, while allowing unrestricted use can undermine the purpose of independent assessments.

A more practical strategy would begin by identifying the learning objectives of each assignment.

For tasks designed to develop foundational skills, instructors could require independent work and use supervised assessments to verify understanding.

For assignments intended to teach AI literacy, students could be encouraged to use AI while documenting their prompts, checking the accuracy of the output, and explaining their decisions.

Teachers could also develop assignments that require critical analysis rather than simply producing a correct answer.

For instance, students might compare two AI-generated solutions, identify errors, explain which method is more efficient, or determine whether a proof depends on an unjustified assumption.

Such activities can turn AI into an object of analysis rather than merely a shortcut to completion.

Assessment should also involve more than one type of evidence. Homework, classroom participation, quizzes, projects, and examinations can each reveal different aspects of learning.

This approach reduces dependence on any single measure.

Finally, universities should regularly review their policies as AI technology changes. Rules that work well today may need to be revised as systems become more capable.

The objective is not to preserve every traditional practice unchanged. It is to protect the educational outcomes that make those practices valuable.

What Students Can Learn From the Berkeley Debate

Students can draw several important lessons from Borcherds's experience.

First, submitting a correct answer is not the same as understanding the underlying material. Genuine learning requires the ability to explain, apply, and evaluate concepts independently.

Second, AI should be used responsibly. When permitted, it can provide useful explanations and feedback, but students should verify its output rather than accepting every response without question.

Third, course policies matter. Students should understand their instructors' rules before using AI for assignments.

Fourth, foundational skills remain important even as automation advances. Students who understand mathematics deeply will be better positioned to evaluate AI-generated solutions and recognize mistakes.

Finally, career planning should account for technological change without assuming that every profession will disappear.

Students interested in mathematics can strengthen their prospects by developing problem-solving abilities, learning relevant computational tools, and cultivating the judgment needed to evaluate complex results.

AI may change what mathematicians do, but students who combine subject knowledge with adaptability will be better prepared for an uncertain future.

Richard Borcherds's decision to stop grading mathematics homework for correctness illustrates the growing pressure artificial intelligence is placing on traditional education.

At the University of California, Berkeley, the professor observed students using AI to achieve high homework scores and concluded that conventional grading was no longer providing the information he needed. He now awards credit based on submission, while acknowledging that this is an imperfect solution.

His experience highlights a problem that extends far beyond one university or one mathematical discipline. When AI can generate convincing answers, educators must reconsider how they measure knowledge, encourage independent thinking, and maintain academic integrity.

AI-detection software has not provided a complete solution, and classroom policies continue to vary. Universities will therefore need to explore alternative assessments, clearer rules, and new ways of evaluating students' understanding.

The debate also raises difficult questions about the future of mathematics careers. Although AI is becoming more capable, predictions about the complete replacement of human mathematicians remain uncertain.

What is clear is that education cannot rely solely on traditional assumptions about how students complete assignments.

The next stage of educational development will require institutions to distinguish between producing an answer and understanding it. By combining responsible AI use with meaningful assessment, universities can adapt to technological change without abandoning the fundamental purpose of learning.

The central question is no longer simply whether students can obtain the correct answer. It is whether they understand the answer, can explain why it works, and know how to use that knowledge when facing a new problem.

Frequently Asked Questions (FAQs)

1. Why did Richard Borcherds stop grading homework for correctness?

Borcherds observed students using AI tools to achieve consistently high scores. He concluded that grading take-home assignments for correctness was no longer a reliable way to assess independent mathematical understanding.

2. Who is Richard Borcherds?

Richard Borcherds is a mathematician and professor at the University of California, Berkeley. He received the Fields Medal in 1998 for his contributions to mathematics.

3. Does UC Berkeley prohibit students from using AI?

AI policies vary by course and instructor. The university asks faculty members to clarify their expectations regarding AI use in course syllabi, so students should check the rules for each class.

4. Can AI-detection tools reliably identify AI-generated homework?

AI-detection tools have limitations and can produce misleading results. Berkeley has examined such tools but has not adopted them as a standardized campus-wide solution for identifying AI-generated coursework.

5. Will artificial intelligence replace human mathematicians?

AI may automate or transform many mathematical tasks, but the complete replacement of human mathematicians is not an established fact. The future of the profession will depend on technological progress and how people adapt to new capabilities.

Visit OUR Website:

Website:

https://news-reflects.blogspot.com ❤️❤️

https://newsreflects.com ❤️❤️

https://songbadreflect.blogspot.com

https://newsforall13.blogspot.com

Visit SOCIAL Media

YouTube:

https://www.youtube.com/@WhisperedKingdoms-1981 ❤️❤️

Entertainment nest - YouTube

https://www.youtube.com/@sonbadreflects1981 (Bangla Language)

https://www.youtube.com/@farhana12fm