Professor Stops Grading Homework for Correctness as AI Spreads

Yes: University of California, Santa Barbara professor Daniel L. Schwartz stopped grading some homework primarily for whether answers were correct, saying generative AI makes a correct answer less reliable evidence that a student understands the material. The change does not mean homework is ignored or that students are excused from learning; it shifts attention toward effort, reasoning, reflection, and what students can explain.

What the professor changed

Schwartz, a professor in UCSB’s Department of Psychological and Brain Sciences, described the approach in public commentary about teaching in the age of generative AI. The central idea is to treat homework as practice rather than as a dependable measurement of mastery. Instead of awarding most of the credit for getting an answer right, an instructor can assess whether students attempted a problem, showed their process, considered feedback, and learned from the work.

This is a teaching choice, not a university-wide policy. It should not be read as a claim that every professor at UCSB—or at any other university—has stopped checking answers. Nor does it mean correctness has no value. Correctness still matters when students need accurate knowledge, but a polished answer by itself may tell an instructor less than it did before tools could generate one.

Why AI complicates homework grading

Generative AI systems can produce explanations, calculations, code, summaries, and essay drafts. That can make routine take-home assignments easier to complete without doing the underlying thinking. A correct response may come from a student’s understanding, a classmate, a textbook solution, an AI tool, or some combination. The final page alone cannot reliably distinguish among those possibilities.

The problem is not unique to AI. Students have always had access to notes, tutoring, answer keys, and help from peers. AI changes the convenience and scale: a student can ask for an explanation tailored to a specific question and receive a plausible response quickly. The response may still be wrong or incomplete, but its fluency can make it look convincing.

That creates a measurement problem. If homework is meant to show what a student can do independently, an at-home answer may be weak evidence. If homework is meant to help students practice, however, the process of trying, receiving feedback, and revising can be educational even when the first attempt is incorrect.

How grading for learning can work

A low-stakes practice assignment can give credit for completing meaningful work without treating every error as a failure. Instructors might ask students to show intermediate steps, explain a choice, mark where they became uncertain, or correct an initial attempt after seeing feedback. These features make the work more useful for learning and provide more information about the student’s thinking.

For example, in a statistics course, a student might submit a calculation, identify which formula they chose, and explain what the result means in context. If the calculation is wrong, the instructor can still see whether the student understood the setup or made an arithmetic mistake. In a writing course, students might submit an outline and a revision note explaining what changed and why, rather than relying only on a polished final draft.

• Completion and effort: Did the student attempt the assigned practice and engage with the task?
• Reasoning: Are the steps or decisions visible, rather than only the final answer?
• Revision: Did the student use feedback to identify and address a misunderstanding?
• Explanation: Can the student describe the idea in their own words or apply it to a new case?

These methods do not make grading automatic or eliminate judgment. They change what the instructor looks for and can require assignments designed to reveal learning, not just produce an answer.

Practice grades and mastery grades are different

A key distinction is between work intended for practice and work intended to certify what a student knows. Practice homework can be forgiving because mistakes are part of learning. A quiz, exam, oral explanation, lab demonstration, or supervised performance may be better suited to checking whether a student can apply a skill without outside assistance.

That does not mean every assessment must be in person. Instructors can use a mix of methods, such as short written explanations, in-class problem solving, drafts, project checkpoints, or questions that ask students to adapt a familiar method to a new situation. The appropriate choice depends on the course goals and the skill being assessed.

There is also a fairness consideration. If students are allowed to use AI, the instructor should explain what uses are permitted and what must be disclosed. If a task is meant to assess unaided performance, that expectation should be clear. Vague rules leave students guessing and can lead to inconsistent enforcement.

What the change does not mean

Stopping correctness from being the main basis for some homework grades is not the same as declaring all answers equally good. Students still need to learn accurate facts, valid methods, and sound reasoning. The difference is that a practice assignment can reward learning behaviors while accuracy is checked in other ways or used as feedback rather than as the sole source of points.

It is also not proof that AI has made teaching or assessment impossible. Teachers can redesign assignments, ask students to explain decisions, and compare work across stages. But no single tactic is foolproof: an AI system can generate reasoning, too, and a student may memorize an explanation without understanding it. Strong assessment uses several kinds of evidence rather than assuming that a single submitted answer tells the whole story.

What students should do

Students should follow the instructor’s policy rather than assume that AI use is allowed—or forbidden—because homework is graded for completion. They can use feedback and permitted tools to learn, but should be prepared to explain their choices and identify any help they used. When the policy is unclear, asking before submitting is safer than trying to infer the rules afterward.

• Show work and reasoning when the assignment asks for it.
• Use mistakes to identify what to review, rather than treating a completion grade as proof of mastery.
• Check generated explanations against course materials, since AI output can be confidently wrong.
• Disclose AI assistance if the instructor or assignment requires it.

Frequently Asked Questions

Which professor stopped grading homework for correctness?
Daniel L. Schwartz, a professor at the University of California, Santa Barbara, has discussed shifting some homework grading away from correctness as generative AI makes take-home answers harder to interpret as evidence of individual understanding.

Does this mean students get full credit for wrong answers?
Not necessarily. A course may award credit for completion, effort, or reasoning on practice work, while still checking accuracy through feedback or other assessments. The grading policy depends on the instructor.

Why not just use AI detectors?
Detector results are not a dependable substitute for evidence of learning and can incorrectly flag human-written work. Assignment design and direct demonstrations of understanding can provide more useful information than relying on a detector alone.

Can students use AI for homework?
There is no universal answer. Some instructors permit limited uses, others restrict them, and some require disclosure. Follow the specific course and assignment rules, and ask the instructor if they are unclear.

Is homework still useful if it is not graded for right answers?
Yes. Practice can help students find gaps, apply feedback, and build fluency. Its usefulness depends on students engaging with the problems and on the course providing ways to check whether they eventually understand the material.

Related

• Why Removing Duplicate Files Is Important
• Raspberry Pi Storage: SD Cards, SSDs & More
• Apple Joins AI Research Group: What It Means