Artificial Intelligence (AI) is changing how colleges assess performance and how employers judge professional ability. When a system can solve math problems, revise essays, or complete coding tasks, some established tests become less informative. This affects students and early-career workers alike, because both increasingly need to show what they can accomplish with AI and what they understand without it.
Why assessments are losing their evidentiary value
Richard Borcherds of the University of California, Berkeley, no longer sees a point in grading his students’ homework, according to a report on AI in university mathematics. He considers the use of AI unavoidable. If an instructor cannot tell whether a submitted solution was developed by the student or generated by a system, the grade no longer measures mathematical ability reliably.
The issue extends beyond mathematics. According to the report, a nationwide survey by Darmstadt University of Applied Sciences found that 92 percent of respondents in Germany use tools such as ChatGPT at least occasionally. Common uses include answering comprehension questions and generating, analyzing, or translating text. In the United States, an OpenAI analysis found that college students used ChatGPT more often than any other group examined.
A written assignment asking you to explain several steps in a calculation is one practical example. The final answer may look correct even if you cannot follow the reasoning yourself. The same applies to an essay that has been translated or polished: its quality may partly demonstrate the tool’s ability rather than your language skills alone.
This does not make assessment obsolete, but it does require a closer match between the test and its purpose. If a college wants to evaluate your understanding, a take-home submission may be less useful than a conversation about your reasoning. If competent use of tools is the actual objective, AI would need to be explicitly permitted and its role made traceable.
What subject knowledge still contributes
The declining value of certain assignments does not make subject knowledge irrelevant. More than 1,000 mathematicians have pledged to avoid AI in their research through the Association for Human Mathematics, according to a report on the organization. The group argues that mathematical understanding develops through writing, careful thought, and human expertise, while automated production could undermine those foundations.
This position contrasts sharply with Borcherds’ pragmatic view that AI can no longer be kept out of university study. The two reports agree that AI is affecting mathematics deeply, but they propose opposing responses: one side wants assessment to adapt, while the other rejects AI use in research altogether. The report also says OpenAI published 722 papers in the same month that claimed to prove or disprove mathematical problems; the quality of that large body of work cannot be independently assessed from the supplied information.
Subject knowledge remains essential for checking results, spotting errors, and asking useful questions. If you can only accept an AI response because it sounds plausible, you depend on the model’s reliability. With a solid grasp of the basics, you can judge whether a calculation holds up, a translation preserves the meaning, or a summary omits a crucial limitation.
A discussion at the 2026 CNO Panel in Bern also placed purpose ahead of technology. The contributions on meaningful digitalization called for an educational compass, closer study of digital communication culture, and continued room for printed books. Digitalization should not become an end in itself; in less ceremonial terms, not every available tool automatically produces a better learning method.
What is changing in college choice and hiring
AI affects not only university study but also the choice of institution. For a Swiss visibility index, Advance Metrics examined which colleges ChatGPT, Gemini, and Claude mention in response to questions about where to study. According to the analysis of Swiss universities, the University of Zurich led in seven of 25 categories and was followed overall by the University of Bern and ETH Zurich.
The results varied by subject. The University of St. Gallen led in business and management, while Zurich University of the Arts ranked first for art and design as well as music and theater. Twelve institutions placed first in at least one category, showing that a broad recommendation says little about the best choice for a particular field.
The company also found more than 1,000 name variants for roughly 120 institutions in the raw data. Models apparently did not always recognize names such as “ETH Zürich,” “ETH Zurich,” and “Eidgenössische Technische Hochschule Zürich” as the same institution with confidence. This is particularly relevant in multilingual Switzerland, where different language versions can affect visibility; an AI recommendation is therefore better treated as a starting point than as a dependable university ranking.
Employers are changing their selection methods as well. At Swiss security software provider Proton, employees are reportedly expected to explain their use of AI in a new round of interviews. A report on Proton’s internal AI interviews, citing 404 Media and an internal memo, says the outcome could affect duties, responsibilities, and continued employment. The interviews are reportedly conducted by senior employees who have already demonstrated strong AI skills rather than by the workers’ direct managers.
These details have not been independently confirmed beyond the cited reporting. Even so, they illustrate a new expectation that may emerge: occasionally opening a tool may no longer be enough, as employees could be asked to explain how it produces better products or working methods. For Switzerland, this is a concrete example of AI competence affecting not only job postings but also the organization of existing roles.
Pros and Cons of AI as a performance tool
Pros:
- Faster clarification – Students can use AI to work through comprehension questions and analyze or translate text.
- More emphasis on application – Assessments and interviews can focus on how you verify results and integrate them into useful work.
- New opportunities for newcomers – People who identify problems and use AI purposefully may offset some lack of professional experience.
- Subject-specific guidance – AI can make different institutions visible for different areas of study.
Cons:
- Unclear individual contribution – A correct submission no longer proves reliably that you understand the material.
- Dependence on models – Incorrect classifications and inconsistent names can distort recommendations.
- Workplace pressure – AI ability can become a factor in assigning responsibilities or deciding continued employment.
- Loss of learning processes – Outsourcing thought, writing, and verification too early may weaken durable subject knowledge.
What this means for you
If you are a beginner, the first sensible step is to stop treating AI as an invisible substitute for your own work. You could ask it to explain a difficult passage or compare two possible translations, then verify the result against your course materials. When choosing a college, you can ask AI to suggest institutions, but you should compare its recommendations with subject profiles and official information from those institutions.
Advanced users gain more by evaluating the entire workflow rather than merely writing better instructions. That includes deciding which task the AI should perform, how you will check its output, and where human judgment remains necessary. For academic and professional assignments, you should also document which parts are yours and where a tool contributed, provided the applicable rules permit that use.
Felix Rieseberg, an engineering manager at Anthropic, offers an example from hiring. According to a report on changing selection criteria, he considers small coding exercises in interviews largely pointless because Claude can perform them better than candidates. In his experience, entrepreneurial thinking, identifying overlooked problems, and turning ideas into results now matter more than conventional test scores.
Rieseberg describes a proposed offline architecture at Notion that colleagues considered technically difficult and supposedly unacceptable to the company’s chief executive. The technical obstacle was real, but nobody had verified the assumed executive resistance. The newcomer’s contribution therefore lay not only in implementation but also in questioning an entrenched assumption.
This also reveals a contrast between Anthropic and Proton. In Rieseberg’s account, a traditional test loses importance, while Proton is reportedly adding interviews specifically about AI use. Both approaches shift attention away from an isolated task and toward how people work with AI, but it remains unclear whether they measure competence more fairly or merely create a new form of pressure to conform.
AI does not make education, expertise, or hiring worthless, but it weakens tasks whose output can be automated easily. Assessments become more credible when they combine understanding, verification, judgment, and purposeful application. The unresolved risk is that colleges and employers may introduce new AI expectations before they have transparent and fair standards for evaluating them.
Sources
- KI im Studium: Warum dieser Professor keine Mathe-Hausaufgaben mehr benoten will – Unknown, 2026-10-28
- Diese Schweizer Hochschulen empfiehlt KI am häufigsten – Unknown, 2026-10-28
- Verband gegen KI: Über 1.000 Mathematiker verpflichten sich zum kompletten Verzicht – Unknown, 2026-10-28
- Wegen KI: Proton-Mitarbeiter müssen erneut zum Bewerbungsgespräch – Unknown, 2026-10-28
- Weil KI den Code schreibt, wird bei Anthropic unternehmerisches Denken zur wichtigsten Währung für Einsteiger – Unknown, 2026-10-28
- Warum wir Sinnhorizonte besser verstehen sollten – Unknown, 2026-10-28
Image: RDNE Stock project via Pexels