AT A GLANCE
Artificial intelligence should not automatically be treated as the enemy of education or knowledge work. The more important question is whether people understand the work they submit, whether AI helped produce it or not. Instead of depending on AI detectors or blanket bans, schools and companies can evaluate something much more meaningful: Can the person explain the reasoning, defend the conclusions, recognize limitations, and apply what they learned to a different problem?
- AT A GLANCE
- We Are Measuring the Wrong Thing
- AI Is Already Part of How People Work and Learn
- Why AI Detection Is the Wrong Measurement
- The Real Skill Is Not Typing Every Word Yourself
- The Same Problem Exists Inside Companies
- What Should We Measure Instead?
- Human Understanding Is Harder to Fake
- Preparing People for a Future With AI
- Stop Detecting AI. Start Measuring Human Understanding.
- References
We Are Measuring the Wrong Thing
A few months ago, I decided to run a simple experiment. I uploaded an article that I had written entirely by myself to one of the most popular AI detection tools. The result surprised me: the detector said there was a 70% chance that the article had been written by artificial intelligence.
It was entirely wrong. I knew that because I had written the article myself, but the experience raised a much bigger question. If software can classify genuinely human writing as AI-generated, should schools, universities, companies, or managers use an AI detector as evidence that someone did something wrong?
The more I thought about it, the clearer the real problem became. We are measuring the wrong thing. The important question should not always be whether AI helped produce a document. It should be whether the person submitting that document understands what is inside it.
AI Is Already Part of How People Work and Learn
Whether institutions like it or not, artificial intelligence is becoming an everyday tool. People use it to write emails, summarize meetings, translate documents, organize ideas, analyze information, write software, prepare presentations, research subjects, and explore solutions to problems.
Students are already using it too. The Higher Education Policy Institute’s 2025 Student Generative AI Survey found that 92% of surveyed UK undergraduates had used AI in some form, up from 66% in the previous year’s survey. See the HEPI Student Generative AI Survey 2025 ↗
Pretending that AI doesn’t exist is as unrealistic as pretending people don’t use search engines or cellphones. That does not mean every use of AI should be accepted. Schools can establish rules for assignments, companies can protect confidential information, and organizations can determine when AI assistance is inappropriate.
But prohibition alone does not prepare people for a world in which AI is increasingly available. The larger challenge is teaching people how to use it without surrendering their responsibility for the result.
Why AI Detection Is the Wrong Measurement
The appeal of AI detection software is understandable. Schools want to protect academic integrity, while companies may want to know whether employees are producing original work or simply submitting machine-generated material. A detector appears to offer a simple answer to a complicated question.
The problem is that research has found significant limitations in AI-text detection. Independent evaluations have shown that detectors can produce false positives, fail to identify AI-generated material, and become less effective when generated text is edited or modified. Researchers have therefore warned against treating detector results as definitive evidence of authorship. See the independent evaluation of AI-text detection tools ↗
A false positive can have serious consequences. A student who genuinely wrote an assignment could be suspected of cheating, while someone who used AI extensively and then modified the output might pass undetected.
But there is an even more fundamental problem. Even a perfect AI detector would tell us who or what may have produced the words. It would not tell us whether the human understands them.
The Real Skill Is Not Typing Every Word Yourself
Artificial intelligence can produce paragraphs, summarize research, compare alternatives, generate arguments, and help organize complicated information. In many professional situations, these capabilities can make people considerably more productive.
What AI cannot assume is human responsibility for the final decision. It can evaluate information and suggest whether something appears plausible, but it can also misunderstand context, overlook important facts, or produce convincing answers that are wrong. A lawyer, engineer, consultant, journalist, accountant, teacher, or executive still has to determine whether an answer makes sense in the real-world situation for which that person is responsible.
This is why schools should go beyond teaching Prompt Engineering and develop something broader: AI literacy. People need to know how to ask useful questions, verify information, challenge answers, recognize uncertainty, protect confidential information, understand limitations, and know when not to follow what an AI system suggests.
Education has always been partly about learning how to think. Artificial intelligence makes that responsibility more visible because generating an impressive-looking answer is becoming easier while evaluating that answer can still require genuine knowledge.
The Same Problem Exists Inside Companies
This is not only a classroom debate. Businesses are confronting essentially the same question. Should employees use ChatGPT? Should AI-assisted reports be accepted? Should companies require employees to disclose when AI contributed substantially to their work?
Imagine a consultant delivering a strategic report to a client. AI may have helped organize research, compare possibilities, improve language, or structure the presentation. Does the client primarily care whether the consultant personally typed every sentence?
Usually, something more important is at stake. Can the consultant explain every recommendation? Can that person defend the assumptions, identify the risks, respond when the client challenges a conclusion, and accept responsibility for the advice?
Knowledge work has never been about typing out every word yourself. It has always been about understanding the problem and making good decisions.
What Should We Measure Instead?
If AI can help people produce essays, reports, presentations, and analyses, perhaps the finished document should not always be the only evidence we use to measure understanding. Schools and companies can add something AI detection cannot provide: a direct test of whether the person understands the work.
A teacher, manager, or client can ask the person to:
- Summarize the work without reading from it.
- Explain which evidence shaped the conclusion.
- Defend an important assumption when challenged.
- Explain how the recommendation would change under different circumstances.
- Apply the same idea to a new problem.
- Disclose when AI played a meaningful role in producing the work.
These questions do not automatically punish someone for using AI. They test whether the person remained intellectually involved in the process. Someone who understands the subject should be able to explain the reasoning, recognize limitations, defend important choices, and adapt the idea when circumstances change.
The purpose of education is not to prove that a machine was absent. It is to prove that a human mind was present, engaged, and capable of explaining what it learned.
I believe the same principle applies to the workplace. The purpose of knowledge work should not be to prove that AI was absent. It should be to demonstrate that people know their jobs, understand what they are presenting, stand behind their decisions, and are prepared to defend them.
Human Understanding Is Harder to Fake
A finished document can be deceptive because polished language is increasingly easy to generate. Conversation changes the test. Ask someone why a conclusion was reached, challenge an assumption, introduce new information, or ask how the same reasoning applies to a different situation, and understanding becomes much easier to observe.
This does not mean every essay needs an oral examination or every business report needs an interrogation. It means institutions can combine traditional assessment with methods that test reasoning rather than relying on software to guess who typed the words.
Transparency also matters. Schools and companies can establish clear rules about acceptable AI use and require disclosure when AI plays a meaningful role. The objective is not to pretend that authorship no longer matters. It is to distinguish authorship, assistance, understanding, and responsibility instead of treating them as the same question.
Preparing People for a Future With AI
Calculators changed how people performed calculations. Search engines changed how people found information. Smartphones changed how people communicated. Artificial intelligence is now changing how people create and work with knowledge.
That does not mean every old method should disappear. It means schools and companies need to decide which human abilities they actually want to preserve and develop. Critical thinking, curiosity, judgment, accountability, subject knowledge, and the ability to solve unfamiliar problems become more important when software can generate convincing material almost instantly.
Schools should prepare students who know how to question AI instead of simply trusting it. Companies should develop professionals who can use AI productively without surrendering their judgment to it. Both require us to stop treating the absence of AI as evidence of human competence.
Stop Detecting AI. Start Measuring Human Understanding.
AI detection attempts to answer a narrow question: did a machine probably generate these words? Human understanding asks a much more valuable question: does this person know what these words mean, why the conclusions were reached, where the weaknesses are, and what to do when circumstances change?
As artificial intelligence becomes more capable, separating those questions will become increasingly important. We should care about academic integrity, professional responsibility, transparency, and appropriate AI use, but none of those objectives requires us to confuse human authorship with human understanding.
The future should not be about proving that AI was absent. It should be about proving that human judgment remained present.
References
Higher Education Policy Institute (HEPI). (2025). Student Generative AI Survey 2025.
Student Generative AI Survey 2025 ↗
Weber-Wulff, D., et al. (2023/2024). Testing of Detection Tools for AI-Generated Text.
Research on AI-generated text detection tools ↗
Brynjolfsson, E., Li, D., & Raymond, L. (2023). Generative AI at Work. National Bureau of Economic Research.
Generative AI at Work ↗
