The Honking Goose

I’m observing a trend that has me very concerned, and it’s that knowledge is being applied without an understanding of the underlying foundations.

This is made possible largely by large language models (LLMs) such as Microsoft Copilot and Claude. The sky is the limit when you give a task to these AI tools. However, these tools are increasingly helping their users skip the part of understanding the problem. AI chatbots act like tutors who just do your homework for you while you watch and don’t participate, who doesn’t press when you don’t ask any questions. In practice students can score 100% on their homework, but then they sit down to take their own exam, and they don’t know a single answer.

Questions in software development spaces have shifted from What is a NullPointerException, and how do I fix it? to the user prompting, Your code didn’t work! [Pastes whatever error message, if we are lucky]! Later, the same person will post a rant complaining that their prompt in their preferred AI platform is not working.

When problems require a little human judgment to nudge the AI in the right direction, the lack of an understanding of computer science fundamentals or knowing a single programming language (and thus being able to spot and correct mistakes) makes users stuck, and many will give up or try to recreate the project using another AI product because they lack an understanding of the underlying problem.

These are not confused internet users; they’re working in every industry, sometimes employed as software engineers and increasingly in another role, using AI to try new ideas. This problem is not unique to any major; AI competence is built in the classroom, and it’s time we talked about it.

ai norms in the classroom

Search the websites of the American Psychological Association or the Society for Industrial and Organizational Psychology for guidance on using AI. There is a wealth of information to navigate, but it’s often unclear how psychology students and professionals should adapt to this new technology. Many individuals are seeking clear guidelines to follow, fearing potential repercussions such as being dismissed from their institutions or losing their professional licenses or memberships in their affiliated organizations. Currently, the situation is characterized by uncertainty and a reluctance to discuss AI, which could compromise our work. Many universities ban the use of AI entirely and students are not taught the information they’ll need to use it in their career. This could have dangerous long-term effects.

I think students should have discussions around how they use AI academically even talking about different AI products and their prompts as this is part of the learning process. Some individuals propose attaching a copy of every chat involved in the research and writing process as an appended material. This workaround creates more problems than it solves when it’s implemented at the policy level.

a compliance trap

Disclosure requirements for AI usage are promising in theory, yet numerous challenges can arise that are not the individual’s fault. Many AI tools can organize hundreds of chats into a single folder related to a project you’re working on, providing a potential solution for managing disclosures of AI use. However, modern web browsers, including Google Chrome and Microsoft Edge, automatically convert search queries into AI queries, meaning that your search on Bing is now an AI search. If you used the AI chat in the sidebar for a quick citation, there’s a good chance neither interaction will be included in the project documentation. The repercussions of such oversights can be significant, leading many individuals to play it safe by avoiding AI altogether. However, by doing so, they risk missing the opportunity to develop the skills necessary for professional use in the future. Some may even deny using AI when they did, as unrecorded chats could be construed as academic or professional misconduct. If stringent disclosure requirements are enforced, we might lose valuable insights into how AI is utilized in classrooms and research.

Researchers studying AI use in the classroom have created self-report instruments which can be incredibly invasive and in the linked example ask the student to list every prompt. Beyond being a compliance-trap (I’ll call it what it is), we don’t ask students to record every Google Search or provide their entire browsing history as part of the submission process for assignments, AI should be treated like any other tool, not held to an impossible standard then blamed for the problems in modern education. I think the example of what happens in practice (even just an example from research) is why we should be very careful with disclosure policies.

explain your reasoning

I think that as future papers are written and graded, part of the process needs to include a discussion with the student. A face-to-face discussion (or Zoom for distance learning) does not need to be long; the instructor should be able to read the paper, ask questions about it, and the student should be able to describe their thought process to the instructor. The key idea is that if a student wrote the paper, they should be able to answer questions about it. This should be a part of the paper’s grade, and it should not be optional. The question is not whether you wrote the paper (at least not directly) but whether you understand what was written. Creating an objective process for this will be important, and educational leaders should work on how to go about solving this.

foundational skills matter more than results

I think avoiding AI entirely in universities is going to become harder and harder as more applications adopt AI features in one form or another. However, there are still tasks that students need to learn how to do without AI first.

For example, in an English class, students need to understand how a paper is structured and how to cite sources properly. In a computer science class, students need to learn how to write their first Python program on their own. These foundational skills matter because students need to understand the work before they can responsibly use AI to support it.

This means that some courses will need to be AI-limited, and some assignments will need to prohibit AI use entirely. That is not about rejecting AI. It is about making sure the education students are paying for gives them the skills they need to understand the problems they are trying to solve, so they can later integrate AI effectively and thoughtfully when the time comes.

beyond the laptop

Attending lectures, reading your textbook, and writing your papers to apply your knowledge are all great ways to engage in the academic process but they present the opportunity to be a passive learner, rather than an active one. Johns Hopkins University wrote an article about how to transform mostly passive learning activities into active ones. Active learning is essential to absorbing information and retaining it in long term memory.

One idea I have for improvement is that instead of writing lengthy discussion board postings, instructors should assign students to peer discussion groups, where they will have the opportunity to discuss a topic together in real time. This could happen in the classroom or over a Zoom meeting; the important thing is that the discussion occurs in real time and ideally has the ground rule of not using Google or AI apps. This has the added benefit of not only demonstrating competence to the instructor but also preparing students to present their knowledge to others in contexts such as job interviews.

We Are Teachers published an article that describes classroom activities that build critical thinking skills and aren’t strictly tied to the course material, giving students a break to absorb what was already taught while participating in another activity. These activities, combined with “let’s put away our laptops for now,” can help students build the critical thinking skills they need and gain confidence in their ability to work without AI assistance when needed.

next steps

Now is the time to act, methodically, but promptly. This is a systems problem. There is not a single entity with the power to fix the problem by themselves. Here are my proposed changes that could start to solve this issue.

• Universities need to be involved. Blanket bans and the implied statement of we won’t allow you to use it, and we won’t teach you how to use in your career, isn’t helping. You have those who will avoid AI altogether and be harmed later in their careers. Or those who choose to hide it, which reduces transparency, and make problems harder to fix later on. And we risk developing a workforce that is able to produce answers but not explain them.
• Professional organizations such as the AMA, ABA, APA, SIOP, etc. need to write clear rules and expectations on how AI can be used in their education and professional careers.
• Licensure authorities need to write clear rules and expectations on how licensed professionals can use AI in their teaching and professional careers.
• And this may be the hardest part is asking employers to provide AI training and prepare employees on how to use provided AI tools in the workplace. When employers don’t do this, employees use their personal AI accounts in the workplace and take data with them when they leave. Depending on the industry and applicable regulations that is a major compliance disaster waiting to happen.

Finally, I ask you to think beyond this blog post. This is a discussion that all stakeholders should be involved in. We need to hear the voices of everyone involved to do this right!