Gábor’s AI policy for advanced applied metrics classes and projects
- v. 0.3.2. Date: 16. December 2025
- This is a policy is written for my advanced classes: Applied Economics with Sports Data, Data Analysis 3 Prediction and ML. Data Analysis 4 Causal Analysis.
- Gábor Békés (Central European University, BekesG@ceu.edu)
- Licence: CC BY-NC 4.0 free to share and adapt.
Big picture: Enjoy “AI” with moderation
This policy is aimed at more advanced classes and assume that you already have core coding and statistics/econometrics skills. The abundance of AI means that the process of doing data analysis projects has changed, and these tools are becoming essential help.
AI has become part of our lives, and most people are using it regularly. I assume you will use AI as a tool to enhance your work. Moreover, the ability to use “AI” as copilot, carry out prompt engineering is a an important skill to master.
However, you should use my courses to develop skills in understanding applied economics and data analysis, master econometric methods and get experience working with data. You will learn how to think about connecting ideas, methods and find appropriate data to test them. This will help you interact with AI, and evaluate what it returns.
You are responsible for all aspects of your work. AI is like advanced search, it’s helpful but use it as an input. You must carefully check and debug any input from AI.
AI can help you in all the steps, but you need strong foundations to become a scholar or an applied economist who uses AI but is not just a vehicle for LLMs. My advice is to use new tools but at the same time remain able to do most work without them. Enjoy “AI” with moderation: find a balance between learning the material as well as get practice with AI tools.
For my assignments I will be explicit of what you may use AI for. However, there is no way to really police the use of “AI” tools, so no point in trying. This implies ethical consideration for students – they will be encouraged to show self-restraint for the benefit of their future self.
Definitions
- I define “AI” here as Large Language Models based methods and tools including OpenAI’s ChatGPT, Github Copilot, Google’s Gemini, Anthropic’s Claude, Mistral’s Le Chat, DeepSeek, Facebook’s Llama and other available open source or propriatery models and tools.
- Prompting means communication with “AI” tools such as asking questions or checking code. This may happen in a chat or via API calls. It may be direct to operators of the model or via third party intermediaries like Azure AI Factory or OpenRouter.
Policies for assignments, papers, presentations
Reading papers:
You must read papers on your own first. Try suffering through harder bits to develop your economics and econometrics skills. But you are encouraged to use AI to ask questions about technical aspects of the model or the methodology as well as the institutional context.
Some suggestions: how can AI help
- Learn more about topics by having a conversation with AI on the material. You may copy text into it and ask for help in understanding details, and clarifications.
- Ask about context, background, relationships to other models, methods or theories.
- Help with derivations, maths.
Presentations.
You are expected to understand the paper you present deeply and create the content of the presentation on your own. But, as the presentation is expected to be well crafted, engaging and pretty and you are encouraged to use AI to assist in creating and enhancing the presentations. (You must use a markdown or tex based editor, no PP, please.)
Some suggestions: how can AI help
- Technical help re learning the use of markdown editors, quarto or latex/beamer.
- General help re formatting, simplifying slides
- Add graphics, videos. Avoid typical bland AI stuff, by now they are boring.
Short essay
For the short essay, my advice is start with review papers but also use AI to help you find papers. Come with the core idea first. Then you can also use AI to pin down (make it precise) the main idea for the paper. You shall go back and forth between thinking and AI review. Then, I suggest you write a full draft. Most people do the thinking while writing. Thus, the first draft should not be outsourced to AI. Then you can ask AI to give you feedback, like a referee report. You may well do a rewrite based on ideas (pick and choose and act on what you think is relevant, don’t accept all suggestions, as some are rubbish.). Then you can use AI to polish. However, beware that most AI have a somewhat artificial and bullshit-y language. So, focus on correct grammar, but make sure your voice remains.
Some suggestions: how can AI help
- Idea generation, thinking about causal pathways
- Language improvement, translation
Coding
You may use any AI assistance for coding. Note that you carry all responsibility for the correctness of your code. Please test and debug your code relentlessly. Moreover, the availibility of AI for coding means, I expect high quality code, well commented. Clear structure, functions, good coding practices.
You are responsible for your code no matter how and how wrote the actual lines.
You will be expected to understand all aspects of your code and be able to explain it clearly. I will sometimes ask you to do that in class.
Some suggestions: how can AI help
- Drafting code snippets, turning code bits into functions,
- Improving code, commenting it
- Translate from one language to another, like stata to R.
- Help create tables and graphs
- Debugging
Data Analysis report
You shall design data wrangling and plan do the analytical work first without AI. Decide on how the work data shall look like. Think about which methods you’d like to pursue and compare. Then comes AI, get help re actual work, find packages, ask help re comparing tools. In many cases, especially with common issues, AI will suggest great solutions. In other cases, such as graphics, AI will be great to get to about 80\% of what you’d want.
In terms of discussing your results, AI will be very good (like “A-“, sometimes even “A”) in simpler tasks, such as descriptive statistics and good but not perfect (like A- to B) in econometrics. It will be convincing, but often imprecise. I will always ask for maximum precision. Thus, it is vital that you check all code, output, texts.
Here as well, I urge you to try do as much as possible without AI first, and use AI as tutor (who can I do better) and reviewer (check if this makes sense). See the first point: you shall learn to do most alone so that you can confidently evaluate AI later.
AI attribution statement
I will ask all submitted work to include an AI attribution statement. In it you shall list what AI tools you used, and in what capacity. You can use IBM’s version, see description here I will also ask for a short (5-8 lines) AI interaction summary, where you describe how you used AI, how it worked, what was the experience, what you learnt from it.
It is mostly for statistical purposes. It will NOT affect your grade. It might also help you clarify how you are using AI.
More
Check out my Data Analysis with AI course