AI Guidelines

Generative AI is here, it is useful in many situations and is quickly getting more capable than most of us could imagine just a few years ago. Personally, I experience AI mostly in a positive way, perhaps most importantly by the new and exciting feeling that so many more ideas and projects are within reach, because much of what stopped me before (lack of time, lack of specific skills) is not really relevant anymore. When learning something new, however, we need to keep in mind that the struggle and the friction, the slow and deliberate process of trial and error, are the exact conditions that will make us succeed. Artificial Intelligence is almost certainly useful in the learning process as well, but it does require a lot of self-control, likely more than you think, to avoid creating a situation parallel to paying a personal trainer to do the workout for you.

I will attempt below to indicate how generative AI can be useful in this course, so that you can make sensible choices when nobody is watching.

The formal rule for the compulsory project is on the assignment page, and that one is binding. Everything else is advice.

What this course is for

The content of this course is not hard to obtain. Every definition, every theorem and most of the proofs in the lecture notes can be produced by a chatbot in a few seconds, and they will usually be correct, more correct, even, than if we were to establish some of the results on our own. If the purpose of the course were to end up in possession of these statements, there would be little reason to run it.

The purpose is that you can do the reasoning yourself, and more fundamentally: To achieve some basic understanding of the way we think when doing statistical inference. When we spend a lecture pushing through the sampling distribution of \((n-1)S^2/\sigma^2\), the value is not the formula at the end. It is that you have seen how a long constructive argument is assembled, how the logic works, and that you believe it because you followed every step. The arguments in this course are, as I write in the notes, not very advanced, but sometimes a bit long. Getting through them is the training, and it is not the kind of training that can be outsourced.

Where AI tools might help

There are places where I think such tools might work well in a learning situation:

A second explanation. If a step is still opaque after you have honestly tried it, asking for another route through it is sensible, and usually faster than waiting for help by the instructor.
Criticism of your own work, after you have done it. Write the proof or solve the exercise, then make the AI model critique your work.
Orientation. “What is the difference between convergence in probability and almost sure convergence?” is a perfectly good question to put to a machine after having reflected on the issue yourself. You can ask for a wider landscape (What is the bigger context where this and that concept belongs?, Where can I read more about it?), a narrower focus (Help me work out the details of this particular step in a proof), or a new connection (Is there a common way to think of concept A and concept B?).
Language. If English is not your first language, use whatever help you need to say clearly what you already think. The language is not what I am assessing, but clear writing supports clear thinking.

Sometimes you will miss the point by using AI

My worry is not cheating, because you would mainly be cheating yourself, not me. The main worry from a pedagogical point of view is that the tool is good enough to remove the productive struggle without you noticing that you have lost something important.

Use these tools deliberately. Before you open a chat window, know what you are asking for, and why you could not get there yourself. If it is just a matter of time, then stop. Take your time, think and work, because that is the whole point. Afterwards, check whether you could now reproduce the argument on a blank sheet of paper. If not, then reflect on what you do not yet understand and work from there.

And do talk to each other. Discussion with other humans is allowed and encouraged at every stage of this course, and it remains, in my experience, the better tool.