This book has specific content depending on your educational programme and operating system. Please select them in the bar above!

Artificial intelligence and chat services

We felt the need to put this in a separate sub-chapter because this is a very important topic. This chapter is our way of looking out for you!

This will start negatively, but we will have a part of this section dedicated to good uses for AI in your studies! The debate around AI is complex and intricate. It's not a black-and-white issue which is what this chapter's about. We want to highlight it's weaknesses, it's strengths, and the risks of using AI excessively.

In recent years, the usage of AI assistants in the field of computer science has truly exploded. Several options are available, all of which can assist you in your programming. Asking a chatbot a question such as "Write a program that calculates X digits of pi in Python" can provide a useful scaffold for your own code.

However, AI tools also come with risks. Most courses expect you to be able to program, write, and demonstrate your engineering skills without AI tools. They are very rarely allowed during exams and only sometimes permitted in labs. The rules for using AI in a course are determined by the examiner for that individual course. Chalmers currently has no overarching rules, so ask your examiner before you use any AI at all. Information regarding AI specifically for a bachelors/masters thesis can be found here but the same typically applies for ordinary courses as well.

These are the most important aspects taken straight from the above page but we want to include them here because this is very important:

• Being cautious of over-reliance on AI, which includes the risks of uncritically accepting AI-generated answers that can be biased or wrong. For example, do not uncritically accept a rewrite of your text, code, literature review, or analysis of your data.
• Understanding the potential for plagiarism and copyright infringements. For example, GitHub Copilot has used copyright protected code when helping programmers to code, so always be wary of such problems.
• That the use of AI tools must be clear and transparent. At a minimum, provide a description detailing how and to what extent AI tools have been utilized in your work. If you have used AI generated content in you thesis, this should be cited as other works.
• Recognizing the need to secure sensitive or proprietary information. For example, you might have access to valuable information in the organization you are writing your thesis and sharing it with the AI tool might not be advisable.

Using unapproved tools could lead to serious consequences—including suspension from your studies. It is better to talk to your teacher and try to submit something late than to submit something you do not understand yourself.

You might have heard about how big companies want to use AI and how they expect programmers to be able to use these chat services. But what usually isn't included in the headlines is that they mean as a tool.

Just like google can be a tool to help you learn things, so can AI chat services. But just like google, if you search up code that you didn't write and turn it in, that's a form of plagiarism.

You might question calling it plagiarism, but the way AI, or large language models work is they are trained on large sets of data. For programming this means they are trained on other peoples code, this is all the context they have for programming. Which means all they can do is rewrite and reuse that code they were trained on. This lets them come up with more complex functions and string together functions they have been trained on, but it also means someone else originally wrote that code.

Our goal here isn't to scare you, but we do want you to take this seriously. This university, like all universities in Sweden takes plagiarism and cheating very very seriously, and there can be very dire consequences.

The limits of AI

Additionally, AI doesn't actually understand what it's doing or saying. That might sound like a surprise, but what AI does is string together information based on it's training using statistical probability to determine a "correct" answer. This is why it can be very bad for math, or for something very basic like counting objects or elements.

As an example the ChatGPT models o4-mini and o4 give different answers for how many letters are in certain words, like strawberry. They're generating a response based on patterns, not based on reasoning.

On top of this, this is why it can be bad for programming and why you need to understand the code it tries to give you. For more esoteric, or uncommonly used languages like Haskell or VHDL, it might not actually know how a function works and just guess. These sorts of answers are what we call hallucinations.

It also often can't tie together several files or big projects, because it gets lost in the information it has received. When there's too much information it's easy for AI to get focused in on one thing or to misunderstand what you've written. It's great for small tasks and self-contained tools, because it's often similar to something it has read or been trained on before.

Using AI as an effective tool.

That said, learning to program with AI rather than relying on it can make you a much stronger and a very sought after developer. AI can help you get quick feedback, it can help you understand why a function isn't doing what you expect. It can help you troubleshoot errors that you don't understand yet. It can give you an example function, similar to what you want to write. Seeing what a working function looks like in this language using the same concepts can help you write your own code. In the same vein it can give you a starting point, what we usually call a skeleton. It won't contain all the code you need but it can help you with filler code or to create a structure to follow.

Sometimes documentation can be unclear, or have very few words and just be code. Copy the documentation and ask AI to explain it in plain language. Instead of asking it to write code to do a certain task, ask it to give you an idea for how that sort of function could be written! Do you have repetitive code you need to write but you just need to fill something in? AI can usually do the exact same thing. It's great at that sort of thing! When you give it something to write again and again, or something to summarize, it can be very effective. These ideas for how to use AI are how we want you to learn to use AI if you want to do so. AI is a supportive tool, it's not a shortcut.

But you need to understand what it does. You need to be able to write the code yourself. You are the programmer, not the AI. It is easy to write code that looks good, especially for inexperienced programmers. But teachers and examiners can often recognize AI generated code, and it's usually obvious when a student can't explain or replicate what they've submitted.

As an example, in lots of AI generated text you can see lots of em-dashes (—) being used. These are often used in fiction which AI is often trained on. In informative texts like this, excessive and out of context use can often be an indicator of AI. Stylistic choices like this aren't always an indicator of AI, some people do write like that, but it can definitely suggest the use of AI.

Although usually helpful, sometimes AI can get things very wrong. Your calculator won't correct you if you wrote the wrong number and will instead happily output the answer, not knowing that the inputs were bad. AI works the same way but works on language instead of math.

As an example for how AI can be used we've asked ChatGPT to restructure the above text for clarity. But read it as an exercise, see what it has rewritten, notice how it commonly styles text. There are many tells in the below text which an experienced reader can find in an instant. Some sentences are formatted very poorly because AI is based on text written by humans, who are also prone to errors. At times it also subtly changes the meaning of the text and what it actually means.

We also had to go through and reformat part of the text to make the source-code look readable at all and to even produce a correct result. A direct copy-paste will almost always be incorrect in some way, and formatting like that can be very important for code.

The exact prompt we gave was: "Can you rewrite this text for clarity? Note that the block under the chalmers.se link is directly quoted from that page." Without the comment regarding the quotes it also rewrites quotes, which could lead to incorrect citations.