Artificial Intelligence

What is generative AI?

Generative AI is the term for a type of language models (‘chatbots’) that, with the help of artificial intelligence, creates a response based on the textual input you have given. Unlike a search engine like Google, which presents existing information from the web based on your textual input, generative AI comes up with answers based on the information it has been trained on. Your input can be a question or a task that you pose in a special way (called a ‘prompt’). Based on which chatbot you use and your input, the chatbot’s output can be text, images, tables, visualizations, graphs, code, or other.

Be aware!

The answers of generative AI seem very, very convincing. It’s important to know that the technology has limitations: The output can contain errors and even false data. 
There are several reasons for this: The language model does not necessarily have access to qualified knowledge within your field of study and at the level you as a student are expected to have. It may be that you do not describe enough context in your prompt, which makes the language model fill in the gaps itself.
It is not possible to see which sources the chatbot generates its answers from - and even if you incorporate a requirement for a source in the prompt, it is not always a correct source that is mentioned.

• Generative AI is NOT a search engine - you cannot validate output on the basis of source-critical method.
• Generative AI is a well-formulated conversation partner, where you should reflect critically on the validity and professionalism of the ‘conversation’.

How can I use chatbots in my studies? 

You can incorporate generative AI in many different phases of your study project. For example, you can use the technology:

• to form an initial overview of a complex topic, a theory or a concept, before you investigate it in depth.
• in the project’s idea phase to generate ideas and provide creative input. 
• as help to get started with your writing, by generating an outline with inspiration for relevant sections.
• as a sparring partner, who in the writing process generates suggestions for sentences or pieces of text, which you can work further with.
• to give feedback on your texts. Insert a piece of text and ask for feedback on e.g. grammar, structure or the clarity of the text. In this way, you can improve your writing skills.
• to make a summary of longer articles and texts, which are on the internet or which you insert yourself into the chatbot.
• to create illustrations and models for your assignment (certain tools).
• to translate text to speech (certain tools).

Mini guide to prompting

A prompt is the text input you give an AI chatbot, where you ask it to create a particular output. There are several ways to prompt: You can make a short prompt and ping-pong further with the chatbot, where you continuously ask to have the output trimmed. Or you can start by making a very comprehensive prompt. Generally, the more precise you can be about the desired output, the more precise answers you get.

In the article 'Getting started with prompts for text-based Generative AI tools' from Harvard University, you get tips for making good prompts. Here you can read, among other things, that you should:

1. be specific and set a concrete task
   The more precisely you can formulate the task you give the chatbot, the better output you will get.
   
2. assign roles to the AI and yourself
   Who is asking and who is answering - are you e.g. a meeting facilitator or a student, and should AI answer as a coach or a professional expert? Example: “You are an AI assistant who helps a student understand the concept behind quantum mechanics. Explain it in a simple way.”
   
3. control the output
   If you want a specific format, you should write it: Text, image, list, table layout or…? If it’s a text: What is the style and tone of language, what is the scope and layout of the output? Should it be sorted with headings? If it’s an image: What is the visual style, which angle is the image seen in (bird or frog perspective)? Feel free to use comparisons, it helps AI to find the desired expression (“like in a horror movie”).
   
4. give do’s & don’ts and examples
   If the chatbot may include something, but should omit something else, you should mention it. Also, feel free to give examples of something that resembles the output you would like to have.
   
5. adjust the output in a further dialogue
   Ask for adjustments and add further instructions until you are satisfied. These adjustments simulate a dialogue with the AI.

Prompt examples

Ethan Mollick is an American researcher who, among other things, investigates generative AI. In his blog post ‘Working with AI: Two paths to prompting’ you get examples of how you can prompt.

Do you need inspiration for how to start a prompt to get the chatbot to do something specific? Check out these prompt starters (pdf)

Be critical!

Generative AI is not a source in itself but a mishmash of many original sources, which you cannot get insight into. Therefore, you should be critical.

• Remember, it generally requires background knowledge before you can assess the quality of the output.
• You should always make several prompts on the same topic and not just accept the chatbot’s first output. You can easily experience getting different answers when you ask in different ways about the same topic.
• You can incorporate a requirement for source references in your prompt, but since an AI-generated answer is not necessarily correct or particularly qualified, neither answers nor references need to be correct. And - if the reference actually exists, it does not necessarily have the content the chatbot claims.
• Therefore, check the output with several real (practical as academic) sources you have found yourself.

There are many kinds of chatbots - generative AIs - and they have different strengths and weaknesses. The most well-known are probably:

• ChatGPT (developed by OpenAI)
• Bing (developed by Microsoft)
• Bard (developed by Google) 
  
  

They can do different things and ChatGPT and Bing are available in a paid version that can do more than the free version. Bard is still listed as an ‘experiment’ in Google and is free. Here you can link to a comparison of Bard, Bing and ChaGPT (pdf). You can also learn what you get out of upgrading to the paid version of ChatGPT. 

You have recorded one or more interviews for your project. It is not optimal to analyze the content across multiple audio files, so you need to have the file transcribed, i.e., 'translated' from audio to text.

This is a time-consuming process. Thankfully, there is help available! Use the Good Tape tool. It is developed by the Danish media company Zetland, and there are several advantages:

• It's free (create a profile: there are benefits for registered users compared to use without registration).
• It is Made in Denmark - we expect it meets with GDPR requirements.
• It is super easy to use.

This way, you can spend your time on the task instead of on tedious work!

Academic AI powered research tools

Elicit & Consensus

Helps you get started with your academic desk research! You pose a research question and the system searches for relevant literature based on your question. You can upload articles to generate summaries and ask article-specific questions.
Please note: Only upload articles you have obtained legally 😉

In the free versions, you have unlimited searches, but a limited number of advanced features.
You can read about the articles, but like in Google Scholar, you need to find the full version yourself. Read our article on how to access academic articles.

Elicit and Consensus both source information from the Semantic Scholar platform, whose content is based on partner publishers. Therefore, not all existing academic literature can be searched and presented! 

ExplainPaper

Upload an academic article - highlight text you don't understand, and get an explanation at a different academic level, ie. high school level.

The free version can help you understand sections in academic articles and is based on chatGPT 3.5. The paid version can do more, such as summarizing main points.

Please note: Only upload articles you have obtained legally 😉

Background Knowledge and Critical Thinking

We have mentioned a fraction of all the AI-based study tools you can find. They are selected based on reviews and usability. They can make parts of your work faster and more efficient than if you were to do the same manually, but they are not flawless. A common limitation is that they inevitably reproduce errors and biases from the data they draw on, and may be trained on a limited pool of articles.

To determine if the tools' analyses are good enough, you need to have a basic knowledge of the subject you want to investigate. You can get that from a textbook on the topic!

Visit the library and talk to the librarians, or search & order through EAAA's library base - or search & order from all of Denmark's libraries via bibliotek.dk.

Artificial intelligence can generate illustrations for your assignment. Using an AI, you don't need to worry about violating the copyright of an artist.

Here are 2 free tools:

• Bing Copilot
  Microsofts browser Bing includes the chatbot Copilot (link to Bing and press 'Chat'). Copilot is based on ChatGPT4 and generates illustrations and tables using the DALL-E technology.
  Download an example of a Copilot generated illustration and table.
  You'll need to create an email-based account to use Copilot.
  If you're logged in to your EAAA Microsoft 365-account already, the free version of Copilot is part of the package.
  
  
• Adobe Firefly
  Adobes Firefly is a powerful (and still free) image generator. You can edit the generated images using pre-defined styles.
  If you have a license for Adobe Photoshop, you can drop the Firefly images directly and do some more creative work.
  Download an example of a Firefly generated illustration.
  You'll need to create an email-based account to use Firefly.

If you use an AI generated image or table in your assignment, you'll need to refer to the AI below the illustration: 'Generated in Bing Copilot/Adobe Firefly' and remember to register in your source list. Read how to in the 'Student guide on using AI' (StudyUpdate)

Generative AI may become more than just a sounding board.

If you're studying graphic design and using an AI-created illustration, the chatbot is a creative co-creator. 
If you're using AI-generated personas or interviewées, in your study report, the bot is also more than a sounding board.

When generative AI becomes a creative co-creator, you'll need to give reasons for your choices in the method chapter. Furthermore you need to document the process in the appendixes.

How to document the process in the appendix section:

1. Choose the most important bits of the process, where AI have been co-creator - ie as screenshots.
2. Comment  the curated parts:  In which way were they essential for your problem solving?

As a student, you'll find all the official rules and regulations in the StudyUpdate menu 'Examinations' (login needed). 

We'll point to the guides 'Generative AI' (updated every semester) and 'Worth knowing about exams'. You'll need to be thoroughly updated on both before heading into AI (and don't (!) use AI during exam unless using AI is the goal of the exam).

Artificial intelligence has rapidly become an indispensable technology throughout society and in many workplaces. Therefore, it is important for students and staff at Business Academy Aarhus to become familiar with using artificial intelligence where it makes sense - and in a responsible manner.

At the academy, we continuously strive to stay abreast of developments, including:

• Establishing an internal research forum comprised of staff and management who monitor technological advancements and experiment with AI technologies in teaching.
• Conducting research on the intersection of artificial intelligence and education, including how the use of generative AI impacts student learning.
• Continuously developing guidelines for the use of artificial intelligence for both students and faculty. These guidelines are revised every six months to align as closely as possible with current developments.