Skip to main content
SDU IT

Search

Dansk
Menu

SDU IT

Recognise opportunities and challenges

Learn about the potentials and challenges of different types of AI technologies, with GenAI being the common element.

Don't get carried away

  • The AI chatbot is not a human. It responds politely and empathetically. It's a robot programmed to create the illusion that you're chatting to a human.

    Remember: Chatbots can't think, feel, remember or empathise - they're programmed to pretend.

  • Getting the AI chatbot to respond in the format and quality you want can be a challenge. Often you already have an expectation of the answer when you ask the AI chatbot a question, and then you're disappointed when it doesn't live up to that expectation.


    Sometimes it's due to the AI chatbot's capabilities, but just as often it's due to you. You get the answers you ask for. AI chatbots are not mind readers, so you need to be clear and specific in what you ask the AI chatbot.

  • There are many AI chatbots on the market, each with their own strengths and weaknesses, such as text processing and programming. In addition, they are constantly changing, as the TECH companies behind AI chatbots are constantly releasing new versions and changing the algorithms and conditions of use.


    They respond differently to the same prompt, and it's a matter of temperament whether you like the AI chatbot's conversational style and way of responding. The impression you get of a given AI chatbot today will most likely change in a month's time.

    Tip: Vary your use of AI chatbots to get the most out of them.

  • AI chatbots are more than just a search engine. Search engines list a wide range of results that you have to sift through to find the answer. The AI chatbot, on the other hand, provides an answer that often cuts across the many search results. It selects what it thinks is the most likely answer to your query.

AI is much more than ChatGPT

Artificial Intelligence (AI) covers many different technologies, but often the entire field of AI is confused with generative AI like Copilot and ChatGPT.

Dive in and learn about the potentials and challenges of the different types of AI.

GenAI refers to AI models that generate content such as text, images or code and are trained on large amounts of data, such as GPT models.

  • General purpose AI chatbots cover knowledge in a broad sense. You can ask anything from food recipes to something very specific.
  • AI chatbots that are specialised for a specific purpose, such as reference tools or teaching mentors.

  • Open source is AI chatbots and language models that are made available for you to work with.

    Closed source are licensed AI chatbots and language models.

  • The majority of AI chatbots are located in the cloud. This means that language models and AI chatbots are located in large data centres around the world and can be accessed from there. The language models and AI chatbots can also be downloaded or developed on your own servers. SDU has several servers in the basement for this purpose.

Image recognition is often used in creative and general contexts, for example to analyse, classify or generate content based on images as object recognition or style transfer.

Models such as Convolutional Neural Networks (CNNs) or vision transformers trained on large datasets fall into this category. Results often require validation because they can be prone to error or bias.

Potentials

  • Administrative tasks, such as automating document preparation, email replies, reporting, summarising meeting notes and generating standard contracts.
  • Research support, such as generating hypotheses, literature summaries or draft articles. It can also help formulate complex ideas quickly.
  • Education, such as creating personalised learning materials, exercises or explanations and can act as a virtual tutor for students.

Challenges

There is a risk of inaccurate or fabricated answers (hallucinations).

Diffusion models are a type of generative AI that uses stochastic processes to gradually transform noise into structured data, such as images or sound.

The model learns to reconstruct data by reversing a diffusion process where information has been gradually added as noise. This approach has proven particularly effective in image generation and is used in models such as DALL-E 2 and Stable Diffusion.

Potentials

  • Education, e.g. creating illustrative images for course materials or interactive learning environments.
  • Research support,  e.g. visualising complex data or simulations - for example in medical imaging or materials research.
  • Creative production, e.g. generating realistic images, artworks or visual concepts based on text descriptions - used in design, advertising and entertainment.

Challenges

High computational requirements and risk of generating unethical or copyrighted content. Requires responsible use and control mechanisms.

 

RAG combines generative AI with a search function that pulls relevant data from an external database or documents to provide more precise and context-specific answers.

Potentials

  • Administrative tasks, fe.g. improving document management by finding and summarising specific information from large archives - for example in HR or legal processes.
  • Research support, e.g. helping researchers find and cite relevant articles or data quickly. Ideal for systematic literature reviews. 
  • Education, e.g. giving students access to accurate, source-specific answers from course materials or knowledge bases, supporting deeper learning. 

Challenges

Depends on the quality of data sources and search algorithms.

Agent systems are AI systems that autonomously perform complex tasks by combining tools, memory and decision-making. For example, an AI that plans, searches and executes actions.

Potentials

  • Administrative tasks, e.g. automating workflows such as scheduling, resource allocation or customer service by coordinating multiple steps - e.g. booking and follow-up.
  • Research support, such as managing research processes including data collection, analysis and visualisation by integrating external tools like Python or databases. 
  • Education, such as supporting project-based learning by guiding students through complex tasks by suggesting next steps or finding resources.

Challenges to overcome 

Requires robust design to avoid errors in complex decision chains.

Edge solutions are AI models - often small and optimised language models - that run locally on devices such as smartphones or IoT devices. This provides low response time, better data protection and the ability to use the solution offline.

Potentials

  • Administrative tasks, enables local processing of sensitive data, e.g. in the healthcare or financial sector - without dependence on the cloud, increasing security.
  • Research support, supports field research by running analyses on portable devices, e.g. in environmental monitoring or data collection in remote areas.
  • Education, provides access to AI tools in classrooms without internet, such as translation, quizzes or learning apps on tablets.

Challenges

Limited computing power and modelling capacity compared to cloud-based solutions.

 

Do you have any questions?

You are always welcome to contact us if you have questions or need sparring about the use of AI.

See also