Skip to content

Latest commit

 

History

History
119 lines (89 loc) · 5.25 KB

2023-02-22-HackyHour-7.md

File metadata and controls

119 lines (89 loc) · 5.25 KB
Raw

How AI is changing the landscape of higher education: Discourse on a contemporary approach

Preface

  • We already talked about AI in HackyHour #6
  • This HackyHour was a collaboration with Hessenhub. They have a excellent website and currently try to foster the debate on generative AI in university teaching with a Open Affinity Group.
  • The slides are also available as pdf

What is intelligence

Intelligence is difficult concept to define. It refers to the ability to acquire,understand, and apply knowledge and skills and apply them in order to solve complex problems or adapt to new situations.

What is artificial intelligence

Artificial intelligence is the attempt to replicate certain decision-making structures of human being in order to solve complex problems or adapt to new situations.

Generative AI:

  • Generative AI uses machine learning technologies and algorithms.
  • The goal is to generate authentic content from existing data.
  • This content should be able to be perceived by humans as being of human origin.

How does generative AI work?

  • Artificial neural networks (KNN) are trained with large amounts of data.
  • Training requires complex and well-chosen training data.
  • After training, KNNs are fixed and can no longer learn.
  • Different KNNs can be combined.
  • All KNNs operate on the basis of probabilities.

Use cases for AI

  • Beauty Filter: used in photo apps and image editing software to optimize appearance, trained on large databases of human faces.

  • Image Diffusion Models (text-to-image): produce high-resolution results, that outperform already human-made artworks and give rise to a new form of digital art. Publicly available.imagen.research.google

  • Video Diffusion Models (Text-to-Video): generate controllable image sequences, but not yet publicly available.Google again, this time text-to-video

  • Voice synthesis (text-to-speech): reproduces linguistic idiosyncrasies and allows "cloning" of one's own voice. The model is trained using a few speech samples. Example: Leonardo Di Caprio at the UN with different voices of famous people

  • A fully trained generative AI needs only a few initial data to achieve impressive and "realistic" results, where the quality of the database and the training goal of the models play a crucial role.

  • Generation of X-ray images for training medical students. Individualized synthetic examples make reverse-search in exams more difficult and enable learning material even for edge cases. Paper on this via NCBI

  • AI as an assistant in searching papers. A kind of enhanced abstract that summarizes the paper itself.Elicit

Use-Case: ChatGPT as an assistant for scientific articles

  • "Setting" of ChatGPT via appropriate prompt
  • Reading in the article via text blocks
  • After that you can "ask" ChatGPT about the article
  • Explains e.g. abbreviations, recognizes translated terms
  • Be careful with numerical data, often go wrong.
  • Beware of eloquent nonsense. Do not forget to think for yourself.

Summary of use cases

Current use cases

  • Analysis of structures/patterns and solution of various tasks (medicine/economy/research/university)
  • Processing of routine work
  • Input provider for creative work
  • Mutual cooperation between humans and AI in development and conceptualization
  • Personalized analytics (more relevant to educators)
  • Personalized learning (more relevant for students/learners).

Three possible courses of action when dealing with AI.

  1. prohibit (very likely not possible)
  2. do nothing (does no work, ignores teaching mission of university)
  3. allow (transformational work necessary. Opportunity to teach critical and productive use of AI).

Implications independent of range of actions.

  • Critical thinking becomes more important
  • Information needs to be tested even more for reliability
  • Scientific work must be promoted
  • Awareness of relativity of knowledge and truth
  • Contextualization of information becomes more important (data literacy and AI literacy=digital literacy)

What does this mean for university teaching?

  • Question of plagiarism/deception & - legal assessment unclear so far.
  • How to react acutely (see handout FB06)?
  • Impact on courses?
  • Impact on examination formats?
  • Can the current (higher education) system handle the advent of generative AI? And if so, how quickly?
  • Which parts of the system are particularly affected?

Rethinking didactic concepts

  • Do not continue old concepts with AI. E.g. a video lecture produced with AI-tools.
  • Try to think of new concepts. E.g. AI learning assistent supervised by a professor.

DISCLAIMER: The above text was created with help from pdftotext (extract text), ChatGPT (summarizing text) and DeepL Translator (translating text). Although I have revised the text myself, errors may have crept in. If you notice any errors, please open a ticket or a pull request.