Responsible AI Use at the St. Pölten UAS

Rules, Data Protection, Tools

Photo: Shutterstock

Generative AI tools such as ChatGPT or Copilot can be helpful in work and study contexts – for example, for generating ideas, structuring content, or increasing efficiency.

However, their use requires special caution: Sensitive or personal information should not be entered, as these tools do not guarantee confidentiality.

Responsible use involves considering aspects such as information security, data protection, legal requirements, copyright, and academic integrity.

Guidance is provided by the data classification guide of the St. Pölten UAS – it supports risk assessment and the selection of appropriate protective measures.
You can find more details further down on this page.

Jump directly to the topic:

AI Tools | Data Classification Guide for AI Tools | AI Glossary

AI Tools

Details on data classification (referred to "Classification" in the table) can be found further below. 

Tool Description Classification Costs?
Adobe Firefly Create images, text effects, and videos within Adobe creative applications using AI Level 4 only Yes
Big Interview AI-assisted interview preparation with mock interviews and personalised feedback systems Level 4 only Yes
Canva AI-supported design platform for creating graphics, images, videos, and presentations Level 4 only Only for the premium version
ChatGPT Conversational AI assistant for text generation, analysis, and creative tasks Level 4 only Only for the premium version
Claude Advanced AI assistant for text analysis, programming, and complex reasoning tasks Level 4 only Only for the premium version
Consensus AI-assisted research tool for finding and analysing scientific literature and studies Level 4 only Only for the premium version
Copilot Conversational AI assistant from Microsoft for text generation, analysis, and creative tasks | safe when logged in with the St. Pölten UAS account Level 4 only No
Copilot for Microsoft 365 AI assistant integrated into Microsoft 365 applications for productivity enhancement  Levels 2–4 Yes
Copilot with Enterprise Data Protection Conversational AI assistant from Microsoft for text generation, analysis, and creative tasks | safe when logged in with the St. Pölten UAS account Levels 2–4 Not for the St. Pölten UAS
Cursor AI-powered code editor with intelligent code completion and generation capabilities Level 4 only Only for the premium version
DALL-E OpenAI's image generation model creating images from detailed text descriptions Level 4 only Yes
DeepL Translator/Write Advanced AI translation service supporting multiple languages with high accuracy Level 4 only Only for the premium version
Eleven Labs AI voice generation and speech synthesis platform for realistic audio content Level 4 only Only for the premium version
Elicit AI research assistant for literature reviews and scientific paper analysis Level 4 only Only for the premium version
Gamma AI presentation creator that produces slides and documents from simple text prompts Level 4 only Only for the premium version
GitHub Copilot AI pair programmer that offers code suggestions and completions within development environments | free for verified students and educators Level 4 only Not for the St. Pölten UAS
Google Gemini Google's conversational AI assistant for text generation, analysis, and creative tasks Level 4 only Only for the premium version
Gradescope AI-powered grading and plagiarism detection platform for educational assessments Levels 2–4 Yes
Grammarly AI writing assistant for grammar checking, style improvement, and content optimisation Level 4 only Only for the premium version
LitMaps Research visualisation tool that creates literature maps to explore academic paper connections Level 4 only Only for the premium version
Midjourney Premium AI image generator known for artistic and creative visual outputs Level 4 only Yes
nanoHUB Educational platform that provides computational tools and simulations for scientific learning Level 4 only No
NotebookLM Google's AI note-taking and research assistant for organising and analysing information Level 4 only Only for the premium version
Perplexity AI search engine that gives comprehensive answers with source citations and references Level 4 only Only for the premium version
PlayHT Text-to-speech AI platform that creates natural-sounding voice recordings from written content Level 4 only Only for the premium version
PopAi Multi-purpose AI assistant for document analysis, chat, and content generation Level 4 only Only for the premium version
QuestionWell AI tool that generates educational questions and assessments from input content Level 4 only Only for the premium version
Research Rabbit Literature discovery platform that uses AI to find and connect relevant research papers Level 4 only No
ScholarAI AI summarisation tool that converts long articles into structured summary flashcards Level 4 only Only for the premium version
Scholarcy An online summarising tool that generates and converts long articles into summary flashcards Level 4 only Only for the premium version
SciSpace Comprehensive research platform with AI tools for paper discovery and analysis Level 4 only Only for the premium version
Scite Citation analysis platform that helps researchers to evaluate and understand scientific literature Level 4 only Yes
Semantic Scholar AI-assisted academic search engine for finding and analysing scientific publications Level 4 only No
Stable Diffusion Open-source AI image generator that creates high-quality images from text descriptions Level 4 only Yes
Wolfram Alpha Computational knowledge engine that provides mathematical calculations and factual answers Level 4 only Only for the premium version

Data Classification Guide for AI Tools

Data classification is a systematic approach to organising information based on its sensitivity level and the potential impact of unauthorised disclosure. It helps ensure the responsible use of AI tools in accordance with the AI guidelines of the St. Pölten UAS and the requirements of the EU AI Act.

The following classification provides guidance on which types of information are suitable for AI-supported applications – and which are not. 

Overview:

Level 1 – Highly Sensitive Information

Most restricted | Highest security required

Examples:

  • Health information
  • Biometric data
  • Financial identifiers

Not permitted:

  • Biometric identification and categorisation
  • Emotion recognition (except for medical or security purposes)
  • Automated collection of biometric data from the internet
  • Processing of special categories of personal data without pseudonymisation and a clearly defined purpose


Level 2 – Sensitive Information

Confidential | Legal protection required

Examples:

  • Employee and student data
  • Examination results, applications
  • Unpublished research data

Not permitted:

  • Social scoring by AI
  • Manipulative systems or those exploiting human vulnerabilities
  • Emotion recognition in educational contexts
  • Risk assessments based solely on profiling


Level 3 – Security-Relevant Information

Internal use | Potential security implications

Examples:

  • Administrative and operational data
  • Infrastructure and security information
  • Unpublished academic content

Required:

  • Transparency about AI usage
  • Compliance with GDPR
  • Avoidance of manipulative applications
  • Human control in risk assessments


Level 4 – Non-Sensitive Information

Public or intended for publication

Examples:

  • General business information
  • Published research and teaching materials
  • Publicly accessible data

Basic restrictions also apply here:

  • Users must be informed about AI usage
  • Misleading applications are prohibited
  • Data protection principles such as data minimisation must be observed
  • The use of AI must be documented

AI Glossary

Familiarise yourself with key AI concepts using the glossary:

Term Definition
Artificial Intelligence (AI) The simulation of human intelligence in machines that can perform tasks such as learning, drawing conclusions, and problem-solving
Artificial General Intelligence (AGI) A theoretical AI system that can perform any intellectual task a human can, including reasoning, learning, and adaptability
Artificial Narrow Intelligence (ANI)

Also known as weak AI, it is designed to perform a single task (e.g., speech recognition, recommendation algorithms)

Artificial Super Intelligence (ASI) A hypothetical AI that surpasses human intelligence in all domains, including creativity and emotional intelligence
Algorithm A set of step-by-step instructions followed by a computer to complete a specific task
Bias Systematic errors or unfair preferences in AI outputs due to biased training data or flawed algorithms
Burstiness The irregular occurrence of high-quality AI-generated content followed by less coherent outputs
Business Value of AI The economic and strategic benefits of AI across industries, including automation and efficiency gains
Chatbot A software application that mimics human conversation through text or voice interfaces
ChatGPT A generative chatbot developed by OpenAI that generates human-like text based on context
Computer Vision A field of AI that enables machines to interpret and understand visual information from the world
Conversational AI AI systems designed to simulate human-like conversations, such as chatbots and virtual assistants
Data Augmentation Techniques used to increase the diversity of training data by modifying existing data samples
Data Mining The process of analysing large datasets to find patterns, relationships, and useful insights
Data Science The interdisciplinary field that uses statistics, AI, and computing to analyse and interpret complex data
Deep Learning A subset of machine learning that uses neural networks with multiple layers to learn patterns in data
Embeddings Representations of words, images, or other data as numerical vectors in AI models
Ethical AI The study and implementation of AI systems that are fair, unbiased, and aligned with human values
Explainability (XAI) The ability of an AI system to explain its decision-making process in a way humans can understand
Few Shot Learning A machine learning technique where a model learns from a very small amount of labelled data
Fine-Tuning Adjusting a pre-trained AI model with additional training data to specialise it for a particular task
Foundation Model A large-scale AI model trained on diverse datasets that can be adapted for different tasks (e.g., GPT, BERT)
Generative AI AI that creates new content such as text, images, music, and code
Generative Adversarial Network (GAN) A type of AI model where two networks compete to improve the quality of generated data
Generative Pre-trained Transformer (GPT) A type of AI model that predicts text using deep learning and a transformer architecture
Hallucination When an AI model generates incorrect, misleading, or entirely fictional information
Heat Map A visualisation tool used in AI to highlight problematic areas in a dataset or a model’s decision-making
Hyperparameters Adjustable settings in an AI model that influence learning, such as learning rate and number of layers
Inference The process of using a trained AI model to make predictions on new data
Internet of Things (IoT) A network of connected physical devices that collect and exchange data
Knowledge Graph A structured representation of information that shows relationships between different entities
Language Model (LM) An AI system designed to understand, generate, and process human language
Large Language Model (LLM) A powerful AI model trained on vast amounts of text data to generate human-like text (e.g., GPT-4)
Latent Space The abstract space in which AI models organise and process complex data representations
Machine Learning (ML) A branch of AI where algorithms learn from data and improve performance without explicit programming
Model Drift The phenomenon where an AI model’s accuracy degrades over time due to changing data patterns
Multimodal AI AI that processes multiple types of data inputs such as text, images, and audio
Natural Language Processing (NLP) AI that enables computers to understand, interpret, and generate human language
Neural Network A computer system modelled after the human brain that processes information through layers of nodes
Neuro-Symbolic AI A hybrid AI approach that combines neural networks and symbolic reasoning
Output The result generated by an AI system such as text, images, or predictions
Overfitting When an AI model learns patterns too specific to training data, reducing its ability to generalise
Parameter A variable that AI models adjust during training to improve predictions
Perplexity A metric used to measure how well a language model predicts text; lower perplexity indicates better performance
Positional Encoding A technique that helps AI models understand word order in text data
Predictive Analytics The use of AI and statistics to forecast future outcomes based on historical data
Probabilistic AI AI that incorporates probabilities in decision-making to handle uncertainty
Prompt The input text given to an AI system to generate a response
Prompt Engineering The practice of designing effective prompts to guide AI models toward desired outputs
Quantum AI The application of quantum computing to enhance AI algorithms and processing capabilities
Reinforcement Learning (RL) A type of ML where AI learns by receiving rewards or penalties for actions taken
Self-Supervised Learning A machine learning approach where AI generates its own labels from raw data
Semi-Supervised Learning A mix of supervised and unsupervised learning where AI learns from both labelled and unlabelled data
Sentient AI A hypothetical AI system capable of consciousness, emotions, and self-awareness
Supervised Learning A machine learning approach where AI learns from labelled training data
Synthetic Data Artificially generated data used to train AI models when real data is scarce or sensitive
Temperature (in AI) A parameter that controls randomness in AI-generated text; higher values lead to more creative responses
Text Classification Categorising text into predefined labels (e.g., spam detection, sentiment analysis)
Tokens The basic units (words, subwords, or characters) that AI models process in NLP tasks
Transfer Learning Using a pre-trained AI model on a new but related task to improve efficiency
Transformer Model An AI model architecture that excels at processing sequential data like text (e.g., GPT, BERT)
Turing Test A test proposed by Alan Turing to determine if an AI can exhibit human-like intelligence
Underfitting When an AI model is too simple to learn meaningful patterns from training data
Unsupervised Learning A type of ML where AI finds patterns in data without labelled examples
Zero-Shot Learning AI’s ability to perform tasks without prior exposure to similar examples

Learn more about AI at the St. Pölten UAS

You want to know more? Feel free to ask!

AI Officers at the St. Pölten UAS