DGB 3: AI in Software and Physical Systems
Artificial intelligence is used in recommendation systems, translation, image recognition, navigation, medicine, industry and increasingly in physical devices. A useful basic model separates the data that is collected, the model that processes it, the decision or prediction that follows and any action carried out by software or a machine.
Curriculum reference
[Bearbeiten]| Area | Classification |
|---|---|
| Competence area | Orientation |
| Perspectives | T, G |
| Focus | Artificial intelligence controls software and physical systems |
| Competence | Describe how artificial intelligence supports or controls software and physical systems and distinguish between data, model, decision and action. |
Learning objectives
[Bearbeiten]Students can:
- explain that AI systems work with data, models and feedback;
- describe AI applications in software systems;
- describe AI applications in physical systems;
- distinguish data collection, pattern recognition, decision and action;
- place sensors and actuators in a simple control chain;
- classify examples from everyday life, mobility, medicine, industry and communication;
- recognise errors, bias and limitations in AI-supported decisions;
- explain why human responsibility remains important.
Learning content
[Bearbeiten]Basic principle of AI control
[Bearbeiten]A simplified AI chain can be represented as data → model → output → decision → action. Not every system contains every step, but the model helps distinguish what is measured, what is inferred and what is actually done.
AI in software systems
[Bearbeiten]Search, translation, recommendation, spam filtering, image recognition and assistance systems can use statistical models to classify, predict or generate results. The output is not the same as certainty and should be interpreted in context.
AI in physical systems
[Bearbeiten]Robots, driver-assistance systems, smart devices and industrial machines combine software with sensors and actuators. The physical consequences of a wrong decision can be greater than in a purely screen-based application.
Applications in everyday life and society
[Bearbeiten]AI-supported systems can improve accessibility, efficiency and analysis, but they may also affect privacy, employment, decision-making and access to services. The same technology can be useful in one context and unsuitable in another.
Data, training and feedback
[Bearbeiten]Models learn patterns from training data. The quality, scope and representativeness of this data influence results. Feedback can be used to improve a system, but feedback loops can also reinforce existing patterns.
Opportunities, limits and responsibility
[Bearbeiten]AI systems can process large amounts of data quickly, yet they may be wrong, biased or difficult to explain. Users and organisations need to check when human judgement, transparency or additional safeguards are required.
Classroom activities
[Bearbeiten]Classify AI applications
[Bearbeiten]Sort examples into software-only and physical systems and explain the distinction.
Expected outcome: A reasoned classification.
Analyse control chains
[Bearbeiten]For a selected system, identify sensors/data, processing/model, decision and output/action.
Expected outcome: A labelled control chain.
Compare software and machines
[Bearbeiten]Compare the consequences of an error in a recommendation system and an error in a physical control system.
Expected outcome: A risk comparison.
Document everyday observations
[Bearbeiten]Find AI-supported functions in everyday digital services without collecting private data.
Expected outcome: A short observation log.
Investigate failure scenarios
[Bearbeiten]Analyse what could go wrong if data is incomplete, biased or misread.
Expected outcome: A fault-and-consequence table.
Discuss stakeholder roles
[Bearbeiten]Debate the responsibilities of developers, providers, users and regulators.
Expected outcome: A responsibility map.
Check media portrayals of AI
[Bearbeiten]Compare a sensational and a factual description of the same AI application.
Expected outcome: A critical media analysis.
Design a simple AI-supported system
[Bearbeiten]Sketch a system with input data, model output, human check and possible action.
Expected outcome: A system design.
Reflect on limits
[Bearbeiten]Write three situations in which AI output should not be accepted without further checking.
Expected outcome: A justified reflection.
Self-check
[Bearbeiten]- I can distinguish data, model, decision and action.
- I can describe AI in both software and physical systems.
- I understand the roles of sensors and actuators.
- I can explain why training data affects results.
- I can identify possible errors, bias and limits.
- I can explain why responsibility cannot simply be delegated to AI.