Real examples of artificial intelligence (AI) in everyday life

Artificial intelligence (AI) is permeating all aspects of our lives: it can write scientific papers or personal letters for you, create and analyse your job application, monitor your health data and control your intake of medication, find you someone to love for the rest of your life - or even be that person!

AI-supported data processing is increasing rapidly. Artificial intelligence is found in many forms, ranging from simple decision trees to complex neural networks. With the promise of increased efficiency and reduced operating costs, more and more businesses and indeed ordinary members of the public are turning to AI for help with their day-to-day work.

Normally, we can choose whether to make use of artificial intelligence or not, and often it can make our lives easier and relieve us of everyday tasks. Sometimes, however, we can't avoid it and inadvertently or even unwillingly enter its sphere of influence.

If only factual data or information that does not refer to specific people is being processed in AI systems, the processing is generally not relevant under data protection law. However, as soon as personal data is processed using AI, the applicable data protection requirements must be met. The Data Protection Act (FADP) is formulated in a technology-neutral manner and so also applies to new technologies.

The use of information to train AI models and the processing of information using AI are therefore relevant under data protection law if the information relates to natural persons. AI can also be used to create references to specific individuals by linking different data sets, i.e. data that have been pseudonymised or even anonymised can be reassigned to a specific person with the help of AI. In all these cases, the Data Protection Act provides the legal framework.

The principles relevant to processing, which must also be observed for the use of AI, are set out in article 6 FADP. Processing must be:

• recognisable / transparent
• proportionate
• for a specific limited purpose
• carried out in good faith
• carried out using data that are accurate

• Art. 6 FADP

The principle that the purpose must be specific and limited is often compromised when training AI models, as the data used were originally collected in a different context for a different purpose. Any change of purpose requires either the consent of the data subject or appropriate justification, in particular on the grounds set out in Article 31 paragraph 2 letter e FADP.

• Art. 31 FADP

Examples of the use of AI in everyday life:

• You use a service that suggests suitable sentences for replying to messages, adapting them to your personal circumstances and writing style in response to your input.

  There may be personal data in the messages. In order to make suitable suggestions, the data are stored, at least temporarily, usually in a cloud. In order to adapt to your writing style, the service also collects data on how you have used the service in the past. As a result, user data are also processed on a personalised basis.

• You ask ChatGPT or another Large Language Model (LLM) to edit your cover letter and CV when applying for your dream job.

  These documents contain your personal data, which are stored at least temporarily.

• A website’s customer service chatbot offers to assist you.

  If the chatbot is supposed to respond to customers’ personal requests, the information provided by the customers will usually include personal data. It must be clear to you as the user how and why your data are being processed. You should also be aware that when the chatbot communicates with you, it may reproduce incorrect information. There is a right to information if the data are stored in a personalised form.

• A credit card provider blocks a transaction because it suspects that your card has been misused.

  Credit card providers use AI to analyse patterns in their transaction data. This may result in automated individual decisions. You will usually be informed of this in the terms of use.

• My smartwatch or fitness tracker monitors my sleep patterns and helps me to manage a chronic illness.

  AI can facilitate personalised health monitoring by evaluating the data collected on heart rate, exercise and sleep. Health data are regarded as sensitive personal data.

• Smart cameras are used at a shopping centre to alert a private security company as soon as certain activities (such as a physical altercation or a theft) are detected in the images.

  Any video surveillance must be justified and the general principles (proportionality, good faith, transparency, etc.) must be observed. Smart cameras are not per se more intrusive from a data protection perspective; under certain circumstances, they can even be more data protection-friendly (e.g. if they pixelate images and only allow identification or storage if predefined patterns are recognised on the image). The specific use is therefore the decisive factor. The proportionality of camera use depends on the purpose and on the algorithms used.
  It becomes particularly difficult to assess proportionality when automated individual decisions are made, when biometric features are used (e.g. facial recognition) or when the cameras are used to track people. A preliminary risk assessment must always be carried out to decide whether a data protection impact assessment must be prepared for this data processing.

Tips for people whose data are processed (data subjects)

• Find out about the product manufacturer and in particular the intended use and assess the risks.
• Who has access to your data? Are they passed on to third parties? Data security deserves special attention.
• When using data for training purposes, what data are being used? Are the data anonymised? Is it possible to object to this data processing?
• Do you consent to your data being processed under these conditions? Otherwise, avoid using the services or, if possible for the specific application, remove any personal data before entering data into an AI-supported program.

Communications related to AI: