Artificial intelligence and society

Perplexity, can I trust you?

“It is right to question things critically. I am an artificial intelligence and work with information that I have learned. My goal is to provide you with helpful, factual and as accurate answers as possible. Here are a few things you should know about me: I can provide you with information, but you should always make your own decisions. I may be mistaken or using outdated/incomplete data. I do not guarantee accuracy or completeness. I try to respond neutrally and impartially. It is best to use me as a source of information and a benchmark, but always check important information yourself and, if in doubt, ask experts.”

[Answer from Perplexity.AI]

In its response to the question of whether it can be trusted, the AI system Perplexity already provides several examples of ethical challenges associated with the use of AI – such as data protection, the origin and quality of training data, non-objectivity, or the question of who is liable for possible consequences if the AI makes a wrong decision. These questions arise wherever AI is already being used in society.

The debate about what AI can and cannot do often fluctuates between utopian promises of salvation and dystopian horror scenarios. However, one thing is clear: AI is already impacting our everyday lives in many areas. It can help translate texts, save energy in the home, make it easier for us to navigate through traffic and monitor our fitness data. However, at the same time, the following applies: Caution is advised in sensitive areas of application.

Advancing digitalisation is leading to the reorganisation of service and production processes. For example, data on work processes is used to make them more efficient or even automate them. Automation means that certain tasks are performed either completely without human intervention or with the help of digital assistance systems. 

This development affects numerous industries and occupational groups – and poses major challenges for the labour market as a whole. The exact long-term consequences of this are still unclear. However, frequently expressed fears that rapid automation, particularly through the use of artificial intelligence, could lead to mass unemployment are considered exaggerated. It is more likely that the requirements for qualifications and skills will change in many occupational fields. However, human labour remains indispensable.

Digital, AI-supported assistance systems have enormous potential to improve working conditions in many areas and increase productivity. In industry, for example, they guide employees in assembly or logistics through their work steps and support them in maintenance and problem solving with respect to machines or systems. In administrative processes, the software used increasingly provides recommendations on how to perform certain tasks. Assistance systems are now also being used in agriculture – for example, in planning the use of resources or in maintenance tasks. Large language models such as GPT-5 (ChatGPT) can also be used effectively in complex work processes such as data analysis or programming.

At the same time, however, new challenges are also emerging: If employees rely too heavily on assistance systems, they may lack problem-solving skills – for example, when it comes to malfunctions that the system cannot fix itself. Therefore, the use of digital assistance systems should always be linked to appropriate training measures. 

In addition, it is important to prevent abuse through control and monitoring. Many systems collect data on work processes, time, location and sometimes even on individuals – this brings with it risks of surveillance and control that are difficult for employees to assess.

Algorithms are increasingly being used in situations where statistical methods can form the basis for a decision, e.g. in application procedures, when granting loans or in insurance decisions. It is often claimed that AI-based decisions are fairer because, unlike human decision-makers, AI does not harbour prejudices and does not discriminate. However that is a fallacy. Generative AI systems tend to reproduce established stereotypes and forms of discrimination. Since they are trained using existing data created by humans, they consequently also learn the prejudices, judgements and inequalities contained therein. Worse still: Discriminatory decisions are given a supposedly objective character because they were made on the basis of AI.

Studies show that AI-based decisions often exhibit a bias that disadvantages already disadvantaged groups. This is particularly worrying in the following areas, for example:

• Predictive policing: People from certain groups are more likely to come under suspicion in proactive policing than others, even though there is no reason for this.
• Finance: Despite having the same credit rating, women receive fewer loans from banks than men.
• Working environment: Applications with certain surnames are less likely to be considered by AI systems – despite equal qualifications.

Active countermeasures are needed to counteract systematic distortions. This is referred to as “debiasing”. However, both the decision to engage in debiasing and the choice of specific debiasing methods are of great ethical and political significance. Therefore, such decisions should not be made exclusively by developers – a broad, interdisciplinary discourse is needed, involving the groups of people potentially affected by discrimination.

Generative AI enables users to obtain answers to all kinds of questions or generate texts, images and videos within a very short time. This can be extremely useful – but it also carries risks. With the help of AI, deceptively real videos or voices can be generated (so-called “deepfakes”), and it is becoming increasingly difficult to distinguish between what is true and what is not.

AI-driven algorithms determine what information is displayed to us on social networks. Back in 2021, the Leopoldina, the German National Academy of Science and Engineering (acatech) and the Union of the German Academies of Sciences and Humanities issued a joint statement criticising the fact that the algorithms used by digital platforms prioritise commercial content over independent information. Although the selection mechanisms of the algorithms would have a major impact on the flow of public information, they would not be subject to any democratically legitimised control.

The existing legal regulations and their enforcement for the prevention and prosecution of undemocratic or derogatory expressions of opinion and targeted misinformation on the internet are inadequate. Particularly with regard to targeted misinformation, platform-independent tools are needed to support users in accessing and evaluating digital information

As already described in the chapter on technological developments in artificial intelligence, it is often impossible for users of large language models such as GPT-5 (ChatGPT) to determine the extent to which the generated texts and information are actually correct and reliable. To circumvent this problem, scientists working in the field of explainable AI are currently working on enabling large language models to explain their own procedures and results. However, a certain lack of transparency cannot be resolved at present – it is therefore necessary to question in which cases decisions can be left to AI systems and in which areas a human being should ultimately make the decision.

Although politicians and society are aware of many risks associated with the development and application of artificial intelligence, it is questionable how these risks can be managed responsibly without hindering innovation in the field of AI.

Many countries have already adopted ethical guidelines or policy strategies for AI. However, these are usually voluntary. The EU is the first regulatory authority worldwide to introduce an AI regulation that provides a legally binding framework for the responsible development and use of AI in the EU: the EU Artificial Intelligence Act.

In addition to the EU AI Act, the EU has also introduced other regulations that include legal rules for AI. The Digital Services Act (DSA) regulates online platforms when they use AI in the context of online services, for example in algorithms or personalised advertising. The Digital Market Act (DMA) regulates gatekeepers, i.e. large digital platforms that offer digital services such as search engines, app stores and messenger services. The aim is to create fair competitive conditions. The General Data Protection Regulation (GDPR) regulates the collection, processing and storage of personal data and also applies to AI systems as soon as they process personal data.

The EU regulations apply not only to companies within the EU, but to all providers worldwide as soon as they offer or use AI systems on the EU market. This gives them the potential to develop into a global standard. Nevertheless, critics fear that excessive bureaucracy will hinder innovation in the field of AI in the EU. It is also not yet clear how consistently and effectively the regulations can be implemented.

The science academies of the G7 countries have also dealt intensively with regulatory issues in preparation for the annual G7 summits. Among other things, they recommend measures for data protection, transparency and the prevention of deception, for example through the labelling of deepfakes. They advocate that governments promote legally enforceable standards for AI-generated content. They also recommend cooperation between the public and private sectors in order to further develop AI in a responsible manner and embed it in democratic processes.

Powerful AI systems that pose a risk to the public if misused must be adequately protected against cyber and physical attacks. For sensitive AI applications, data protection frameworks that are appropriate to the level of risk (e.g. the EU General Data Protection Regulation) and privacy-enhancing technologies are essential to protect personal data. When interacting with AI systems, users need clear privacy notices about how and for how long their data will be used, reused and stored.

The main areas in which generative AI has a major impact include the creative industries and research. Generative AI offers enormous potential for creative content creation and software development, but it also raises questions about intellectual property protection when it generates seemingly new content from existing works. A revision of the existing legal framework for copyright and intellectual property protection for the AI era is urgently needed. This requires a combination of clear legislation, self-regulation by the industry and independent supervision.

For a long time now, social media use has been a part of the everyday lives of children and adolescents in Germany. The impact this has on the psyche of children and adolescents is currently the subject of intense debate in science, politics and society. Ever more powerful algorithms and AI systems are being used to predict and steer user behaviour and create emotional attachment. Social media is also increasingly being used to influence users' political views. The psychological consequences of this digital influence on the socio-emotional development and mental health of children and adolescents are still largely unknown. However, there is compelling evidence that social media use can significantly impair the mental health, well-being and developmental opportunities of young people. Effective protective measures are therefore needed.

One measure could be, for example, the establishment of a “digital education canon” in nurseries and schools, which prepares young people for key issues in digital life. Children and adolescents should be taught to understand the business models of social media. In addition, power structures on the internet and social media should be addressed and media literacy skills taught, for example how to distinguish fake news from trustworthy sources. In addition, the widespread use of AI and its consequences should be discussed.

Such a digital education canon should also include teaching children and young people about the environmental impact of AI technologies (e.g. sustainability, consumption of rare earths, etc.). AI technologies are very resource-intensive and therefore also have an impact on our climate.

Published: September 2025