New expert interviewsWhat impact does AI have on schools, our environment and data protection?

1. will we still have to memorize things in the future if the AI knows everything anyway?

Falck: I would answer this question with "yes" and "no": No, because we can access the world's knowledge in a personalized and omnipresent way. In many places, this opens up opportunities to outsource knowledge and not always have to have it available. And yes, because knowledge is the basis for skills development and we need a whole bundle of skills to be able to work effectively with AI-generated output. Of course, generative AI tools such as ChatGPT or Gemini are powerful and helpful tools: they answer us in natural language and have a high level of expertise in many areas. You can ask them almost anything and, in many cases, get helpful answers. At the same time, these "answers" are based on probabilities. This means that chatbots don't "know" anything at first. They are good at combining data plausibly. This means that every AI output must ultimately be assessed by us humans. And that, in turn, is not possible without knowledge.

Flick: I would even go one step further: I believe that the more of a role AI plays in everyday (school) life, the more important it becomes to have a basic knowledge of how to use these tools responsibly, critically and reflexively. This is because the effective use of AI tools such as chatbots requires solid basic and specialist knowledge as well as a stable foundation of fundamental skills. This knowledge also includes practical knowledge in order to understand contexts and to be able to critically examine and classify AI content.

2 Why should students still learn if AI can do everything?

Flick: Even if AI systems are already very powerful in many areas, there are profoundly human skills that cannot be replaced by AI in the future, or at least only to a limited extent. I'm thinking of authentic communication, empathy, genuine collaboration and creativity. These skills need to be emphasized more than ever in the context of learning. In addition, I experience in practice that students want to understand and master things above all when they realize that the content taught has a real connection to the world of work or life. Then the motivation is there. Pupils do not want to become completely dependent on technology - they want to perceive themselves as competent! That's why the question "Why are we actually learning (this)?" should always accompany lessons.

Falck: Learning and, above all, education are more than just the accumulation of knowledge. The fact that we learn has something to do with curiosity and with how we understand the world, how our personality develops and how we expand our options for action. We also learn because of certain personal goals or to avert something. Educational institutions should therefore focus more on conveying meaning and creating activating learning environments. We need more components of self-determination, individual goals, experience and personal meaningfulness. This can be achieved, for example, through greater participation and project-based forms of work that are closer to reality than textbooks, but also through an increasing focus on future skills such as communication, collaboration and creativity.

1 Why have AI systems experienced such a "boom" in recent years?

Lorig: The fact that AI systems have become better and better in recent years is largely due to the huge amounts of data they have been trained with. For an AI to be able to create human-like texts, the system must first understand how people express themselves, how they communicate with each other and what their views are on certain topics. To this end, the systems have been and are being trained with vast amounts of data that we as users leave behind on social media, for example. This applies not only to texts, but also to images that we upload there - they too have been and are used to train the image and video generators.

2. what particular risks arise with regard to data protection when AI systems are used in schools?

Lorig: Data from children and young people is particularly interesting for training AI systems, as their way of expressing themselves is rarely found in publicly accessible texts. If public bodies, such as schools, use AI systems, it must therefore be ensured that the pupils' input is not used for commercial training and further development of the provider's product. Some systems make it possible to deactivate the processing of input for training purposes in the settings. Some tools can also be used without logging in or in "incognito mode", which means that less information is saved from session to session. One problem with AI systems, however, is the lack of transparency. We do not know which algorithm is used to process the data. This not only makes it almost impossible to remove data that has been fed into an AI system for training purposes. We can see

1. what does artificial intelligence have to do with sustainability?

Doria: Much more than it seems at first glance. But first we need to clarify what the words "sustainability" and "artificial intelligence" actually mean. The term "sustainability" was coined in the 18th century by Hans Carl von Carlowitz, who supplied the mining industry in the Ore Mountains with wood. His basic idea was that if you take more wood from the forest than grows back in the same period, you cannot operate a "sustainable" timber industry. Based on this, the aim should therefore always be to use today's resources in such a way that they are also preserved in the future. The term "artificial intelligence", on the other hand, is misleading - because if you take a closer look at these technologies, you realize that they are neither "intelligent" nor "artificial". AI systems imitate intelligent behavior and cannot function without human work. Talk of alleged "artificial intelligence" should therefore not lead to the people who develop and operate them no longer wanting to take responsibility for them. Because behind the colorful apps on our cell phones, there is unfortunately a very problematic side that is often ignored.

2. what is the problem with "AI and sustainability"?

Doria: When it comes to "AI and sustainability", Prof. Aimee van Wynsberghe distinguishes between two areas: On the one hand, the use of AI to achieve sustainability goals (AI for sustainability) and on the other hand, the sustainability of AI itself (Sustainability of AI). Two problems are particularly relevant here: The main problem lies in the high energy and resource consumption of AI.

To address this, we must first understand how AI works. When we talk about "AI" today, we usually mean "machine learning". This means that an AI system is taught to recognize patterns in large amounts of data. The "bigger-is-better" approach usually applies here: the more data an AI system receives for training, the more effective it becomes. A trained model can later apply the patterns and correlations recognized there to new, unknown data in order to arrive at a result.

Let me illustrate this with an example: if an AI is to "learn" to recognize lung cancer on CT images, I first have to show the system many lung images that vary in terms of health status, age and gender. If cancer is visible, I always give the AI the hint "This is lung cancer". Over time, the model becomes more and more accurate at differentiating between diseased and healthy lungs. After the training phase, it should be able to recognize lung cancer on new CT images.

And now back to energy consumption: in order to be able to use an AI model for a specific purpose, it must first be "trained" with very powerful computers. How much energy is consumed for this depends on various factors - for example, how complex the model is, how powerful the hardware is and how sustainable the energy mix at the location is. But to get a feel for the order of magnitude, I'll give you two examples: Experts estimate that training a predecessor model of ChatGPT caused as many CO2 emissions as around 300 return flights between New York and San Francisco. What's more, the AI data centers that chip manufacturer Nvidia equips with hardware consume about as much electricity per year as the whole of Sweden.

In addition to training, the "inference phase" - the point at which a trained AI system is actually used - also requires a lot of computing power. This is because every time I interact with an AI, for example via a prompt, it costs more energy. Especially if the model is supposed to "learn" from my interactions (user data) and improve itself. Again, here are some examples to put this into context: Generating an AI image requires about as much power as charging a smartphone. And the total annual energy consumption of ChatGPT is roughly equivalent to the power required to charge over three million electric cars.

The word "cloud" completely ignores the fact that we are talking about power-intensive data centers that require valuable raw materials as well as energy. For example, silicon for computer chips, copper for cable connections and enormous quantities of fresh water to permanently cool the AI chips installed. In the 2015 Paris climate protection agreement, we actually agreed globally to limit global warming to 1.5 degrees. However, this seems increasingly questionable to me in view of these figures.

Another problem is the outsourcing of certain work processes to so-called click workers in low-wage countries. This is because before data can actually be used for AI training, it must first be viewed and "labeled" by real people (e.g. labeling images with "this is cancer" / "no cancer"). This is a very time-consuming process that is usually done manually. Large AI companies such as OpenAI (ChatGPT) like to outsource this to clickworkers in the Global South, as wages are significantly lower there. We know, for example, that OpenAI paid its clickworkers in Kenya only 2 US dollars per hour via subcontractors. From an ethical point of view, however, it is not only the low hourly wage that is problematic, but also the psychological strain of this work. This is because these people also view content that would fall under the protection of minors in Germany for good reason.