AI in Development Cooperation

Artificial intelligence (AI) is now widely and successfully used, including in emergency aid and development cooperation. But its hunger for data, electricity, water and labor has dark sides. A search for clues. 

You don't need a "smart refrigerator" that suggests recipes based on its contents and indicates expiration dates to come into contact with artificial intelligence (AI) in everyday life. AI has long been an integral part of our lives – be it in the form of voice assistants on smartphones, AI-assisted medicine, or intelligent systems that control traffic. It goes without saying that AI is also being used in efforts to combat hunger and promote greater education and equality, security and integration. More on this later. 

AI creates precarious work 

First, the downside: AI requires extensive training. This does create jobs, especially in the Global South. In Bangladesh, Uganda, Madagascar,and India, refugees, people living in poverty, and people with disabilities receive money by teaching AI how to tag data. At the same time, working conditions are often precarious , underpaid and lack opportunities for advancement.  

Training AI is repetitive and, depending on the task, even traumatizing. For an AI to learn to recognize videos of (suicide) murders, sexual assaults, or child abuse, for example, a human must train the machine. This means that a human must view disturbing content, assess it, and flag it accordingly. Employees who perform these tasks rarely receive psychological support . The consequences are anxiety, depression, and trauma. The "clicks" performed are not monitored to protect employees, but to control speed and efficiency . Anyone who is too slow loses their job.  

Hungry and thirsty AI 

AI training is an ecological problem: Applications like ChatGTP consume ten to thirty times more energy per query than traditional internet searches. By 2026, we are estimated to need twice as much electricity for data centers worldwide as in 2024. Operators are increasingly turning to nuclear power for this purpose . To prevent overheating, data centers are cooled with vast quantities of water – drinking water to prevent bacteria and corrosion. Microsoft alone used over 6.4 billion liters of water for this purpose in 2021, equivalent to 2,500 Olympic-sized swimming pools.  
With AI, the immense hunger for raw materials for batteries and microprocessors has also reached new dimensions. AI competes directly with the expansion of green technologies such as photovoltaics, heat pumps, and electromobility. 

A more considered approach is offered by open source initiatives that are accessible to developers worldwide. With open source, once trained, AI can be reused multiple times and modified to meet individual needs. This massively reduces resource consumption. A cooler location for data centers also requires less water for cooling. 

Concentration of power promotes Western worldviews 

What matters is the data an AI accesses. Language models learn through training, website content they access, and user search queries. Globally, white men in the US perform the most searches . Answers that match their needs are therefore rated as more relevant and are more likely to be displayed. Thus, generative AI primarily reproduces Western, male perspectives and sexism and stereotypes from other cultures. 

In addition to this bias, AI can also encourage disinformation. Language models often produce false or misleading political results : Microsoft's Bing Chat invents scandals, fakes poll numbers, and provides false election dates. China's DeepSeek remains silent about the 1989 Tiananmen Square massacre or human rights violations against Uyghurs. Language models are improving, but it can be difficult to obtain sound political information and form a sound opinion. 

In addition, many languages ​​are not (sufficiently) available digitally , for example because they have an oral tradition. Because the machine is usually trained in English and Western European languages , and there are comparatively more data sets for these, queries in these languages ​​​​provide more detailed answers than those in indigenous African, Asian or South American languages. However, there are local initiatives (e.g. African ones such as Lelapa , Masakane or EqualyzAI) that address this underrepresentation. Only when these languages ​​and knowledge bases are incorporated can AI also serve people in the Global South. However, for smaller initiatives to develop an impact, investment and strong regulation are needed to break the power of the tech giants and promote AI that is oriented towards the common good. 

At the moment, tech giants have the most resources to further develop AI, and they can incorporate these innovations into countless applications. In doing so, they expand their power and reinforce the worldviews and values ​​that go with it. AI critics speak of digital or AI colonialism here – and due to the aforementioned exploitative working conditions . And the concentration of power continues to grow when tech giants like Musk, Altman of OpenAI , and Zuckerberg in the US – as we are currently observing – have such close ties to the Trump administration, which has no regard for social and ecological sustainability. 

Humanitarian aid: Gaining time and saving lives 

Despite all the problems and risks, there are two sides to the coin. AI's potential is also enormous. It is already being used in a variety of humanitarian aid applications because of its ability to quickly recognize patterns, predict scenarios, and evaluate satellite images. This makes it possible, for example, to predict crop failures and thus food shortages and famines, and to plan appropriate measures in a timely manner.  

An AI developed by Google was able to calculate when and where a river in Nigeria was likely to overflow its banks – based on a combined analysis of temperature fluctuations, changes in air pressure, topographical maps, historical data, previous floods, rainfall, and river levels. In a country where floods occur repeatedly and more than 600 people died in floods in 2022, AI bought valuable time when another flood threatened: Many people were able to bring relatives, livestock, or important documents to safety in time. 

Thanks to the combination of satellite images with AI, armed conflicts can also be better analyzed and identified, revealing how they shift geographically , where the largest populations live and where help is most urgently needed. Self-piloting drones can explore areas autonomously without people having to put themselves in danger. This is an advantage that warring parties also take advantage of, but also a feature that relieves the burden on rescue services. Rega is also testing self-flying drones in mountainous regions. These drones use an algorithm to learn how to recognize and locate people in the terrain. This is particularly useful in poor visibility, when a helicopter flight would be too dangerous. 

Voice bots are increasingly being used in emergency aid: They can answer simple inquiries from refugees or triage requests as needed – a significant relief when time is of the essence. People who are illiterate can express their concerns verbally and receive a spoken response. This makes advice easier to understand and reaches more people, around the clock.  

Development cooperation: Optimizing irrigation and preventing illegal deforestation 

Long-term development cooperation also benefits: An AI system from ETH Zurich categorizes 3.2 million development projects by thematic groupings and identifies global trends. It shows how funding is distributed across topics, countries, and years – and where funding gaps exist. This allows for more effective global project coordination.  

AI also has a local impact: In agriculture, AI optimizes irrigation . AI-supported image analysis allows for better control of plant pests or – for example, in Colombia – faster intervention in cases of illegal deforestation that threaten ecosystems and the livelihoods of indigenous communities.

AI improves processes and identifies savings potential – in building technology, transport and logistics, supply chains, and power supply. In solar and wind energy , for example, AI helps analyze wind speed and solar radiation, optimally adjust systems, and generate more electricity. 

After weighing all these pros and cons, the question is not whether we should use AI, but how to use it responsibly. Conscious consumption remains crucial: in moderation. As little as possible, as much as necessary.