The E-Bezpečí team from the Faculty of Education at Palacký University in Olomouc, in cooperation with Charles University and the CEDMO hub, has prepared a new electronic publication entitled Risks associated with generative artificial intelligence, which addresses the very important and topical issue of potential and real risks associated with artificial intelligence (AI).
The introductory part is mainly devoted to the issue of hallucination and the so-called AI bias, including, for example, stereotyping and distortion of information, where artificial intelligence, in practice, may not tell users outright false information, but information that is distorted in a certain way. Prof. Kamil Kopecký from Palacký University in Olomouc, the author of the publication, explains, “Artificial intelligence makes factual errors, of course, but it can also commit distortions of reality and reinforce certain kinds of stereotypes associated with, for example, race, nationality or gender. For example, according to several studies, AI image generation tools (Dall-E, Midjourney) are guilty of stereotype distortion by showing people of higher social status with light skin while those of lower social status with dark skin (or Asian features). This is all due to the nature of the databases on which the models are trained, which are often not balanced. But it may not just be stereotypes associated with race, but also nationality, for instance – Indians are usually generated by AI wearing a turban, Mexicans are hairy, wear a sombrero, wear a poncho, and so on.”
Another key theme of the book is the misuse of generative AI to create disinformation and fake news. Despite the existence of ethical constraints, AI users can circumvent these rules and generate texts of a disinformative nature. The problem is not only hoaxes but also extortion letters or fraudulent spam.
A significant part of the book is also devoted to deepfake technologies that allow the creation of realistic videos of fictional events, which can lead to the erosion of trust in public institutions or influence political processes (including elections). Deepfake videos are also used for internet fraud, which is encountered on a daily basis in the environment of social media in the context of investment fraud – for example, in the context of cryptocurrencies. These scams very often exploit celebrities (their appearance, voice, gestures) and are generated by AI. “According to the findings of our CEDMO Trends research, three-fifths of the Czech population cannot correctly identify that deepfakes are AI-generated images, recordings or videos that look real but are not based on truth. Public awareness of the principles of generative AI language models is also low,” says Anja Grabovac, CEDMO project manager, illustrating the need to strengthen AI media literacy.
The publication appeals, in particular, to the need for regulation and a critical approach to the use of generative artificial intelligence and also highlights some of the legal and ethical issues associated with its use. These issues are constantly evolving and changing in the context of ongoing court cases. It also mentions the environmental impact of operating AI systems, which require significant amounts of energy and water, and addresses the risk of leakage of sensitive data, where users unknowingly provide personal information when using AI tools.
You can download the manual in Czech for free from the CEDMO website or from the E-Bezpečí website.
Are you interested in artificial intelligence in relation to the elderly? If so, we also recommend the publication Risks associated with artificial intelligence for the elderly.
