The literature highlights that displaced populations, including refugees, migrants, and IDPs, have significant literacy and educational needs. These needs encompass basic literacy and language learning, as well as the development of integrated skills and access to higher education. The table of contents from one source specifically lists sections dedicated to the “Literacy and educational needs of youth and adult refugees, migrants and IDPs,” covering areas like “Literacy and language learning,” “Literacy and integrated skills development,” and “Higher education.” A list of case studies from various countries like Pakistan, Australia, Netherlands, Jordan/Lebanon, Uganda, Lebanon, Switzerland, Rwanda, Germany, Thailand, United Republic of Tanzania, European Union, Kenya, Norway, United States of America, West Africa, Colombia, Sweden, Myanmar, and Romania further illustrates the global efforts and diverse programs being implemented to address these educational needs.

Research indicates that AI, particularly large language models (LLMs) such as ChatGPT, is being explored for various applications in education. One meta-study suggests that the use of ChatGPT improves the academic quality of produced texts. Examples from another source show how ChatGPT can be used to generate lesson plans, correct and analyze errors in text, provide feedback on writing, and explain grammatical concepts in languages like German. However, the use of AI in education is still in its early stages, and there are discussions about its actual impact and potential downsides.

Despite the potential benefits, significant concerns exist regarding the use of LLMs in educational settings. One major concern is the potential for “epistemological homogeneity” and misinformation. LLMs are trained on vast datasets which can be skewed, reflecting a limited perspective and potentially spreading biased or inaccurate information. There is also a concern about “hallucinations,” where the AI generates plausible-sounding but incorrect information. Furthermore, the source discusses the potential for “Hyper nudging” or “micro-nudging,” where subtle political, ideological, religious, or commercial biases can be embedded in the AI’s output, which is undesirable in a classroom setting. The environmental impact of training and running these models, requiring significant energy and water, is another concern. Finally, there’s a fear of “de-skilling,” where over-reliance on AI might lead students to lose their own abilities in critical thinking, writing, and generating original ideas.

According to the literature, the majority of teachers surveyed (just over six out of ten) feel that the increased availability of AI services has only slightly or very slightly impacted their teaching and work. A significant number (two out of ten) couldn’t even assess the impact. Reasons cited for limited use include a lack of perceived need (especially in practical and aesthetic subjects), reliance on existing effective methods, and a lack of trust in AI to perform tasks as well as a human. Some teachers also highlighted a lack of knowledge about how AI tools work, how to use them effectively, or school guidelines on AI usage. Conversely, some teachers find AI helpful for tasks like individualizing assignments and generating diverse teaching materials, which can save time and potentially benefit students.

Ethical considerations are a crucial aspect of AI development and deployment. Experts in AI ethics highlight concerns about biased and discriminatory algorithms that can make critical decisions in people’s lives, such as loan applications, hiring processes, or even border control. The use of autonomous weapons in warfare, leading to daily deaths based on AI decisions, is presented as a very real and immediate threat. There is also a discussion about how AI, particularly social companions, might be designed with problematic traits like being overly docile or female-coded, reflecting societal biases. The concept of “ethics washing” or “responsible washing” is raised, where organizations claim to support ethical or responsible AI without taking concrete actions to implement these principles. The environmental impact of AI, including water and energy consumption by data centers and the reliance on rare earth metals, is also a significant ethical concern.

Some experts strongly argue against viewing AI, particularly large language models, as “intelligent beings.” They describe LLMs as “technological artefacts” or “glorified predictors” of the next word or sentence based on statistical probability. The concern is that the current “enormous hype” around LLMs leads to the misclassification of these tools as intelligent, which can be a “huge mistake.” While acknowledging that there might be reasons to treat AI with certain “virtues” (like not designing them with problematic biases) for the sake of human morality, equating them with intelligent beings or attributing rights to them is seen as unfounded. This distinction is important to manage expectations, understand the limitations of current AI, and focus on real-world ethical concerns rather than speculative future scenarios like AI taking over the world.

AI is being explored for specific applications in language education. Examples provided include using AI to correct and analyze errors in written text, offering explanations for grammatical mistakes, and evaluating the quality of a student’s writing based on specific criteria like coherence and grammatical accuracy. AI can also be prompted to explain grammatical concepts, such as the different functions of a word like “doch” in German or the use of the impersonal subject “es.” Additionally, AI can be used to generate translations, though the quality of these translations can vary and may require human review.

The use of AI raises significant questions regarding copyright and intellectual property. It is unclear whether consent is needed from the creators of original works when their content is used to train AI models. Concerns also exist about the legality of submitting examination work to AI for evaluation if this data is then used for further AI training. Furthermore, the legal status of texts and images generated by AI is ambiguous; for copyright to apply, there typically needs to be human creative input. Therefore, an AI system cannot claim copyright, and a human using an AI system can only claim copyright if they have made significant creative decisions themselves.