Romanian students use artificial intelligence at a higher rate than their foreign counterparts, according to a study published in the Economic Amphitheater, but “too many respondents (43%) say that the results obtained by AI are ‘rarely’ or ‘never’ is incorrect, which is an alarm signal,” the quoted document says.

About 10% of them use AI content generation models to search for information (for example, for “statistics”, “restaurants in Bucharest” or “historical and geographical facts about different countries”) and, in addition, consider the generated content to be reliable.

​The majority of research participants have used artificial intelligence models for content generation at least a few times (52.5%), a significant number have heard about these models and expressed interest in trying them (22.8%).

Only 5.2% have never heard of AI models for content creation

However, there is also a category of 14% of participants who have heard of but do not want to use AI models. A relatively small percentage regularly use AI-generated content (5.5%), while at the opposite pole, 5.2% have never heard of AI models for content creation.

By comparison, in a May 2023 study by Byles, Lea, and Howe of the University of Northampton, 62% of respondents said they did not use AI models, while in a July 2023 study by Chan and Hu, 33% of respondents said they did not. state that they have not used AI models such as ChatGPT.

Reasons for using or not using IAG models

When analyzing the reasons for using or not using the IAG models, they were found to range from “interesting” to “misinformed”. To better understand what the key elements are, we first look at the 90% of students who are interested in the IAG models used:

• For 33% of them, the interest is “curiosity”, which indicates that they are intrigued by the technology and that they have explored or will soon explore its possibilities.

• A similar percentage considers the use of AI content generation models to achieve greater efficiency (32%). They point out that IAG tools are ways to save time and streamline efforts in content creation and learning.

• About 14% of respondents interested in these technologies look to AI-generated content as a source of inspiration, highlighting the potential of these tools to stimulate creative ideas and innovative potential in an academic context. It also means that students are not intended to use the content as is, but will add their own ideas.

• At the same time, an alarming percentage of 10% use AI content generation models to search for information (for example, for “statistics”, “restaurants in Bucharest” or “historical and geographical facts about different countries”) and, in addition, look at the generated content how reliable Students report that they used AI when accurate information was not available on Google or when they wanted accurate information. This indicates a high risk associated with the possibility of disinformation.

In addition, we may see an increase in students’ trust in AI-generated content, which can have both positive and negative aspects, the study says.

On the positive side, AI can fill knowledge gaps where traditional sources are insufficient, and on the negative side, there is an increased risk of misinformation and over-reliance on created content, which can lead to a lack of actions such as critical thinking and information verification.

In contrast to those who place too much trust in IAG-generated content, we also record 10% of participants who state that they are not interested in using IAG for reasons such as: incorrect information generated, predicted negative effects on cognitive abilities , a preference for using one’s own skills, considering the use of AI in the “plagiarism” category, or a perception of a lack of added value.

As far as IAG models available to students go, ChatGPT stands out. Yes, it is mentioned by 76% of students who have used IAG and 53% of those who have no experience but are interested or have limited experience using AI tools. This finding complements the results of a study conducted in Germany in 2023 involving 6,311 students from 395 colleges (Garrel, Mayer, & Mühlfeld, 2023). According to this study, when asked to describe how they use AI tools in their learning and for what purpose, 63.2% of students said they have used or are currently using AI-based tools for learning, and 48.9 % – already used ChatGPT.

Many students with no or limited AI experience expect lower quality than their own texts

Most students say they use them for academic purposes “rarely” (42%) or “sometimes” (37%), or expect to use them “rarely” (37%) or “sometimes” (46%). This fact suggests that students believe they will continue to rely on traditional methods of creating academic content.

Analyzing students’ perceptions of the quality of AI-generated content, it is interesting to note that more of those with no or limited experience expect lower quality than their own texts – 30% – compared to only 15% among those who used AI simulations and perceives quality as lower than its own.

The majority of surveyed students (43%) are relatively positive about AI-generated content – ​​they believe that AI-generated results are “rarely” or “never” wrong or irrelevant, which is a red flag. Text-generating AI models are also known for their “hallucinations,” or “confabulations” (if we don’t attribute human characteristics to the AI). When used as a brainstorming tool, IAG’s leaps in logic and confabulation can lead to creative breakthroughs, but when used as a factual reference, IAG can do real damage.

On that note, we used a question at the end of the questionnaire to gauge students’ interest in formal training in the use of AI. To our surprise, a relatively high percentage of respondents, namely 21%, said they were not interested in learning or using artificial intelligence models, the cited study also shows.

Source: țălă, ML, Müller, CN, Albăstroiu Năstase, I., State, O. and Gheorghe, G., 2024. Examining the perception of generative artificial intelligence in education by university students. Economic Amphitheater, 26(65), pp. 71-88,

DOI: https://doi.org/10.24818/EA/2024/65/71