Fostering AI Literacy Through Simple Prompting Exercises Using Dall-E

This innovative practice full paper describes an AI prompting exercise to enhance students' AI prompting skills by providing immediate visual feedback.The exercise was executed within an introductory systems development and organizational security course, consisting of approximately 54 students who were enrolled in a cybersecurity program at a large Midwestern university. The exercise was conducted in the classroom setting, followed by a discussion of observed patterns in successful and unsuccessful prompts. This paper delves into the exercise's methodology, offering an in-depth exploration of the prompts generated by the cohort of 50 students, supplemented with a content analysis. Majority of students initiated their prompts with elaborate descriptions of the intended subject, showcasing a clear inclination towards a subject-centric approach. There was also variance around explicitly addressing the task of image generation, with only a minority of prompts articulating the core objective. Spatial details of background features were prevalent, particularly in terms of their positioning relative to the subject or within the overall image composition. Moreover, the comprehensive nature of subject descriptions encompassing elements such as gender, age, physical features, ethnicity, and clothing, was a ubiquitous trend observed across nearly every prompt. We also consider future directions for enhancing and extending the exercise in courses with larger enrollments.

I. Introduction

With the proliferation of artificial intelligence (AI) tools into different facets of everyday life, there is a renewed emphasis placed on developing AI literacy [1] and incorporating it as a part of digital literacy skills [2]. AI literacy has been defined as “ … a set of competencies that enables individuals to critically evaluate AI technologies; communicate and col-laborate effectively with AI” [3]. AI literacy also emphasizes the ability of individuals to utilize AI tools both at home and the workplace [4]. Applications of AI, as well as ethics and safety elements associated with AI are considered to be subsets of AI literacy [5]. There have been efforts in recent years to promote AI literacy but most of them have taken a computer science-centered approach [6] and have been situated in a K-12 context. It is also worth noting the computer science centered approach towards AI literacy may not be suitable for most audiences since they will likely not need to know how to program AI to interact with it [4]. Efforts to improve AI literacy are needed given how organizations from around the world are recognizing the importance of AI literacy at an individual and organizational level [7]. Image generation systems utilizing generative AI have also risen to mainstream prominence in recent years [8]. The primary allure of this AI-enabled image generation is the ability for anyone, regardless of artistic ability, to create digital images using prompts in a natural language. It must be noted that the quality of the images generated depend greatly on the nature of the prompts provided with users often needing to interact with the AI tool in an iterative fashion to arrive at the desired output [9]. This approach for interacting with AI tools isn't limited to the realm of image generation and as such, there has been a renewed emphasis on the notion of prompt engineering across various disciplines [10] [11]. Prompt engineering encompasses not just the design and implementation of prompts but also the refining of prompts to shape the output of large language models (LLMs) or AI tools [11]. Prompt engineering plays an integral role in the effective utilization of LLMs or AI tools as it can be leveraged to control parameters such as length, complexity, or style of the output [12]. Some studies have also indicated that novices may not be able to construct effective prompts and often end up creating ambiguous prompts or those prompts lacking in context [10] [12]. Given the emerging nature of AI, there is a relative dearth of literature pertaining to AI literacy specifically in the topic of promoting AI literacy or the conjunction of higher education and generative AI have for the most part only provided frameworks, systematic reviews, or a better understanding of what other researchers perceive as “fostering AI literacy”. There is also a lack of public datasets detailing college student interaction with generative AI or LLMs. This is in turn reflected in the lack of research analyzing such data. Very few studies in the field of human-computer interaction, specifically with educational applications of AI tools, have sampled their own data for analysis of human behavior and as such leaves gaps for this study to have relevance. This study was guided by the research question: What are student approaches to prompt engineering when tasked with image creation using a generative AI tool? This paper presents the results of a qualitative data analysis on image generation prompts from students.