This article is contributed. See the original author and article here.
Introduction
Welcome to Part 2 of our exploration into generative AI, where we delve deeper into the practical applications and creative potential of this innovative technology.
This article highlights concrete examples from students projects of the course ‘Prompt Engineering’ at Fondazione Bruno Kessler (FBK) in Trento (Italy). The aim is to showcase how students leveraged generative AI in unique ways. In particular, we’ll focus on two fascinating projects: “Generative Music” and “Personal Chef,” which exemplify the versatility and impact of generative AI in diverse fields.
Core element of these projects is the use of a structured framework known as the Card Model to define and organize generative AI tasks. In the context of this course, a card refers to a structured format or template used to define a specific task or objective for generating content or output using generative AI techniques. The Flow of these cards, meaning the logical sequence and interaction between them, is crucial for the coherent generation of complex outputs. For a detailed explanation of Card and Flow concepts read the 1st part of this blog series.
Our students have been actively experimenting with generative AI, producing remarkable results in their projects. Here, we present detailed insights and experiences from their hands-on work, demonstrating the practical applications of prompt engineering with non-tech students.
Generative Music
The “Generative Music” project leverages generative AI technology to innovate the music creation process. Central to this project is the use of Generative AI Cards that define various musical parameters and guide the AI in generating unique compositions. Generative AI Cards specify key musical elements such as genre, number of chords, melody length, key, and instrumentation, including bass and guitar (Fig. 1). Each card represents a distinct aspect of the music, allowing for precise control over the generated content. By configuring these cards, the team can tailor the AI’s output to meet specific creative goals.
Card Configuration
The process begins with the selection and configuration of these cards. Initial configurations often require multiple iterations to achieve satisfactory results. Each card’s parameters are adjusted to optimize the music generation, focusing on refining the elements to create a harmonious and appealing output.
Fig 1: Example of Cards from the Music Project.
Flow Generation
Flow generation involves the structured combination of these AI Cards to produce a coherent piece of music. This stage is crucial as it dictates the sequence and interaction of different musical components defined by the cards. The project utilizes tools like Canva to aid in visualizing and organizing the flow of these components, ensuring a smooth and logical progression in the music. During the flow generation process, the order of the AI Cards is experimented to explore different musical outcomes. However, the team found that altering the sequence did not significantly affect the final output, indicating that the cards’ individual configurations are more critical than their order.
Iterative Refinement and Human Interaction
A significant aspect of the project is the iterative refinement process, where generated music undergoes multiple evaluations and adjustments. Human intervention is essential at this stage to validate the quality of the output. Listening to the music is the primary method for assessing its adequacy, as human judgment is necessary to determine whether the AI’s creation meets the desired standards. The team continuously modifies the prompts and configurations of the AI Cards based on feedback, refining the generative process to improve the music quality. This iterative cycle of generation, evaluation, and adjustment ensures that the final product aligns with the creative vision (Fig. 2).
Fig 2: genAI music showcase (youtube.com)
Lessons Learnt
The “Generative Music” project demonstrates the potential of generative AI in the field of music creation. By using Generative AI Cards and structured flow generation, the project showcases a methodical approach to producing unique musical compositions. Despite the need for substantial human involvement in the refinement process, this innovative use of AI represents a significant step forward in integrating technology with artistic creativity.
Personal Chef
The “Personal Chef” project utilizes generative AI to assist individuals in planning balanced meals efficiently. The primary goal of this project is to save time and resources, enhance creativity, and provide valuable insights for meal planning. Generative AI Cards are central to this project, serving as modular components that define specific meal planning parameters. Each card encapsulates different aspects of meal creation, such as the type of dish (e.g., balanced dish, vegetarian alternative), the ingredients required, and the nutritional composition (Fig. 3). These cards help in structuring the meal planning process by providing detailed instructions and alternatives based on user preferences and dietary needs.
Card Configuration
For instance, one AI Card might focus on generating a list of high-protein foods, while another might ensure the meal components are seasonal. These cards are iteratively refined based on user feedback to ensure they deliver precise and relevant outputs. The language used in these cards is carefully chosen, as even small changes can significantly impact the results. The feedback loop is crucial here, as it allows continuous improvement and ensures that the AI provides more accurate and context-specific suggestions over time.
Fig 3: Example of Cards from the Personal Chef Project.
Flow Generation
Flow generation in this project involves the logical sequencing and combination of Generative AI Cards to create coherent and balanced meal plans. This process ensures that the output not only meets nutritional guidelines but also aligns with the user’s preferences and constraints. The flow of these cards is designed to cover various stages of meal planning, from selecting ingredients to proposing complete dishes (Fig. 4). For example, a flow might start with an AI Card that provides a balanced dish recipe, followed by another card that suggests a vegetarian alternative, and then a card that customizes the dish based on seasonal ingredients. This structured approach ensures a logical progression and maintains the relevance and coherence of the meal plans.
Fig 4: Example of Flows from the Personal Chef Project.
Iterative Refinement and Human Interaction
The project emphasizes the importance of human feedback in refining the AI-generated outputs. Users can interact with the system to customize the generated meal plans, adding or removing ingredients as needed. This iterative process ensures that the AI’s suggestions remain practical and tailored to individual preferences and dietary requirements. By continuously incorporating user feedback, the project aims to enhance the precision and utility of the Generative AI Cards, ultimately making the meal planning process more efficient and enjoyable.
Lessons Learnt
The “Personal Chef” project showcases how generative AI can be leveraged to support everyday tasks like meal planning. The use of Generative AI Cards allows for a modular and flexible approach, enabling users to create personalized and balanced meal plans. While the AI can provide valuable insights and save time, human interaction remains essential to validate and refine the outputs, ensuring they meet the users’ specific needs and preferences. This integration of AI and human expertise represents a significant advancement in making daily routines more manageable and creative.
Students Survey Results
The class consisted of 11 students (average age 23, 5 female) from various university faculties, i.e. Psychology, Cognitive Science, Human-Computer Interaction.
As said, it was a class of non-tech students. Indeed, most of them (7 out of 11) stated that they rarely (1-4 times in the last month) used tools such as ChatGPT. Only one student stated that he/she regularly (every day or almost every day) used these tools, either for study-related and unrelated purposes. One student admitted that he/she was not familiar with ChaptGPT, had only heard about it but had never used it.
In order to investigate the students’ knowledge of GenAI and its potential and to assess the effectiveness or otherwise of the course in increasing their knowledge and ability to use GenAI, we administered a questionnaire at the beginning and end of the course and then made comparisons.
The questionnaire consists of 50 items taken from existing surveys [1,2] investigating various dimensions concerning AI in general. These included: (a) AI Literacy, based on the level of knowledge, understanding and ability to use AI, (b) Anxiety, related to the fear of not being able to learn how to use AI correctly, as well as of losing one’s reasoning and control abilities, and (c) Self-Efficacy, related to confidence both in one’s technical capabilities and in AI as good aid to learning.
As evident from the graph below (Fig. 5), by comparing the answers given by the students before and after the course, it is evident that on average the students increased their literacy and self-efficacy, and decreased their anxiety.
Fig. 5:. Average scores of students’ literacy, anxiety and self-efficacy gathered at the beginning and the end of the Prompt Engineering course for non-tech students.
Furthermore, at the end of the course we asked the students to answer 10 additional questions aimed at gathering their feedback specifically about Generative AI. In particular, we asked them to score the following statements using a 7-point Likert scale, where 1 means “strongly disagree” and 7 means “strongly agree”:
I increased my knowledge and understanding of GenAI |
I can effectively use GenAI |
When interacting with technology, I am now aware of the possible use of GenAI |
I am aware of the ethical implications when using AI-based applications |
Taking a class on Prompt Engineering for Generative AI made me anxious |
I am afraid that by using AI systems I will become lazy and lose some of my reasoning skills |
AI malfunctioning can cause many problems |
If used appropriately, GenAI is a valuable learning support |
When using GenAI, I feel comfortable |
I significantly increased my technological skills |
Fig. 6 presents the average score and standard deviation of each item. As evident from the graph, after the course the students recognised the value of GenAI as a valuable tool to support learning (item 8). They also showed to be more aware of the possible uses of GenAI (item 3) and ethical implications of such uses (item 4), whereas they showed a low level of anxiety in attending the course (item 5) and a low level of fear of losing reasoning skills (item 6).
Fig. 6: Average scores on a 7-point Likert scale given by non-tech students at the end of the Prompt Engineering course.
Conclusions
Generative AI is widely used in higher education and skills training. Articles like [3] demonstrate that Generative AI is widely used in higher education and skills training, highlighting its benefits for productivity and efficiency, alongside concerns about overdependence and superficial learning. In Part 2 of our blog series, we delved into the practical applications and creative potential of this innovative technology. Through projects like “Generative Music” and “Personal Chef,” our students demonstrated the versatility and impact of generative AI across diverse fields. Central to these projects was the structured framework known as the Card Model and a Flow of the identified cards, which helped define and organize generative AI tasks.
The course significantly enhanced students’ understanding of prompt engineering, reducing their anxiety and increasing their self-efficacy. Survey results indicated improved AI literacy and decreased anxiety, with students feeling more confident in their technical abilities and recognizing the value of generative AI as a learning tool. Utilizing atomic cards to define and organize generative AI tasks facilitated the learning process. This structured approach allowed students to better grasp and control various aspects of content generation. In the “Generative Music” and “Personal Chef” projects, the cards provided a flexible and modular framework, enabling iterative refinement and improved output quality.
Looking ahead, future developments could further enhance the effectiveness of teaching generative AI. Developing specific tools and editors for configuring prompts could simplify the process, making it more intuitive for students. Establishing standard guidelines and metrics for evaluating generative outputs could provide more structured feedback, improving the learning process. Additionally, expanding the course content to include a broader range of diverse and complex case studies could help students explore more generative AI applications, deepening their understanding and innovative capabilities.
These advancements would not only improve the teaching of generative AI but also promote greater integration of technology and creativity, better preparing students for their future professional career.
Antonio Bucchiarone
Motivational Digital System (MoDiS)
Fondazione Bruno Kessler (FBK), Trento – Italy
Nadia Mana
Intelligent Interfaces and Interaction (i3)
Fondazione Bruno Kessler (FBK), Trento – Italy
References
[1] Schiavo, Gianluca and Businaro, Stefano and Zancanaro, Massimo. Comprehension, Apprehension, and Acceptance: Understanding the Influence of Literacy and Anxiety on Acceptance of Artificial Intelligence. Available at SSRN: https://ssrn.com/abstract=4668256.
[2] Wang, Y. M., Wei, C. L., Lin, H. H., Wang, S. C., & Wang, Y. S. (2022). What drives students’ AI learning behavior: a perspective of AI anxiety. Interactive Learning Environments, 1–17. https://doi.org/10.1080/10494820.2022.2153147
[3] Hadi Mogavi, Reza and Deng, Chao and Juho Kim, Justin and Zhou, Pengyuan and D. Kwon, Young and Hosny Saleh Metwally, Ahmed and Tlili, Ahmed and Bassanelli, Simone and Bucchiarone, Antonio and Gujar, Sujit and Nacke, Lennart E. and Hui, Pan. ChatGPT in education: A blessing or a curse? A qualitative study exploring early adopters’ utilization and perceptions. Computers in Human Behavior: Artificial Humans, Vol. 2, N. 1, 2024. https://doi.org/10.1016/j.chbah.2023.100027
Brought to you by Dr. Ware, Microsoft Office 365 Silver Partner, Charleston SC.
Recent Comments