Aiming to support creative work of students: The ChatGPT application as part of a Master’s level online course

Computer on the table with beautiful scenery seen in the window

This blog post is written by professor Kalle Juuti and university lecturer Vilhelmiina Harju from the Faculty of Educational Sciences.

Currently, in the field of education, the hot topic is the applications of generative AI and how they impact learning and teaching. During the rapid technological changes, it needs to be discussed, how we understand new generative AI applications, their possibilities and potential drawbacks in education. Further, it is important to consider learning and studying with these tools. For example, do we understand new applications as tools for producing essays and other learning assignments, or do we see them as an opportunity to ideate and develop our own thinking? Do we use these tools in a way that actually excludes ourselves from a learning process or can we use them in a way that in which we actively seek to develop our understanding and self-regulate our learning process? In higher education, where students have traditionally written a lot of texts to prove what they’ve learned, the development of generative AI tools means we need to rethink what and how we teach, as well as how we evaluate students’ learning. In particular, it is important to develop pedagogical approaches that exploit new tools in a way that support students’ new creative work and development of understanding.

In this blog post, we describe, how ChatGPT3.5 application “CurreChat”, operated by the University of Helsinki, was used in a Master’s level education online course in spring 2023. One aim was to integrate the use of a generative AI application into course assignments in a pedagogically meaningful way. ChatGPT was considered as a tool to support students’ creative work. Another aim was to practice reporting on the use of the AI application according to university guidelines.

In the course, students were asked to construct a solution concept to a problem they had identified in the field of education. Weekly course assignments were linked to different phases of concept construction process, and finally the whole working was documented and reflected in a portfolio. Students were given an opportunity to use the AI application in doing the assignments (e.g., identifying problems, ideating solutions, getting feedback on ideas, and reflecting on impact). For each assignment, students were given tips on how to use the tool in a way that would support their work. Students were also asked to write a short description each week on how they used the tool. The main principle was that a student were asked first to send text to ChatGPT and then to human to read.

The university’s own interface “CurreChat” connected to OpenAI’s ChatGPT3.5. The use of university’s own interface was seen as important because it was not wanted that students would have to log in to the services with university’s external IDs. In addition, the interface ensured more secure connection to language model. The assumption was that the material students entered into the application would not be reused elsewhere.

Students reported that they used the tools in a variety of ways as part of their course assignments. Some tried the application in a wide range of ways while others were more cautious. Some reported that they benefited from using the generative AI tool at different phases of their work, while others found it rather useless in their work. A key observation we made from the teaching experiment is that joint practicing and instruction in the use of generative AI tool is important for the tool to best support students’ creative work and learning.

Vilhelmiina Harju, University Lecturer, Faculty of Educational Sciences

Kalle Juuti, Professor, Title of Docent (pedagogy of science), Faculty of Educational Sciences

November 13, 2023November 13, 2023

Introducing the Visual Consistent Character Creator: A New Era with GPT Builder

Introduction

Welcome to an exciting exploration of the innovative new GPT Builder tool (by OpenAI) and my ambitious first project with it — the Visual Consistent Character Creator, an attempt to unlock the holy grail of the text to image sphere. This pioneering tool represents a major leap forward in AI-assisted creativity, combining GPT Builder’s capabilities with the user’s unique imagination.

What exactly does it do? The GPT Builder assists you in creating your own personal AI assistant tailored to your needs – in a nut shell.

In this case, my aim was not just to create a character generator, but to come up with a tool that allows me to construct aesthetically pleasing and above all consistent looking AI-generated characters. As a cherry on top I had the GPT Builder formulate the prompt in a Midjourney readable syntax.

John, a fictional character generated using a tailored GPT with OpenAI’s GPT Builder. The initial generation of John by DALL-E.

The Innovative Process: Enabling Detail and Consistency

My Visual Consistent Character Creator enables this through three key capabilities:

Comprehensive trait selection — this allows for diverse and highly customised characters.
Sequential, step-by-step trait selection — this ensures (or at least strives to achieve) coherence and precision in line with GPT Builder’s innovative approach.
Quality check through AI-enhanced portraits using DALL-E — as a first step to ensure some level of consistency has been achieved before moving to Midjourney (or other text to image generators).

By combining these strengths, my tool can cater to a wide spectrum of creative needs while maintaining visual consistency and artistic flair. Let’s have a look at the process.

The Innovative Process: Detail and Consistency

1. Comprehensive Trait Selection At the outset, I focused on defining a wide array of character traits, mainly physical attributes (and some secret ingredients I am not revealing). I created a template for this, a matrix of sorts. This was done keeping in mind the need to match the high standards of character portrayal as seen in Midjourney. Every trait was carefully chosen to ensure that my GPT Builder could cater to diverse creative needs while maintaining a high level of detail and visual consistency.

2. Sequential Interactivity for Enhanced Precision A standout feature of the Visual Consistent Character Creator is its methodical, step-by-step trait definition process. Reflecting the innovative approach of the GPT Builder, this process ensures that each character trait is not only distinct but also contributes to a coherent overall portrait. This phase was somewhat tricky as the GPT Builder, although always complying, not always “remembered” my instructions and occasionnally would show signs of hallucinations.

3. AI-Enhanced Portrait Prompt Meeting Midjourney Syntax In the final step, after the Visual Consistent Character Creator has summarised the character’s traits and the user has confirmed them I instructed the tool to generate two types of portraits — a detailed close-up and a full-body image with DALL-E which conveniently sits in this workflow as it is part of OpenAI’s ecoystem, so the user never has to leave the browser window. At the very end the Visual Consistent Character Creator creates a prompt the user can copy&paste into Midjourney.

After some testing however, I have to say that the consistency of the characters is quite impressive, but only when generating with DALL-E. When exporting the prompts to Midjourney, the consistency is less evident. A major advantage of integrated DALL-E in ChatGPT is the possibility to discuss the result with GPT and ask for modifications on the generated image. This is huge!

The Potential Impacts: Opportunities and Challenges

By significantly enhancing the character design process, a visual character designer assistant like I just build with GPT Builder could revolutionise creative sectors like gaming, animation and graphic novels. The ability to quickly generate consistently looking, detailed and high-quality characters could greatly accelerate production and encourage more experimentation.

However, this also risks reducing human input in creative roles. As AI becomes increasingly capable of mimicking human artistry, important questions around originality and authenticity arise. While AI art tools offer exciting new opportunities, maintaining a balanced perspective regarding their applications will be vital so that we can benefit from their potential while responsibly managing their risks.

Overall, as an innovative new frontier in AI-assisted creativity, this tool promises to take character design into an exciting new era. By harnessing its capabilities thoughtfully, we can unlock immense creative potential.

Screenshot of ChatGPT interface creating iterations of John, a fictional character generated using my tailored GPT with OpenAI’s GPT Builder. Generated by DALL-E.

November 8, 2023

Serendip – an Immersive Sustainability Learning Adventure

Launching Serendip project on November 9, 2023!

When we started the Global campus project in 2022, we were given the freedom to experiment the limits of online learning. We were expected to do really bold, even risky EdTech experiments. So, we thought very carefully how we could use our time wisely. We wanted to know what this university wants or needs? What could be something bold that would benefit all the members of the university community regardless of the faculty and beyond?

One of the strategic goals for the University of Helsinki is to advance ecological sustainability and responsibility. The University is dedicated to integrate the themes of sustainability into all education programmes.

Well-designed digital and physical environments for work, teaching and learning will enhance ecological sustainability and promote encounters with others, support creativity, renew forms of collaboration and improve accessibility.

(University of Helsinki Strategy 2021-2023)

Following this mission, sustainability became a topic that would be the glue of our work. In the design process, we asked from teachers and students what they are missing regarding sustainability education. We learned that a virtual space where students would gather together around the world to solve the sustainability challenges would be the secret wish of the sustainability teachers.

Students, on the other hand, wanted to travel in 3D worlds and learn how to influence stakeholders. They wished to improve their skills in finding the intervention points in decision-making processes. Students also desired to see hope and use their all senses. We knew we wanted to do this. And this was the foundation for a bold EdTech experiment, the project called Serendip*.

Based on our pedagogical framework, we believe that learning should be engaging and fun but also at the same time personalized and efficient. By offering students a visually appealing virtual reality learning environment with diverse multi-disciplinary learning content and a chance to actually train the sustainability competencies, we can help students to become the change agents this world needs.

The learning content has been developed together with researchers, teachers and students from different disciplines. The research-based content together with state-of-the-art technologies make an engaging learning experience. In virtual reality we could make impossible possible, travel in time and place and practice empathy.

Also, we identified that by taking the AI tools to the next level, we could increase the interaction between a student and the learning content. Therefore, we designed virtual AI-powered characters for different pedagogical purposes for the game. Each discussion is different and personalized, based on the student´s own interests.

The first game episode, the Boreal Forest, one of the tipping elements in earth´s climate system, is an adventure through snow and woods. It combines forest economy, forest ecology and well-being with Indigenous studies. It helps the students to practice their systems-thinking, values-thinking and intrapersonal skills.

We see that you have a role to play in sustainability, so we are happy to invite you to participate as a teacher, a student or a subject-matter expert and co-create with us the further episodes. Learn more on serendip.fi and join the adventure by sharing us how you would like to take part by filling in the form. Can a learning environment for the sustainability education look like this?

* Serendip = The word serendipity, originating from an old Persian fairytale “the Three Princes of Serendip”, means unplanned fortunate discoveries. The Serendip Learning Adventure is based on serendipitous learning approach where, through exploration, learners might discover unexpected and interesting connections among phenomena which can lead to meaningful learning. Serendipity, as valuable unexplored sources for learning, can be fostered through engagement and interaction. We see that sustainability challenges need innovations which can be results of serendipitous events.