AI-Driven inclusive learning in a co-creative environment

Traditional academic programs often assume a homogenous set of learners, pushing students to adapt to a fixed structure. However, learners come with a range of challenges, from sensory impairments and emotional barriers to linguistic obstacles and other learning difficulties. The prevailing one-size-fits-all strategy falls short of addressing these diverse needs.

Universal Design for Learning (UDL) (Rose 2000), rooted in cognitive neuroscience, fosters a versatile educational landscape. Instead of forcing students into a predetermined mold, UDL shapes the curriculum around each student’s needs.

Opportunities for integrating generative AI in UDL

The expanding realm of artificial intelligence, particularly generative AI, holds transformative potential for the UDL paradigm. Generative AI, a subset of Artificial Intelligence, autonomously produces content ranging from text and images to videos, other multimedia, and even software code, all in response to prompts articulated in simple language. This technology powers platforms like ChatGPT, enabling dynamic human-like interactions through conversational mediums.

The educational landscape, particularly within the UDL framework, is ripe with opportunities for integrating generative AI and ChatGPT-styled interactive agents. As the democratization of AI accelerates and open-source models grow increasingly available, there lies an opportunity to weave these innovations into the UDL fabric. Essentially, AI can dynamically address the multifaceted profiles of learners, ensuring tailored support for each individual’s requirements.

While initial apprehensions suggested that tools like ChatGPT might undermine student creativity or facilitate academic dishonesty, a deeper understanding of generative AI has unveiled its transformative potential for learning (Angela et al. 2023). Today, this technology has catalyzed an expanding research domain dedicated to exploring the advantages of integrating generative AI into the educational sphere.

The realm of AI in education is vast and constantly evolving. Current areas of interest include adaptive personalized learning (Sayed et al. 2023), fostering critical thinking and problem-solving abilities (Ashley et al. 2023), imparting metacognitive skills (Grubaugh et al. 2023), promoting self-reflection (Afzaal et al. 2023), facilitating idea generation (Selker et al. 2023), curriculum development (Tavakoli et al. 2022), experimenting with diverse problem solutions, and utilizing conversational agents like ChatGPT as tailored coaching and mentoring tools (Su & Yang 2023), among other avenues. Indeed, AI heralds a major shift in the educational paradigm.

As a testament to this transition, at Haaga-Helia, we are conducting research on teaching metacognitive skills using generative AI (Khan & Alamäki 2023). Our focus is on leveraging generative AI tools that enable students to introspect and analyze their cognitive processes. The essence is to investigate how AI can be used to equip students with the skills of self-directed, lifelong learning.

Addressing inclusivity and diversity

Every student is a unique individual with distinct needs, strengths, learning preferences, and challenges. For example, for students who grasp concepts better through visual representation, generative AI tools such as Synthesis, DiaChat, DiagramGPT, ChartAI, Craiyon, Dall-E, and mid-journey can create visual content. Meanwhile, for those who benefit from audiovisual materials, textual information can be seamlessly converted into dynamic lectures and presentations with AI avatars using tools like Elai, GliaCloud, and Pictory. Advancements in AI also ensure inclusive access to content for hearing-impaired students through automatic captioning or real-time translation into sign language (e.g., signer.AI and live transcribe).

Additionally, for those facing learning challenges, complex concepts can be simplified (e.g., Eli5). By introducing these intricate topics into AI platforms, educators can obtain clearer, more accessible explanations. This content can then be molded into engaging audiovisual presentations using platforms like Elai.

Moreover, generative AI tools such as ChatGPT can assist in the pedagogical technique of scaffolding, where initial support is provided to solve a problem (Exintaris et al. 2023). Later on, the students are asked to solve a similar problem generated by ChatGPT without any support.

The Large Language Models (LLMs) like GPT-4 are not just enhancing traditional search engines but are creating platforms that cater to diverse learning modalities. Platforms like Perplexity AI, Hugging Chat, YouChat, Bing Chat, Elicit, and Google’s Bard ensure that educational content is not only accurate and updated but also accessible to all. Their interactive conversational abilities can be harnessed to design lecture materials and presentations that cater to varied learning preferences.

For instance, when preparing content for visually inclined learners, an instructor might use these AI tools to fetch comprehensive visual data. Alternatively, for auditory learners, the same content could be transformed into interactive discussions or lectures. This AI-driven approach ensures that every learner, regardless of their unique challenges or preferences, finds an entry point into the content, promoting a truly inclusive learning environment.

On the other hand, specialized AI utilities like ChatPDF and Typeset can summarize lengthy academic papers and books, rendering them into more digestible formats. These tools can also cater to specific queries related to the documents, offering precise responses. Such capabilities can immensely benefit students in grasping complex theories and can assist educators in streamlining content, ensuring it is tailored to specific educational objectives and easily understandable by their target audience.

Generative AI-based assessment and reflection

The scope of generative AI tools extends beyond merely crafting personalized assessments and tasks for students. They hold the potential for assessment and reflection (Natalie et al. 2023).

Take ChatGPT, for instance. By integrating detailed student profiles and comprehensive course materials, the AI can generate a myriad of problem scenarios and education cases uniquely suited for diverse student groups, allowing students to consult these intelligent platforms for insights, solutions, and even real-time assistance. Once students submit their responses, ChatGPT can be employed for assessment and feedback.

Specialized AI Tools

ChatGPT is just being used as an example here. The expanding field of AI in education is giving rise to more specialized and domain-specific tools tailored to meet specific needs across varied educational levels and demographics.

For instance, an education-specific conversational agent, called EdGPT caters specifically to young learners aged 3 to 7 by providing age-appropriate answers. EduChat, an open-source platform emerging from China and serving a trifold audience – students, educators, and parents – offers course tutoring, advising on potential career paths, and offering emotional support. A Chinese group, called TAL Education Group, has recently developed a large-scale mathematical model, called MathGPT, that focuses on mathematics-related problem-solving and lecturing (Soumyakanti et al. 2023).

These examples showcase the multifaceted roles a well-designed conversational agent can play in the educational ecosystem. The proliferation of AI tools and the availability of open-source LLMs are propelling the educational sector into an era of unprecedented opportunities. Platforms like HuggingFace host models like Llama-2 and the growing compendium of rich data sources provide a robust starting point.

Dynamic classrooms: AI and human synergy in co-creative education

AI-driven education has the potential to transform classrooms into dynamic ecosystems fostering collaborative innovation. This paradigm shift will see stakeholders – teachers, education experts, students, parents, and AI researchers – come together in a co-creation.

Envision a classroom where educational content undergoes a transformation, driven by AI but with the human touch of educators, aided by the wisdom of education experts and AI specialists. The process commences with a meticulous AI-assisted analysis, discerning each student’s unique requirements, difficulties, and merits. Subsequently, the contents are transformed into multiple representations using generative AI to resonate with the AI-based assessment.

The students actively interact with AI-generated assignments and projects, immersing themselves in the content through digital devices. They harness tools that convert text to speech (e.g., Acoust), transform speech to text (e.g., Diplop and Dragon for students and teachers with difficulty writing or typing), or even translate text into video (e.g., Elai). This two-pronged approach of human teacher and AI facilitator maintains the irreplaceable value of the human touch. Moreover, with the capability to monitor students’ emotional states and engagement (e.g., Intel’s Class and myViewBoard Sense), AI can discern their passions and hurdles. This analysis coupled with the feedback from students, teachers, and parents can be used for fine-tuning the teaching contents and methods.

At the heart of this ecosystem is a dynamic repository hosting AI tools and diverse learning resources. Based on each student’s unique profile, the repository recommends and adapts, ensuring that every learner’s journey is tailored to their individual needs.

Harnessing AI’s potential, navigating ethical terrain, and promoting human ingenuity

AI’s purpose in education is not to eliminate human instructors but to augment their capabilities, bringing a synergy that elevates teaching to unprecedented heights. However, this fusion doesn’t come without its challenges. Ethical considerations are paramount. The limitation of LLMs to occasionally produce misleading or false information underscores the importance of fostering critical thinking in students. The landscape where AI plays a role in education is not one where students blindly accept all that is produced, but one where they are empowered to interrogate, validate, and leverage the outputs intelligently.

It is important to note that generative AI tools like ChatGPT do not truly “think” or “innovate” in the human sense. Their ‘creations’ are sophisticated predictions, a mosaic of their vast training data. They do not craft something out of nothing: they remodel what they have been fed. This distinction is pivotal. It underscores that human creativity remains a unique, unparalleled force. AI cannot eclipse it: at best, it can serve as a catalyst, sparking or refining human ideas. The ultimate vision for AI-driven education should thus hinge on this principle.

As digital literacy was perceived as a foundational skill a couple of decades ago, we now stand on the cusp of another transformative era: the age of prompt engineering which is the art of effective interaction with AI. This skill will be pivotal for educators and learners alike. The dynamic classrooms will be the training grounds for mastering this discipline.

With the higher integration of AI in education, encapsulated by the dynamic classroom’s vision, several other ethical considerations might surface. Foremost among these are data privacy and security, and the proper sharing of student progress and feedback. And beyond the data, the very nature of AI-mediated content requires careful design. We must avoid fostering an over-reliance on AI, preventing a mindset that is excessively narrow or overly reliant on technology, which could hamper cognitive growth.

For illustration, the following images of a dynamic classroom were produced by Synthesis. Initially, the entire article was fed to ChatGPT Splitter, which automatically divided it into manageable chunks suitable for ChatGPT. After feeding the chunks to the ChatGPT, it was tasked with crafting an appropriate prompt to guide a generative AI in visualizing a dynamic classroom. The prompt formulated by ChatGPT was then fed into Synthesis, resulting in the images presented below.

Images of a dynamic classroom produced by Synthesis

This illustrates the utilization of various AI models to achieve a specific outcome. Since ChatGPT cannot directly generate images (at the time of writing this article ChatGPT did not have Dall.E 3 Beta), it was employed to comprehend the article’s semantics and produce a prompt for the image-generating model, Synthesis. Such a prompt can subsequently be refined through an iterative approach to obtain the desired results. Similarly, a combination of models can be adeptly applied to manage different modalities.

Picture: shutterstock