Using Generative AI in Higher Education

Insights from cognitive science

Generative AI (GenAI) is here, surprising us with its human-like abilities and prompting us to consider how it compares to human intelligence.

Of particular interest is how GenAI might affect developing minds and its potential role in teaching and learning.

The challenge in integrating GenAI in education is double or even triple: professionals may adopt it into their workflow, they may think about how to use it for teaching, and last, how to introduce it into the curriculum and scaffold students' abilities to integrate it efficiently as they gradually develop their expertise.

This might be quite overwhelming, and it's not surprising that some big questions are often raised by many, for example:

Should GenAI be integrated into academic teaching? and How?
Which current skills will become obsolete?
Will academic teaching undergo a complete transformation?

While some answers to these questions imply the need for a paradigm shift, I argue that we already have the tools to find concrete answers and develop solutions. The proposed approach relies on two bodies of knowledge that we already have (or should have) and already use (or should use) in teaching:

Understanding of human cognition and its process of development from novice to expert.
Pedagogical frameworks for designing and applying efficient instruction.

In education, human learning is both the goal and the limiting factor. This stays true even when other things shift. This is the main reason that cognitive science and good old pedagogy remain the backbone of planning and decision-making when approaching this challenge. Let's explore both aspects:

How do people learn?

To tangibly understand essential principles of human learning, let us use a pyramid model, that captures basic ideas about human learning (follow from left to right in the image):

Building Blocks: New ideas (red) are connected to prior knowledge (grey) forming new meaningful structures, depicted as a larger triangle.
Practice is essential for the newly created structure to become increasingly functional, i.e. elicits an efficient behavioural response.
Mastery before combination: Following repeated, spaced and various opportunities to use it, the pattern is deeply embedded and allows fluent and variable use. This is when it can be combined with other structures for further construction.
Higher cognitive functions form by this gradual iterative process: consolidating one layer in order to further construct a more complex one... This idea is modelled by a Sierpinski triangle (right) that represents the everlasting ability to combine smaller-scale pyramids to create more complex ones, as we develop mastery.

The Sierpinski triangle is used below to demonstrate how novices and experts may acquire a new skill - this time we consider integrating GenAI use into our complex tasks.

Using GenAI as experts and s novices - what is the difference?

The key question when integrating GenAI into our routines is: what human qualities do we bring to the table? Let's focus on integrating GenAI in relatively complex task in our daily academic or professional lives. Adopting GenAI is essentially acquiring a new skill that relies on the existing foundational knowledge and skills. GenAI can produce pretty good results when provided with clear goal and precise instructions covering all possible aspects of the task and final product, and when the processes are constantly guided and monitored. Otherwise, results may appear impressive at first glance, but much less so upon closer examination. This aligns with our experience in the academic and professional fields: GenAI can be a powerful tool for experts if they learn how to use it well. However, the big question that concerns educators, is how helpful it may be for novices without those basic knowledge and skills. Let's delve deeper...

What does it take for an expert to acquire GenAI skills?

For those who have ventured beyond initial attempts with GenAI in their field of expertise, the learning curve becomes apparent. Good quality, truly helpful results hinge on some basic knowledge and practice:

What is needed to build foundational GenAI skills (represented by the yellow pyramid):

Understand the technology: how Large Language Models (LLMs) work, how they mimic human language and importantly, their capabilities and limitations.
Learn how to prompt: learn methods like multi-shot prompting (conversational style prompting where you monitor and adjust according to each response), then progress to crafting comprehensive prompts containing all necessary information.
Extensive Practice: deliberately practice using different LLMs for diverse tasks within your field. Trying different ways and methods to achieve different goals, crafting your sequences.

By merging your professional knowledge with these newly acquired GenAI "driving" skills, you can leverage GenAI to streamline tasks, enhance outcomes, and eventually save time, Naturally, consistently ensuring the quality and reliability of the results is a huge part of it.

To conclude, experts need to invest some effort to acquire GenAI skill, and when they do, it becomes an effective tool combined seamlessly with their existing expertise.

But what does it mean for novices?

Novices face a much steeper climb. While foundational GenAI skills are essential, they're insufficient. For non-trivial tasks, novices also need to acquire discipline-specific knowledge and skills, that in turn will allow them to clearly define the desired outcome, to guide the generation process, steer GenAI in the right direction, and crucially, to evaluate and adjust the results.

Simply relying on GenAI will likely yield low-quality results, as we're already witnessing. It is also clear that in such cases, learning did not take place, it was bypassed. This is what we worry about, and this is where we have to rethink how to design learning assignments and assessments.

How should we adapt academic teaching and learning?

GenAI is a newly accessible technology that can provide shortcuts and can yield good results provided that the user has:

1) The knowledge and skills to define, guide and evaluate the process of production
2) Basic GenAI skills - knows how to work with the technology

Experts, have the first and need to acquire the second, it's a relatively small challenge that requires some learning and practice.

Novices, need to acquire both, which is a much bigger challenge, focusing solely on GenAI skills will yield poor results and hinder learning at the same time.

Discplenary knowledge and skills is the bigger challenge, but this challenge is not new. The challenge we face is how to integrate AI responsibly, while avoiding the risk of bypassing crucial human learning.

What is the best way forward? We do know where to start!

When it comes to education, as stated above, human learning is not just the limiting factor, it is also the objective. Thankfully, we have much experience working within those boundaries - and in that sense, the challenge is not transformative but constructive.

The main challenge that we now face is to be more precise about our expected outcomes and to ensure we can guarantee that these outcomes are indeed achieved by human intelligence. That not a new challenge, so what has changed? Since learners can now bypass learning phases more easily, teachers should be more pedagogically careful, not to bypass essential checkpoints. For example, submitting a short essay as a checkpoint for reading and understanding a paper may not be sufficient anymore.

The tools to apply careful pedagogical design exist, our challenge is to use them more and use them properly. We should start with well-defined learning outcomes as backward design suggests, and continue with a careful distinction between formative and summative assessment tools.

Below, is a (very general) suggested pedagogical workflow for the design or redesign of an academic course, that considers GenAI challenges on one hand, and the gradual and effortful nature of human learning on the other.

1️⃣ Review and redefine learning outcomes: what should students really be able to do (by themselves) by the end of the course? Do the goals truly reflect the most advanced skills students are required to demonstrate? Does Gen AI change things? In that respect, there are two major points to be careful about:

For a whole group of basic skills (like writing), GenAI may seem to render them less critical. However, we must ask ourselves if this basic skill is not required as part of the skill set that is needed to guide, drive and evaluate GenAI results as a professional. And if it does (and writing does) we should not skip it. Moreover, we should not ask for something more advanced (e.g. editing) just because GenAI can do the basic stuff. We should focus on the human skills.
For identified areas where GenAI integration is relevant, experts should first explore the possibilities by themselves. And then deduce the expected outcomes, the skills and knowledge that learners will need to drive drive GenAI responsibly.

2️⃣Reconsider and design a summative assessment that aligns with the outcomes. The well-established distinction between summative and formative assessments takes on renewed significance. Summative assessments should primarily gauge student mastery, or "exclusive human abilities". As a lot is being written and said about how to integrate GenAI in different ways in assessment, I wish to highlight two cases where we should emphasize summative assessment without GenAI:
1. When we assess basic human skills GenAI can do better than students.

2. When GenAI can be integrated to yield better results, but the ways to do it were not explicitly demonstrated and taught throughout the course.

We should make sure that summative assessment clearly signal to students the course expectation, that is fair, valid and reliable,

I hear claims that not using GenAI in academic assessment is becoming irrelevant as GenAI is immersed in the professional world. Conversely, I think academic studies are about building the basic skills that will allow students to act in this world, and those basic skills are still extremely valuable. This is where the expert-novice distinction is crucial: novices need displinary knowledge and skills in order to use GenAI responsibly.

3️⃣Scaffold Formative assessment and assignments. Formative assessments, designed to support the learning process, become even more crucial with GenAI integration. If we accept the underlying assumption that effective GenAI use builds upon existing professional knowledge and skills, then scaffolding formative assessments to guide students toward this goal becomes paramount. While designing assessments specifically for GenAI may require some trial and error and some creative ideas, we are already familiar with the core principles and systems of effective scaffolding to build upon, they may be more relevant than ever. For formative assessment, grading methods that favour effort over success (like zero or low-stakes grading) may be critical to signal that GenAI, in case used, is supposed to support learning not to bypass it.

4️⃣Teach the knowledge base, While AI might seem like an attractive shortcut for knowledge acquisition, a solid foundation remains essential. As the pyramid analogy illustrates, advanced skills crumble without a solid knowledge base. While we can guide students to use GenAI to support and supplement their learning (by providing answers, explanations, and examples), we should not assume that this crucial stage can be trimmed or skipped. Teachers are irreplaceable for paving learning pathways: selecting the relevant knowledge, organizing, sequencing and using pedagogical tools to support students learning and practice of essential knowledge. This message should be embedded in assignments and assessments and communicated clearly - knowledge is power.

The way forward - GenAI and higher education

GenAI's arrival might seem transformative, but the pace of this transformation is dependent on the human adoption rate. We've seen that professionals can potentially integrate the new technology into their work with a relatively moderate amount of practice, though still effortful and deliberate. When it comes to education, we must consider an additional significant factor - GenAI's potential to seemingly bypass essential parts of the learning process. The opportunities to shortcut and bypass learning are not new, but they are more accessible, tempting, cheap and rewarding with the emerging technology.

This means that the challenge for professionals who also happen to be educators is at least double:

Actively and deliberately explore GenAI's capabilities within the discipline, and how to exploit them in combination with existing knowledge and skills.
Adapt and refine the pedagogical approach to address the challenge, while emphasizing established effective pedagogy to support learning: accurately define learning outcomes, align appropriate summative assessment (without GenAI if needed!), scaffold formative assessment and supporting assignments and teach all necessary knowledge.

This is a true challenge, that undoubtedly requires exploration, innovation and creativity. However, the way forward is clear as human-intelligence-focused pedagogy remains highly relevant. In fact, the recent advancements challenge us to apply it with even greater rigour than ever before.

This post started as a Twitter/X thread:

Sierpinski Tetrahedron image source: https://en.m.wikipedia.org/wiki/File:Sierpinski_tetrahedron_by_George_W._Hart.jpg

Published: May 2024, Updated: December2024

Google Sites

Report abuse