Linking Kolb’s learning theory and GenAI-mediated learning: a brief literature review

A search for research connecting experiential learning cycles with generative AI in education

A focused, quick literature review examining papers that try to connect Kolb’s Experiential Learning Theory and GenAI-mediated learning somehow, with the intent of comparing our own to-be-written framework.

Keywords

experiential learning, Kolb’s learning theory, generative AI, GenAI-mediated learning, higher education, learning frameworks, literature review, educational technology

1 What is experiential learning?

To make sure I’m more caught up with the latest thinking on experiential learning, I chose to read the 2023 book by Cecilia Chan (Chan, 2023) on the topic rather than relying on David Kolb’s original 1984 book - even though it has a more recent, 2015 edition (Kolb, 2015).

Kolb’s notion of experiential learning is that knowledge is created through the transformation of experience and is conceptualised as containing four stages:

1. Concrete Experience (by experiencing from prior or from new experience),
2. Reflective Observation (by reflecting on the experience),
3. Abstract Conceptualization (by rethinking and clasping the experience), and
4. Active Experimentation (by reapplying and testing what you learnt from the experience and reflection).

In (Chan, 2023, sec. 1.1), Cecilia adds the perspective of other educators and researchers of experiential learning. In its simplest form, experiential learning is framed as ‘learn by doing’ thing but it may also be characterised as a consideration for the learner’s experience (prior, current life events, or those arising from the teaching context) - which is sometimes labelled ‘experience-based learning’. which if you ask me, it feels like a slightly different, although connected, concept. We are also presented to two categories of experiential learning: Field-Based (learning in the field) and Classroom-Based (learning in the classroom) (Chan, 2023, sec. 1.5).

Interestingly, Cecilia also lists the criticisms of this learning theory (Chan, 2023, sec. 1.4), which I copy-pasted here and annotated so I have that always in mind:

• Experiential Learning tends to focus on retrospective reflection, fails to address the “here and now” experience (Vince, 1998); Isn’t this just a reflection of how you apply this framework though? A Reflective Observation stage could very well be a reflection of the here and now experience.
• fails to account for social, political-cultural and institutional infuences on the learning process (Reynolds, 1999); Kolb’s six principles takes that into account, though
• places too much emphasis on the role of the individual learner and decontextualises the learning process (Holman et al., 1997); that critique makes sense to me; indeed the stages are very focused on the experiencie of the individual learner
• does not take into consideration the role of a learner’s intention and desire in learning. The focus on cognitive reflection is simplistic (Britzman, 1998); (Chan, 2023, sec. 1.3) touches on that by adding the student perspective on this learning theory. She cites other researchers who proclaim the obvious: “students’ attitudes infuence their motivation to engage in learning” to make the point that “it is vital to ensure a clear understanding and consideration of students’ perceptions.”. agreed
• does not account for the interaction between cognition and the environment, how each individual’s cognition converges with others,’ or “how individual knowledge co-emerges with collective knowledge” (Fenwick, 2000, p. 263). Kolb’s principle of ‘environment learning’ and ‘constructivism’ take that explicitly into account, though

2 Kolb’s learning theory in a nutshell

Here are some annotations on top of excerpts from (Chan, 2023, sec. 2.2) which is all about Kolb’s specific learning theory.

2.1 The six principles of experiential learning

The six principles of experiential learning include:

1. Learning as a process: “Kolb does not believe in looking at learning in terms of outcomes.” “It is best to assume that learning is without an end.”
2. Learning is relearning: learning is a continuous process.
3. Emotional reflection: “[…] to truly learn, internal and external agreements, conficts and disagreements may exist and drive the learning process.”
4. Holistic learning: it’s not merely about acquiring knowledge but it involves how one behaves, thinks, feels (and how one perceives that feeling)
5. Environment learning: “Learners may perceive the same event differently based on their previous experience.”
6. Constructivism: “Learning is the process of creating knowledge, which is understood as the result of the transaction between social knowledge and personal knowledge.”

2.2 The two continuums and the cycle with the four stages of the learning process

Kolb conceptualises the Processing Continuum - which depicts the learner’s personal approaches to a task - and the Perception Continuum - which depicts the learner’s emotional response to the task. These two continuums are the axes of a diagram in which the cycle of the four learning styles are represented.

“[…]According to Kolb and Kolb (2013), learning occurs in the “resolution of creative tension among these four learning modes”” (Chan, 2023, sec. 2.2.2).

“Kolb views the learning process as people moving along the two continuums, and the effectiveness of learning depends on the balance between these stages.” (Chan, 2023, sec. 2.2.3).

The diagram below shows the four stages of the learning process and the two continuums:

Figure 1: Kolb’s experiential learning theory and learning styles. Source: (Chan, 2023, p. 22).

2.3 The four learning styles

“In layman’s terms, the cycle has four steps – you do something, you reflect upon that action, and based on that reflection, you conclude and modify your understanding before doing that activity or related activities again to see if you have improved, and followed by a further reflection.” (Chan, 2023, sec. 2.2.2).

Referring to the four stages identifed, Kolb proposed four learning styles:

1. Diverging (concrete, reflective) – This learning style emphasises an innovative and imaginative approach in learning and favours observation rather than action. Learners of this style look at things from different perspectives, gather information and solve problems through imagination.

2. Assimilating (abstract, reflective) – This learning style compiles different observations and thoughts to form an integrated view of an event. Learners of this style adopt a concise and logical approach and prefer clear explanations in learning over practical opportunities.

3. Converging (abstract, active) – This learning style emphasises the practical application of ideas and problem-solving. Learners of this style prefer technical tasks and look for practical applications of ideas.

4. Accommodating (concrete, active) – This learning style is defned by solving problems through trial and error. Learners of this style rely on intuition and experience to solve problems.

Note

These four learning styles, grounded on Kolb’s experiential learning cycles, are probably a lot more descriptive and informative than the 4+1 R’s we’ve been exploring in our GENIAL paper.

3 Literature on GenAI-mediated learning <-> Kolb’s learning theory

I searched for papers that connect Kolb’s learning theory to GenAI-mediated learning to some degree. My intent is both to see what’s out there but above all, I wanted to see if other scholars have been trying to develop a conceptual framework for GenAI-mediated learning using Kolb’s learning theory - just like how we are trying to do in our GENIAL paper. For this, I used Elicit and Web of Science to searchfor the most relevant papers.

Click here to understand how I searched for papers

I ran two Elicit research reports to try to gather the most relevant papers to help me understand how Kolb’s experiential learning theory could help us decode the patterns of GenAI-mediated learning. Here are the two research questions I used:

• Report 1: (poor results) “How does the traditional theory of Kolb’s experiential learning cycles explain the cognitive processes of student learning when using generative AI as a learning tool in higher education settings?” 🌐 (link)

• Report 2: (few salvagable references)“What cognitive transformations occur in students’ learning processes when generative AI is used as a supplementary learning tool, examined through the lens of Kolb’s experiential learning theory?” 🌐 (link)

I configured Elicit such that Report 1 returned 40 most relevant papers and Report 2 returned 10. Elicit automatically screens papers for relevance (educational setting of study is higher education, the study is about GenAI tools impact on learning, it is empirical, and Kolb’s learning theory should be explicitly mentioned, etc.).

I was unhappy with the results of Report 1 and Report 2 (the LLM tried to find things that were experiential-learning-like rather than explicit references to Kolb’s learning theory) so I used a search query instead:

• Search Query: (good results) “What research has been done in applying Kolb’s learning cycle as a framework for understanding how learners construct knowledge when interacting with Generative AI technology?” 🌐 (link)

After gathering the reports above, I used Scimago’s Journal Ranking to select only for papers publised in journals that are featured in the Scimago’s Education category (2024). I know I might end up removing some genuine research papers that just didn’t happen to make it to the top 1000 journals in the field but I think it’s worth it so I don’t risk reading papers that have not been scrutinised to a high standard.

• Web of Science search: Then I did another search, this time using Web of Science to find more relevant papers. Using ‘generative AI in education’ and ‘experiential learning’ as topics, it returned 19 papers, several of which I had already come across from the previous search.

Below I highlight the papers I found most relevant and which I think are the most promising to read while I’m working on our GENIAL paper.

Important

⚠ You don’t need to read the rest of Section 3!! ⚠

You can just skip to the Conclusion section below.

In summary, there were few papers that tried to link Kolb’s learning theory as a way to study and understand GenAI-mediated learning. Most papers that mentioned experiential learning in their title or abstract either thought of it as a way to design activities or were referring to it in a loose sense to imply the importance of “real-life experiences” in learning.

I could not find any paper that tried to bring Kolb’s learning theory to the analysis of GenAI-mediated learning or that tried to update the cycles to reflect the new learning styles that GenAI-mediated learning might bring.

(This is good for our GENIAL paper because it would be an innovative contribution to the field.)

I did a close read on a few selected papers that seemed most relevant. You can find my interpretations below - and some in-place annotations in our shared Zotero group genial-project.

3.1 Chiang et al. (2024)’s paper

Perhaps the most relevant paper I found was (Chiang et al., 2024)’s paper, titled “Can generative AI help realize the shift from an outcome-oriented to a process-outcome-balanced educational practice?” published in the Educational Technology & Society journal (rank: 50).

You can find the paper in our shared Zotero group genial-project, with my in-place annotations. The paper is shelved under the “Learning Theories” > “Applications - ELT” folder.

Although they do not use Kolb’s learning theory for analysis of data, they propose how Generative AI can be used to design learning activities at each stage of the learning cycle. Their paper is focused on how to design assessment such that we harness the potential of GenAI to make learning more engaging and meaningful. As part of their proposal, they also suggest the incorporation of a multimodal learning portfolio to capture the learning process and they advocate for a formative-summative mixed assessment approach. All of those are great ideas and great suggestions but it’s not comparable to what we are trying to do in our GENIAL paper (Good!).

Figure 2: Chiang et al. (2024)’s diagram depicting the stages of their activity (Figure 3 in the paper).

I found it unusual that they consider ‘Active Experimentation’ as Stage 1, unlike all the other sources that put ‘Concrete Experience’ as Stage 1. But well, it’s a cycle anyway, so I guess it’s fine to start from any stage.

Just below the diagram, they list the type of activities students would be doing and how GenAI can be used to support them. I replicate their table below:

Table 1: Chiang et al. (2024)’s table depicting the stages of their activity and the roles of GenAI (also Table 1 in their paper).

Stages of experiential learning cycle	AI’s roles in assisting Teachers	AI’s roles in assisting Students	AI’s roles in facilitating the elements of learning portfolio
Stage 1: Active Experimentation	• Crafting hands-on tasks • Providing immediate feedback to students • Designing rubrics for evaluation students’ work • Evaluating students’ works based upon the rubrics	• Providing instruction relevant to task completion • Answering students’ questions relevant to completing tasks	• Collaboration • Mentoring
Stage 2: Concrete Experience	• Designing guidelines for students to recall their experience • Providing instant feedback to students’ work	• Guiding students to document/recall their learning experience	• Documentation • Mentoring
Stage 3: Reflective Observation	• Designing prompts to facilitate students’ reflective practice • Providing feedback to students’ observation notes	• Working as a peer to facilitate students’ reflective observation • Working as a mentor to facilitate note-taking of reflective observation • Documenting students’ works	• Reflection • Collaboration • Mentoring • Documentation
Stage 4: Abstract Conceptualization	• Designing the instruction of higher-level thinking tasks • Designing rubrics for evaluation students’ work • Evaluating students’ works based upon the rubrics	• Working as a mentor or collaborator to facilitate the higher-level thinking tasks • Documenting students’ works	• Collaboration • Mentoring • Documentation

What is nice is about this paper is that in (Chiang et al., 2024, sec. 4) they show a ‘deep dive’ into how these activities were carried out in practice. It’s a nice paper to cite and relate to.

Note

What does this mean for our GENIAL paper?

Here are my thoughts:

• We should cite this paper as a way to show that others have been thinking about connecting Kolb’s learning theory to GenAI-mediated learning.
• We should also point out that their focus has been on designing activities and not on uncovering and analysing the learning process. In our observational study, we will be doing the latter.

3.2 Sun and Deng (2025)’s paper

The Sun and Deng (2025) paper, titled “Using Generative AI to Enhance Experiential Learning: An Exploratory Study of ChatGPT Use by University Students” and published in the Journal of Information Systems Education (rank: 736), also seemed relevant. From the title and abstract it was clear they were going about it the other way around (using GenAI to improve experiential learning) than us (using Kolb’s theory to make sense of GenAI-mediated learning) but still, just like with Chiang et al. (2024), it shows how others have been connecting the two topics.

You can find the paper in our shared Zotero group genial-project, with my in-place annotations. The paper is shelved under the “Learning Theories” > “Applications - ELT” folder.

Their methodology

Once again, the focus was on designing activities rather than analysing the learning process. Oddly, though, they label the stages differently and for no apparent reason. This is where some of my criticisms start. Here is what they did:

1. [Experiencing] The instructor would first ask students to read an article written by Karen Hao ¹ (Hao, 2023) about ChatGPT. Next, the instructor would show a ChatGPT prompt specific to the course they were teaching. Students were expected to imitate those steps ² and were encouraged to re-generate responses for the same prompt to observed how the output differed each time.

2. [Reflecting] At this stage, students were asked to answer (anonymously): “Did the ChatGPT-generated response to the FIRST question meet your expectations? Please explain.” which I don’t think is a superb question to ask if you ask me.

3. [Thinking] Students were then tasked with coming up with a course-related prompt themselves. I wouldn’t call this ‘Thinking’. I guess this would be the stage students would be doing Active Experimentation.

4. [Acting] In the final stage, students answered another survey question: “Did the ChatGPT-generated response to the SECOND question meet your expectations? Please explain.” Again, inadequate labelling of the stage! Sure the students would have already actively experimented with the prompt, but they were not ‘acting’ in the sense of Kolb’s theory after they have finished the activity.

And here’s the diagram they used to depict the stages above:

Figure 1: Sun and Deng (2025)’s diagram depicting the stages of their activity. Re-styled for this blogpost by Jon Cardoso-Silva because the original was too blurry.

The diagram does not closely mirror Kolb’s learning theory and even, in fact, seems to represent a deviation from Kolb’s learning theory, perceiving learning as a linear (the arrows do not depict a cycle) and outcome-based process, one that culminates in learning (as opposed to a continuous process).

Findings

Instructor-provided prompts vs student-generated prompts

One of their findings is that although students were presented to specific style of instructor-provided prompts (“37 students (55.2%) practiced an instructor-provided ChatGPT prompt that involved metacognitive knowledge, followed by 22 (32.8%) for procedural knowledge and eight (11.9%) for conceptual knowledge.”), their own prompts had a different distribution: “Unlike their instructors, only nine (13.6%) of the student-generated prompts involve metacognitive knowledge. The majority (65.2%) of the student-generated prompts require conceptual knowledge, followed by procedural knowledge (21.2%).”

Another contrast was that although the 70 instructor-provided prompts had a full range of Bloom’s Taxonomy levels, the student-generated prompts gravitated more towards the lower levels of Bloom’s Taxonomy (~85% were in ‘Understanding’ and ‘Applying’) and much less towards the higher levels (‘Analyzing’, ‘Evaluating’, and ‘Creating’). The authors do provide a cross-tabulation table to show the divergence in Bloom’s Taxonomy levels between instructor-provided prompts and student-generated prompts with the added factor of ‘No Prior ChatGPT Experience’ vs ‘With Prior ChatGPT Experience’ - which was nice.

They find that those students who had no prior ChatGPT experience were more likely to mirror the instructor’s prompt. “In particular, if the first prompt provided by the instructor is about metacognitive knowledge, compared to other knowledge types, students are more likely to propose a ChatGPT question involving conceptual or metacognitive knowledge.” (Sun and Deng, 2025, p. 58).

Note

What does this mean for our GENIAL paper?

The biggest contrast is that they rely on the experiential learning theory only as an anchor to design their activity and not at all when they analysed the data or when they discussed the results.

For the analysis they stuck to Bloom’s Taxonomy to investigate how students approached the different instructor-provided prompt as labelled by their qualitative coding (“Conceptual Knowledge”, “Procedural Knowledge”, and “Metacognitive Knowledge”).

• We could cite it together with Chiang et al. (2024) to show that others have been thinking about connecting Kolb’s learning theory to GenAI-mediated learning.

3.3 Annamalai and Bervell (2025)’s paper

Finally, the third paper I read more closely was (Annamalai and Bervell, 2025)’s “Exploring ChatGPT’s role in English grammar learning: A Kolb model perspective” published in the Innovations in Education and Teaching International journal.

Just like the previous papers, this one also referred to Kolb’s cycles as the model to use when designing activities but they also explicitly reflect on GenAI’s impact at each stage. Theirs just wasn’t as rigorous a mapping as I saw in the previous papers.

Problem: They ran interviews with 20 of the 28 students who participated in the study with questions aimed at capturing the different stages of Kolb’s learning theory but the way they framed the questions didn’t seem very accurate or robust to me. Under the ‘Reflective Observation’ section they do ask students to reflect on their experience, both on learning with ChatGPT and on how they approach uncertainties in general (fitting of RO I think) but the question used for the ‘Abstract Conceptualization’ section was: “What are your conclusions about learning English through ChatGPT?” which does not capture AC very accurately. AC should be more about “The ability to make sense of what has happened and put this into context of other experiences and learning” ³. I also don’t think “How do you envision utilising ChatGPT for English language skills?” captures AE very well either.

Findings

Perhaps one thing that can be taken away from this paper is under their ‘Data Analysis’ section. They report themes that emerged from running the interview, broken down into the different stages of Kolb’s learning theory:

• CE: the researchers observed and students also reflected on the ‘Interactive Nature’ of ChatGPT, the perception that it can give ‘Immediate Feedback’ (this should always be tempered with scepticism given how unreliable LLMs can be) and valued the opportunities for ‘Bite-sized learning’ it provides.

• RO: students talked about ‘Innacurate information’ (“T19 [a participant] asserted that ‘checking grammar using ChatGPT is so easy, but we found some of the answers are unrelated to the sentences’.”) and the ‘Seeking guidance from peers, teachers and additional resources’ this should not have been presented as an emergent ‘theme’ because it was precisely one the questions

• AC: NOTHING. They don’t have a heading for this stage nor report anything about it!!

• AE: students talked about their ‘Intention to use’ ChatGPT more often after the activity.

Note

What does this mean for our GENIAL paper?

I would only refer to this paper to critique the way they ran the study.

3.4 Other papers

• (somewhat useful) A conference paper by Silva et al. (2025) does not have good content (it’s poor research) but it shows that someone else has tried to find papers that use Kolb’s learning theory as a framework for GenAI-mediated learning and couldn’t find anything.

• (not useful) In a paper by Haywood et al. (2025), published in MDPI’s Education Sciences journal (rank: 377)⁴, the authors describe their experience integrating AI into their experiential learning activities but they just mean it loosely and don’t use Kolb’s framework directly nor describe the different stages at all. Here is a quote that shows how they perceive experiential learning: “The learning methodology in this unit was Experiential Learning, as an effective way to apply the students’ understanding of HRD knowledge and skills to real-world situations. Experiential learning is a student-centered methodology that engages students in developing critical thinking through undertaking an authentic and meaningful project.”

In my search, I found several other papers (a few dozen) that were about GenAI-mediated learning and mention experiential learning in their title or abstract but none of them were worth a close read (as I did with the three ones I listed above) nor worthy of mentioning here.

4 Conclusion

In our GENIAL paper, we want to consider Kolb’s learning theory as part of our framework to make sense of how learning as a (continuous) process is mediated by GenAI. We will also rely on other bits from TPACK, some taxonomies (4+1 R’s, Anthropic’s quadrants perhaps, etc).

Out there in the literature, there are just a few papers that reflect on how Kolb’s learning cycles are mediated by GenAI in a similar sense to how we want to present our study.

• Chiang et al. (2024) is definitely the most relevant of them. Even though its goal is to design assessment that make the most of GenAI (and a learning portfolio), it does assume Kolb’s cycles are a good way to understand how learning happens and presents some useful reflections in there.

• Sun and Deng (2025) shows another attempt to connect Kolb’s learning theory to GenAI-mediated learning, but I find that it deviates a bit too much from Kolb’s stages.

• Annamalai and Bervell (2025), despite being a paper that explicitly reflects on GenAI’s impact at each stage of Kolb’s learning theory, the way they framed the questions for the interviews and summarised their findings was not very accurate or robust.

Take-away: Linking Kolb’s learning theory as a way to make sense of GenAI-mediated learning presents a good gap in the literature that we will be able to fill with our study.

---

References

Annamalai, N., Bervell, B., 2025. Exploring ChatGPT’s role in English grammar learning: A Kolb model perspective. Innovations in Education and Teaching International 1–17.

Chan, C.K.Y., 2023. Assessment for experiential learning, 1st ed, Routledge research in education. Routledge, United Kingdom.

Chiang, Y.V., Chang, M., Chen, N.-S., 2024. Can generative AI help realize the shift from an outcome-oriented to a process-outcome-balanced educational practice? Educational Technology & Society 27, 347–385.

Hao, K., 2023. What is ChatGPT? What to know about the AI chatbot. The Wall Street Journal.

Haywood, S., Padurean, L., Ralph, R., Tobias Mortlock, J., 2025. From Intimidation to Innovation: Cross-Continental Multiple Case Studies on How to Harness AI to Elevate Engagement, Comprehension, and Retention. Education Sciences 15.

Kolb, D.A., 2015. Experiential learning: Experience as the source of learning and development, 2nd ed. Pearson Education LTD, Upper Saddle River, New Jersey.

Lithner, J., 2015. Learning Mathematics by Creative or Imitative Reasoning. In: Cho, S.J. (Ed.), Selected Regular Lectures from the 12th International Congress on Mathematical Education. Springer International Publishing, Cham, pp. 487–506.

Silva, L., Caldas, E., Borges, M., Damasceno, A., Oliveira, F., 2025. Evaluating the Adherence of Synthetic Digital Educational Content to Kolb’s Learning Theory: In: Proceedings of the 17th International Conference on Computer Supported Education. SCITEPRESS - Science; Technology Publications, Porto, Portugal, pp. 717–725.

Sun, R., Deng, X., 2025. Using Generative AI to Enhance Experiential Learning: An Exploratory Study of ChatGPT Use by University Students. Journal of Information Systems Education 36, 53–64.

Footnotes

1. famous tech journalist and author of the ‘Empire of AI’ book

2. something that reminds me of (Lithner, 2015)

3. As greatly expressed in this University of Cumbria document

4. MDPI is considered a ‘predatory’ publisher these days.