The Use and Misuse of Generative AI in Academic Education

Exploring New Developments and Implications

Gerko Vink

9 Apr 2023

Disclaimer

This presentation is a collection of my thoughts and opinions. It does not necessarily represent the views of Utrecht University or the USO Consortium AI in Higher Education.

These materials are generated by Gerko Vink, who holds the copyright. The intellectual property belongs to Utrecht University. Images are either directly linked, or generated with DALL-E. That said, there is no information in this presentation that exceeds legal use of copyright materials in academic settings, or that should not be part of the public domain.

You may use any and all content in this presentation - including my name - and submit it to generative AI tools, with the following exception:

You must ensure that the content is not used for further training of the model

Source:

First of all

Education is a universal right

Everyone has the right to education. Education shall be free, at least in the elementary and fundamental stages. Elementary education shall be compulsory. Technical and professional education shall be made generally available and higher education shall be equally accessible to all on the basis of merit

Source: Universal Declaration of Human Rights as declared by the United Nations in 1948.

Introduction to gen AI

What is Generative AI?

Generative AI refers to a subset of artificial intelligence technologies capable of generating new content, ideas, or data that resemble human-like outputs.

Generative AI operates through advanced machine learning models, particularly deep learning networks.
These models are trained on large amounts of data in a specific domain (e.g., text, images, video) and can then generate new outputs based on the patterns and features they have learned.

Some common examples of generative AI technologies include:

Chatbots and virtual tutors: AI-driven chatbots can provide personalized tutoring, answering student questions and offering explanations on a wide range of subjects.
Content creation tools: Tools like GPT (Generative Pre-trained Transformer) can assist in creating educational content, generating lecture notes, or drafting exam questions based on certain criteria.
Automated essay scoring and feedback: AI models can grade essays and provide feedback to students on their writing.

How Do AI Tools Work (in 10 steps)?

Machine learning enables computers to learn from data, identify patterns, and make decisions with varying levels of human intervention.
Machine learning methods often work by means of training, for example to perform a single task or set of tasks.
Some forms of models are not specifically programmed to perform certain tasks and learn withouth human intervention.
The domain of machine learning that focuses on such human brain-like deep learning strategies is called neural networks (NNs).
Transformers (Vaswani et al. 2017) are a type of neural network that have advanced deep learning, particularly in the domain of natural language processing.
Instead of modeling words one after another, transformers can look at whole sentences or paragraphs at once and take the context of text into account.
It is not hard to imagine that contextual understanding allows for better learning and more flexible applications.

How Do AI Tools Work (in 10 steps)?

Generative AI tools, such as Gemini, chatGPT or GitHub Copilot are applications of transformers.
These tools have been trained on extremely large amounts of data (mostly text) and have learned from the structure, meaning and nuances of the language.
After training, these tools can generate new text that both coherent an contextually relevant.

Don’t be fooled! It’s just a prediction model

Who develops generative AI tools

Generative AI technologies are often developed by publicly traded companies with:
- Deep pockets
- Shareholders
- Motives and goals that may diverge from academic or public interests
These technologies also require significant computing resources for training and operation
- This skews access and development to those with substantial resources.
- If you feel sad or overwhelmed by the low sense of control, read Choose Your Weapon: Survival Strategies for Depressed AI Academics

Impact of generative AI

Impact on education

There is a finite number of options for dealing with generative AI in education. These options translate to the following stages of AI grief:

Ignore the existence of generative AI
Forbid the use of generative AI
Circumnavigate generative AI by using offline assessment modes like pen/paper exams or oral exams
Test around Let students perform tasks that generative AI struggles with
Embrace generative AI and allow it in course work
Rethink the way we assess students and develop new assessment methodologies

Futureproof education?

Could it be

That our defined learning goals, our evaluations and the skills that we teach are not aligned with the future professional needs for our student body?

AI in assessment

It is crucial to think about proper assessment designs for an AI-enabled student body.

Here are some links that I find useful:

Impact of AI on Cognitive Performance

Impact of AI on cognitive performance

Impact on society

There is also a societal impact of generative AI technologies
- They can create divides and inequality
- There is evidence that the workers who curate these models are treated unfairly or even inhumanely by their employers. This interview from last month also paints a good picture of how and where AI work can harm people.
The energy consumption required for training generative models contributes to carbon emissions.
- This also holds for interacting with and applying these models
- There is already a larger than necessary hidden carbon footprint
The Dunning-Kruger effect

Policy

Current at our university

Towards responsible genAI use

I believe that we should demand from the UU community to follow these simple steps when using generative AI tools:

Minimize the use of AI tools, as they are (currently) environmentally unfriendly
Don’t input confidential or personal information
Don’t input information that violates IP or copyright
Don’t violate IP with using output from the tool
Confirm the output accuracy
Check the tool output for bias
Disclose the use of AI tools in your work
Don’t do naughty or unjust things with AI tools
Ask the tool not to use input for training
Open the content you produce by using AI tools as much as possible with permissive licenses

If in doubt about any of the above, don’t use generative AI tools.

For teachers: Put a disclaimer on your materials that governs what is or is not allowed when using generative AI tools.

Promises that have been made

Enhancing Learning with AI

Personalized Learning Pathways: Generative AI tools are paving the way for personalized education, adapting in real-time to the learning pace and styles of individual students. This leads to a more engaging and effective learning experience, tailored to the needs and strengths of each learner.
Interactive Content Generation: These tools can produce dynamic educational content, including interactive simulations, customized quizzes, and virtual labs, making complex subjects more accessible and engaging for students.
Improved Engagement and Motivation: By providing instant feedback and fostering a more interactive learning environment, generative AI tools have the potential to increase student engagement and motivation, crucial factors for successful learning outcomes.
Augmented Creativity and Problem-Solving: With the capability to suggest multiple perspectives on a given topic, AI can enhance students’ critical thinking and creativity, encouraging them to explore novel solutions to problems.
Access to Quality Education: Generative AI can democratize education by offering high-quality, personalized learning experiences to students in remote or underserved regions, breaking down geographical and socioeconomic barriers to education.
Support for Instructors: By automating administrative tasks and offering insights into student performance, AI tools allow educators to devote more time to teaching and personalized interaction with students, enhancing the overall educational experience.

Enhancing Learning with AI

Personalized Learning Experiences: Generative AI can create customized educational content that suits the learning pace and style of individual students. This approach has been shown to improve engagement and comprehension in subjects ranging from mathematics to language learning.
Enhanced Research and Writing Assistance: For academic research and writing, generative AI tools like automated literature review generators and citation managers have been developed.
Simulations and Virtual Labs: In science and engineering education, generative AI has enabled the creation of highly detailed simulations and virtual labs.
Language Learning and Enhancement: Generative AI has been particularly impactful in language learning, offering tools for automatic translation, conversation simulation, and pronunciation correction.
Feedback and Assessment Tools: AI-driven assessment tools can provide instant feedback on student assignments, ranging from essays to complex problem sets.
Facilitating Collaborative Learning: Generative AI can create scenarios or projects where students need to collaborate to find solutions, fostering soft skills like communication, teamwork, and problem-solving.

When asked for proof of enhanced learning?

Intelligent Tutoring Systems and Sustainable Education: A systematic review has explored how AI supports sustainable education by changing teaching scenarios to more remote, virtual, and blended formats. It highlights the application of AI in analyzing learning behaviors, performance prediction, and personalized learning interventions. This integration aims for better adaptation to students’ real-time learning status and early assistance provision (source).
Adaptive Learning Techniques for Personalized Education: The study discusses the use of e-TPCK, an adaptive electronic learning environment designed to support the development of student-teachers’ Technological Pedagogical Content Knowledge (TPCK) in a personalized manner. The e-TPCK system aims to engage learners in personalized learning experiences, addressing their diverse needs and preferences. (source).
Personalized Learning and Academic Assessment: The University of South Australia has developed learner profiles that provide real-time analysis of a student’s learning behaviors and wellbeing. This initiative aims to significantly improve teaching and learning quality by allowing educators to identify and respond to each child’s needs promptly. Furthermore, the OnTask project enhances academic experiences by providing personalized feedback and suggestions for better learning experiences. (source).

Please give me proof that AI enhances learning; I’d like to believe it so badly

Bottomline:

AI is supposed to enhance learning, but the evidence is still scarce. Promises are made about

personalized learning paths
adaptive learning
enhanced writing assistance
improved motivation
accessibility to quality education
support instructors

On top of that:

AI is supposed to make education more inclusive
- this is not the case, as the same level of AI is not accessible to all
- AI itself is not always inclusive, as it can be biased

How Students and Teachers Use AI output

Students and Teachers: Expected Uses of AI

We have seen the promises that have been made in the previous section. These promises assume that students and teachers will use AI in a certain way. Do they?

To realize the potential gains, students and teachers are supposed to take an active, knowledge driven approach to AI. This means that they should:

engage actively
critically evaluate the output
use the output to enhance their learning experience

Reality Check: How AI Tools Are Actually Used

Some general UU survey results

1633 Students
348 Teaching staff
BA/MA balanced; few pre-master
34.4% of teaching staff report not to use GenAl for educational purposes vs. only 13.4% of students
68% of students use genAl for working with texts vs. 47% of staff
Majority of respondents use GenAI for Brainstorming, Text Writing, Clarifying Questions, and Summarizing & Explaining Texts
Teaching staff use it less for Research Design than students (40% vs. 11.9%)*
18.5% of teaching staff report that they use to create study content and 26.9% for demonstration purposes in class

Some interesting results

Both students and staff think that the use of GenAI tools for assignments is widespread among students use.
Lecturers seem to perceive students as not particularly knowledgeable about the use of GenAI
Students seem to perceive teachers as not particularly knowledgeable about the use of GenAI
Lecturers also think that their colleagues are not very knowledgeable about the us of GenAI.
Students don’t think it’s unethical to use GenAI in course work for support and assistance, or for generating a solution for even a small portion of an assignment. Staff views are a bit more varied, but not exact opposite.
There is a huge cry for literacy training in GenAl from both students and staff.
7.1% of teaching staff report that they use it for grading/assessment of work (no significant differences between faculties detected)

April 29 @ 4pm: Results will be presented in Ruppert 002 with drinks afterwards

Inputting student data into AI

Can AI Tools Be Used for Grading?

Sure! I have done that in past, but I did not call it AI

There are a couple of domains where AI tools can be very helpful in grading:

Automated Essay Scoring (AES)
Code evaluation and automated testing

The more structured language is, the easier it is for AI to optimize the language. This is very apparant in the field of programming, where AI can be used to generate all sorts of automated evaluations, code optimization, and unit tests.

How AES Works

Training: AES systems are trained on a dataset of essays that have been graded by human experts. The system learns to recognize the qualities and characteristics that correspond to various grades.
Feature Extraction: The AES algorithms analyze essays for a range of features, including grammar, syntax, vocabulary, sentence structure, and sometimes even the coherence and logic of the argument. The sophistication of feature analysis can vary widely among systems.
Scoring: Once an essay is submitted, the AES system applies the model it has learned to evaluate the unseen essay’s features. It then assigns a score based on its training on previously graded essays.
Feedback: Some AES systems can also provide feedback to students, identifying areas for improvement or strengths in their writing.

This is a completely different scenario from entering student work into generative AI tools to generate output for grading student work!

Cons of AI-Assisted Grading

Much like with AES, there are some limitations and potential drawbacks to using AI tools for grading:

Language is complex; can contemporary AI fully grasp the nuance that a human grader would pick up?
Bias in the training data can lead to biased grading outcomes.
AI tools may struggle with creativity, originality, or unconventional approaches that human graders can appreciate.
How does AI treat continuation errors in exams?
Would you feel satisfied if your work was evaluated by an AI reviewer? Or if an AI would process and respond to your assessment and development portfolio?
Isn’t student work protected by intelectual property and copyright?

As a statistician I can argue that using AI tools for grading opens the door to a class of potential errors that might go unnoticed.

Is It a Good Idea to Use AI in Grading?

Student rights in the AI-Era Classroom

Human contact is a right

Every student has the right to be educated and graded by a human being. That right also includes the right to know a priori to what extend AI is used

in creating the educational materials,
in providing feedback,
in grading student work.

Every student has the right to refuse interaction or any other involvement with an AI for their coursework.

Conclusion

Minimize the use of AI tools, and when you do use them, do so responsibly.

Generative AI can be a great companion in a knowledge discovery journey, but it is not a replacement for scientific rigor, active reading, critical thinking and human creativity.

Contemporary AI should always be viewed as a complement to human actions rather than a replacement.

Be as transparant about the use of AI tools in your work as you would require others to be. Also, make sure that you are not violating any rights or laws.

The only way to sustainable embedding of generative AI in academia is to ensure a sense of community and collective ownership of responsible AI practice.

This requires a shared responsibility that requires the active participation of all involved, whereby we hold ourselves and each other accountable for our actions.

Consortium USO at Utrecht University

Final thoughts

join the consortium
talk to each other
talk to me

Follow the growing collection of information in a more structured resource at www.gerkovink.com/ai