AI in a CS Classroom – An Interview with Dr. Mohan

AI has captured everyone’s attention at IMSA. For optimists, AI signifies a new era in education, on par with the introduction of calculators in math classes or computers in design classes. IMSA is leading this initiative with the establishment of the IMSA AI Center and exploring AI applications in various classrooms, in partnership with Flint. To understand the practical applications and limitations, The Acronym interviewed Dr. Ashwin Mohan, a Computer Science faculty member, to share his views on the subject. Not only has he integrated Flint into his CS classes, but he also possesses deep knowledge on the topic and offers valuable perspectives for students who are ready to embrace this change.

The Acronym: I want to focus on how AI can be used in classrooms, specifically in computer science classrooms, because that’s your area of expertise. We have only experienced Flint once, but it seems like it’s a tool for generating personalized assignments. Is this understanding correct?

Dr. Mohan: Let’s unwrap the question. So the first part is, how can I use it in the classroom?

 AI can aid in learning, as a teaching assistant, almost like a tutor. There are two possibilities in the classroom: it can help students with finding answers, and it can also help teachers generate newer questions. You could create a module such as FAQs on a topic. Students may be able to directly interact with the AI system to find answers, rather than search on the internet. So it could be much more focused and it helps you save time. Importantly, this is a learning model- based on the questions asked from different cohorts of students over time, so in the classroom, it will serve as a digital repository of different questions students have asked over the years that hopefully covers the entire gamut of learning that is required. Moreover, this tool can be customized for assignments, a specific lesson, or a student’s learning ability. Another application could be where people use it to generate questions like a question bank, i.e. it starts helping the teacher curate the set of questions (e.g. a question bank prior to the Quiz). Overall, there is definitely ample scope for AI applications in the classroom for educators and students to have a positive experience 

To explore how Flint AI could be applicable in the CSI classroom, sophomore students Helen Shao and Andrew Bae undertook a unique project. All CSI students experienced building web pages and a website using HTML and CSS. Students were tasked with designing wireframes, submitting proposals for their website before the start of this project. These two students were tasked with building their own website, and additionally, use FlintAI to explore its capabilities to design and deliver a website based on wireframes and textual prompts. They could then compare and contrast their outcomes vs. those generated by an AI engine. Further, they explored the number of specific prompts required by Flint AI engine to reproduce what they had created based on content knowledge from the classroom. So here is another example of how in a CS classroom, students can first hand experience the benefits and challenges of dealing with an AI engine and how it compares to learning in a student-centric classroom curated by the teacher.

The Acronym: All of those are great examples. Could you also share applications in other areas such as moderating tests, evaluating assignments, etc?

Dr. Mohan: In those aspects, I would actually categorize AI tools as a hindrance. That’s because one of the most critical aspects of assignments is a teacher’s ability to provide feedback and guidance that’s specific to the student. When teachers interact with a student, they create a sense of rapport, understanding where they are, along with their understanding of the context and can convey the intent of their questions. An AI system cannot do all those things—it can evaluate a response based on a rubric that the teacher could create, but it cannot provide critical feedback or  an understanding of why a specific approach was taken or help establish rapport with the student. AI is presently used as an aid to detect similarity in code, check for plagiarism but is limited in scope.  Teachers are always cognizant that every student is different, their approach to learning is different, and so interactions are unique, especially to understand student perspectives. An AI evaluation engine literally removes a layer of humanity.

The Acronym: Given that AI is facing issues such as hallucinations, a lack of citations, and providing false information in its responses, is AI a good source to teach students theoretical concepts? Should we wait for the technology to evolve before it can even reach a point of viability?

Dr. Mohan: I would agree with that. First, it would be incredibly expensive to create AI code customized for every student. Then, AI will remove the collaboration from learning. If you’re stuck on something while working, you would go to an AI system for answers instead of going to a table partner, colleague or a friend. Those things begin to affect motivation and importantly impact honing soft skills such as communication, team-building, and self-confidence. For example, if you’re trying to resolve a query, AI is a very quick way to get the solution, but it won’t expose you to the many different ways you can approach and  learn the same concept, something you can do when you collaborate with, say, your table partners. In that sense, it’s like a GPS system—AI systems will find you the shortest or quickest route, but they won’t teach you how to navigate through traffic or become a better driver. An AI system provides the information, but it doesn’t create knowledge you need at any instant to learn. Moreover, as you identified, we know from several sources now that some of the information AI provides is somewhat dubious. We need to understand that we could be trying something that is just past its infancy and claiming it’s ready for primetime. AI is another tool in our toolchest to solve the problems of the work, to assume it is the only way forward is short-sighted as it is prone to biases, errors and inherently dependent on the underlying algorithms and what it was designed based off (training data).

The Acronym: Your answer brought up a really interesting point- AI is saving time, but not providing any learning to the students. What about learning and skills that would become less important with the advent of AI? What skills in your view would become less important as time goes on?  

Dr. Mohan: The evolution we are seeing is not necessarily driven by AI. First, we have become a data-driven society. Before the AI hype, it was all about big data (large volumes and varieties of data created), which drove the need for data mining, machine learning  and ultimately for AI to get benefits/learning out of the data. AI and LLMs offer significant potential, but it’s not clear what’s next or what tools will be replaced because of AI. Moreover, computational thinking has nothing to do with AI or programming languages. When I was your age, there were languages such as Fortran, COBOL, and Pascal, all of which have gone away. And that has nothing to do with AI—it was about the problems we wanted to solve, the technology needed to support it and that implied software needing to evolve in accordance with hardware.

For example, Python was created as a portable, easier, and convenient option for people who emphasized code readability and didn’t want to get into C++ or Java. Therefore, the new language was introduced to meet the needs of a section of society. As with everything, the development of these tools is more driven by things like evolution in computational thinking and technological needs of the times, and not necessarily because of AI.

That said, AI provides significant opportunities—it brings capabilities to a wide array of applications. It allows someone to not only analyze structured data but also unstructured information, emails, social media, images, tweets and verbal languages—the scope is endless. So when you ask, what will AI replace, I think it is a tool that has a lot of potential. It can be revolutionary in its application in how much time is saved, how quickly we are able to analyze data, generate new questions, learn and implement or discard ideas. This is a process that requires careful and ethical consideration at every step of the way. Are solutions that AI generates meaningful, can they be validated, are they entrenched in bias, who is the beneficiary etc. Only when one can fully ascertain answers to deeper societal concerns, can we fully acknowledge the true nature of jobs it will replace. Until then it serves as an evolutionary tool to explore, experiment and engage with.  

The Acronym: What ethical considerations should students keep in mind when using AI? For example, using AI for assignments or understanding a topic using AI? 

Dr. Mohan: There are several ethical considerations regarding AI. As I mentioned, AI suffers from biases, partly because it’s not entirely open-source. It is largely controlled by major corporations like Meta and Microsoft, who have significant influence over these models. Moreover, those who provide data to these AI models and those who use them typically have access to computers and the internet, but they don’t necessarily represent the broader spectrum of society, which introduces inherent bias into the outcomes. 

From a student’s perspective, AI-driven learning doesn’t properly acknowledge sources and isn’t conducive to providing citations; you can’t just cite AI as a source. Validation of information is another area that currently remains unchallenged. For commercially available products, I have little to no way of knowing the underlying algorithms or the datasets that were used to train LLMs because they are not published, not subject to peer-review and so remain a black box in that sense.  Lastly, as the students in my CSI class discovered, AI doesn’t always produce consistent output for the same prompt, thus failing the gold standard of repeatability, reproducibility and reliability, a hallmark of good scientific exploration. 

The Acronym; What skills would a student who is trying to learn or use AI must have to ensure they are effective?

Dr. Mohan: The key skill is computational thinking. Most students learn subjects like reading, math, and science from elementary school, but computer science is often introduced in the 8th grade or high school. As a result, only a few students are exposed to the world of computer science early on.

AI looks like it is here to stay for the foreseeable future, thus, it is imperative for students, irrespective of their fields of interest and career pursuits, to appreciate computational thinking, that is, have the ability to break down complex problems, observe patterns in them, understand different layers of data and focus on filtering out irrelevant information and identifying only that which is most important, and be able to logically approach the problem while distinguishing between good and bad data. This progression is similar to how algebra, pre-calculus, and calculus lay the groundwork for courses like linear algebra in mathematics. A foundational understanding that is based in computational thinking is critical to be successful as independent thinkers and doers in the AI world.