Learning with an Artificial Intelligence

Introduction

One useful role of artificial intelligence or AI is in education. AI can act like a personal tutor that is available whenever you want (as long as you have internet). AI can help with English grammar, writing, math, physics, and even other subjects like software engineering.

In my software engineering class, which my college refer as ICS 314, I personally used an AI called ChatGPT to support my learning and project work. In fact, the class encouraged the use of AI as part of the learning process.

Personal experience with AI

There were many places in ICS 314 where I used AI, and many places where I could have used it but chose not to.

The experiences

We were introduced to many programming languages, techniques, and frameworks. We learned them through free online tutorials and by completing assignments which the class called “experiences.”

I did not use AI while reading tutorials or working on these early assignments, mainly because it was too early for me. I could do them myself, so AI would add very little benefit, and thus using AI here would mostly be a waste of time.

The in-class WODs

In class, we had timed quizzes called “workout of the day” or WODs. WODs focused on implementing a function or webpage using the tools we learned within a strict time limit.

This is the part of the class where I used AI the least. The instructions were often long, and sometimes in the form of images. I didn’t think I could quickly paste everything into an AI tool in a way that would help.

Since WODs are timed, every second matters. In timed quizzes and exams I tend to turn in my work at the very last moment, because I don’t want to waste time I could spend checking my work. Similarly, AI might give faster answers, but it also risks costing time and producing something not reliable at all. For that reason, doing WODs myself was usually the best option.

The in-class practice WODs

Practice WODs were similar: still timed quizzes, but more focused on training rather than evaluating our learning status. I still usually did them myself rather than using AI.

Honestly, I don’t have a logical reason for that. Maybe a habit or self-challenge of the strong one? But the trade-off was the same: AI could maybe provide quick answers, but it could also waste time.

The essays

The class also assigned writing tasks—essays about what we learned and what we thought about it—posted on our personal portfolio.

English is not my first language, and I didn’t learn it well when I was young, so AI was extremely helpful for improving grammar and tone for my essay. One prompt I gave ChatGPT was:

Okay, we finally finished Wonkes! Alright, indeed it is the due date of Wonkes has passed and we have to submit what we have.
Luckily it turns out our grades are nice: we got an A overall. Yay! The next thing is to wickedly utilize Wonkes to make us
more like a professional by writing essays about the effort we spent. Can you read the following essay and improve it, give
it a conclusion, and an interesting title? ## Why effort estimation? In the rougly one…

Yes, I often write long prompts, although large portion of them are confides. ChatGPT usually improved my essay a lot, although I still needed to revise the output manually because ChatGPT sometimes moves too quickly through ideas, making text more like a summary rather than explanation. The cost of using AI here felt like nothing, and the benefit was high.

The final project

The part of the course where I used AI the most was the final project, Wonkes, which I built with teammates and ChatGPT.

Even though it was a school project, I didn’t treat it as something to do in the simplest possible way (even if ChatGPT sometimes pushes that style). I planned a lot for Wonkes, which created a lot of work. My usual workflow was: I do the design and decisions, and I let ChatGPT help with the boring but necessary parts like making a page skeleton or formatting a form. For example, I used prompts like this to speed up sign out page development:

Now the Merch Detail page is finished. The next task is to redesign the Logout page. Once the user clicks logout, they will
be redirected to this page, where the user is being asked whether or not the user is sure to log out. There is a button that
user clicks to confirm logout, and a button to cancel. Please redesign the Logout page based on the following template:

'use client';

import { signOut } from 'next-auth/react';
import { Button, Col, Row } from 'react-bootstr…

When problems happened and I couldn’t find a solution, I asked ChatGPT to inspect code and suggest causes:

There is a problem that we just found on Wonkes: on dev site images can upload normally when creating a new merch, but on
production site only one image can upload successfully and more than one image will fail. The program still notify that the
adding/editing merch is successful, but the image did not actually get uploaded to the database (I found no database entry
created in the database). What might be the problem, please check the code. The code for add merch form is

'use client';
import { useSession } from 'next-auth/react';
import { Button, Card, Col, Container…

I also used ChatGPT for language and UI text improvement, like making search bar placeholders shorter:

I am having a few ideas of what to put as placeholder in seachbars for listings.

<Form>
  <Form.Group
    controlId="search"
    className="mb-3"
  >
    <Form.Control
      type="text"
      placeholder={wordBank[Math.floor(Math.random() * wordBank.length)]}
      value={search}
      onChange={(e) => setSearch(e.target.value)}
    />
  </Form.Group>
</Form>

Here are the placeholders that will be picked at random.

const wordBank = [
  '(„• ᴗ •„) Let\'s search something!',
  '(　-_･) ︻デ═一 ▸ What are you aiming for?',
  '(⸝⸝⸝O﹏ O⸝⸝⸝) What have we got today?',
  '(⸝⸝> ᴗ•⸝⸝) Good luck searching your favorite!',
  '<(￣︶￣)> Proud of what you want today?',
  'ヽ(・∀・)ﾉ Aloha, what can search bar find you today?',
  '(///▽///) Would you embarrass when you search for certain stuff',
  '(×﹏×) RIP',
  '(⊃｡•́‿•̀｡)⊃ Hug what you like.',
  '౦０o ｡ (‾́。‾́ )y~~ Wonkes\' policy banned cigar from sale.',
];

Can you shorten the phrase as much as possible without losing too much of its meaning? Hard words is okay.

In the final project, AI gave me a lot of value. The biggest benefit was time: I saved time on repetitive work and spent more time on design choices, structure, and debugging.

But there is also a cost: if I let AI do the boring parts, I get less practice doing them myself. That can matter if I need to do those tasks without AI, such as during an interview.

Learning a concept or tutorial

Most tutorials in the class were clear enough that I didn’t need AI’s help. However, one exception was connecting pgAdmin 4 on my computer to a local database. pgAdmin asked for many configuration values, and I didn’t know where to find them. I searched online, and an AI tool (I think it was called “AI Overview”) gave a clear step-by-step guide that helped me build the connection successfully.

There wasn’t much cost, and the benefit was huge: I no longer needed to query databases through the terminal with command lines.

Answering a question in class or on Discord

In ICS 314, I did not use AI to answer classmates’ or project teammates’ questions in class or in Discord.

One reason: sometimes I already knew the answer, so AI wasn’t needed. Another reason: many questions were highly course-specific, for example, “What coding software are we supposed to use to zip files to do EUT 1 and 2?” AI wouldn’t know the answer. The last reason is simple: I didn’t answer the question because someone else already did.

Asking or answering a smart question to a classmate

I didn’t use AI to answer classmates’ smart questions. The reason is also simple: I didn’t want to answer them, and I didn’t answer any of them, so I didn’t use AI.

Coding examples

I did not ask AI for small coding examples like “Give an example of how to generate a random number between 10000 to 99999.” If I’m going to ask an AI, I usually prefer asking directly for the code I need in my situation. I’m busy and I try to save time, so example-only prompts would be too time-consuming and therefore aren’t my style.

Explaining code

I generally prefer reading code myself, but if the task becomes time-consuming or confusing, I ask ChatGPT to explain it. For example:

Please explain what are @relation and the fields and references that come after it. Also, please tell me what does idexes
like @@index([AccountID]) do.

What I like most is time saved: instead of researching each keyword or reading long documentation, I can get an explanation quickly.

As for drawbacks, my honest answer right now is that I don’t feel one strongly. Nevertheless, the risk is reduce of ability to read unfamiliar code confidently without help.

Writing code

As mentioned earlier, I frequently let ChatGPT help with writing code for Wonkes, including editing existing code. The benefit is clear: speed. The drawback is also realistic: sometimes ChatGPT’s output doesn’t match the needs exactly, so I still need manual review and edits. But it is faster than writing all code from scratch anyway.

One prompt I used was:

Right now I am to write the code for the right side of the merchandise detail page. I will need a merch object for testing.
The data model for a merchandise in prisma.schema is

model Merch {
  MerchID Int @id @unique @default(autoincrement())

…please create an object name testMerch based on these information.

Documenting code

Honestly, the most documentation I did in ICS 314 was writing comments for parts of the final project. For example:

/**
 * A simplification of the MerchImage data model in the database. Contains basic information
 * needed to identify and display an image.
 */
export type MerchImage = {
  id: number;
  mimeType: string;
  base64: string;
  url?: string;
};

Since this wasn’t very hard, I didn’t use AI much for documentation.

Quality assurance

While working on Wonkes, I ran into many bugs: forms not submitting, missing user feedback, or the wrong UI response.

The debug problem I remember most was the modified sign-in system. The template originally used email address and password as credential. We changed it to use username and password, and then sign in started failing. I inserted print statements to locate where execution stopped and asked ChatGPT what could cause it. For example:

I added some checkpoints inside the authorize function:

async authorize(credentials) {
  console.log('F');
  if (!credentials?.username || !credentials.password) {
    console.log('MISSING CREDENTIAL');
    return null;
  }

…I read the terminal after a submission, and BEGIN QUERY is printed onto the terminal, and then followed by POST
/api/auth/callback/credentials 401 in 178ms. Nothing else printed onto the terminal after that. Did the query failed or
never complete, causing the sign in to fail?

Later, I found an error message about database connection failure:

When I manually insert the password, I inserted changeme as is, not hashed. After catching the error thrown during the
query, the following is printed onto the terminal: BEGIN QUERY E QUERY FAILED: PrismaClientInitializationError: Invalid
prisma.account.findUnique() invocation: Can't reach database server at ***:5432 Please make sure your database server
is running at ***:5432. at $n.handleRequestError

Eventually, the root cause was surprisingly small: an extra semicolon (;) in an environment variable caused the system to misread a value. Probably I wrote too much PHP and TypeScript so I mistakenly wrote a trailing semicolon without noticing.

After adjusting the environment variables, the database is connected. After bypassing the hash thingie, using
credentials.password === account.Password instead of compare(), I logged in successfully! Now let's work around the
hash related stuff.

Without ChatGPT, this debugging process might have taken days. With ChatGPT, it shrank to about two hours. The cost: fewer days of hard-earned debugging experience. The benefit: I could quickly continue working on Wonkes.

Other uses

ChatGPT can draw stuffs. I asked it to design an icon for Wonkes, and it turned out so good that it became our logo.

I also used ChatGPT to create other visuals, such as a background image for the 404 not found page and the coming-soon page. It saved me time I would have spent designing, drawing, or searching for images.

AI’s impact on learning and understanding

Overall, AI had a strong influence on my learning experience. Since I could consult tools like ChatGPT whenever I was stuck, and get clear explanations, AI improved my understanding of the course materials and software engineering concepts. I gained better comprehension of code and better approaches to solving coding problems.

Practical applications of AI

I have also seen AI used in real-world work. In the organization where I worked, a colleague used AI to parse hundreds of manuscripts and convert their information into XML files. That saved him a huge amount of time, and it reinforced my belief that AI is extremely effective at reducing repetitive labor and saving time.

Challenges and opportunities of using AI

A common concern about AI is whether it makes people lose the ability to solve problems themselves. For people who don’t want to think, AI can become a tool for getting answers without thinking or learning.

For example, it is not learning when one ask AI to write code for a task, then repeatedly testing and requesting for modification on the code until it works, because the person never know what didn’t work and what made the final solution work.

AI is like a double-edged sword. AI can be a tutor that supports learning all day, but it can also act like a digital slave that hands out solutions for free. I can ask ChatGPT to explain code, or I could ask it to write me entire Wonkes project. Whether AI helps learning depends on the user’s willingness to learn. I’m proud to say I’m interested in learning, so AI functions more like a tutor for me than a cheater’s treasure.

AI has real opportunity in software engineering education, but students should stay aware of the temptation to stop thinking.

Comparative analysis

The advantage of an AI compared to a teacher in school in tutoring a student is time: an AI can be consulted at nearly anytime, while a student can only consult a teacher during class and office hours. Sometime remote communications like email and Discord channel are available, but still require waiting for response. AI typically answer immediately, which makes learning more engaged. Another reason is that AI can make learning more engaged is that AI is not a human, thus a student may ask questions that they may chose to not ask a teacher because they are shy. I am an example, I ask minimal number of questions in class but talk a lot to ChatGPT.

Regarding knowledge retention or how well a student may remember knowledge, the traditional teaching method may win. If you ask an AI to explain something, it typically give detailed response but without a way to remember it. However, a teacher can gain experience with teaching and develop ways that help students remember things. For example, my physics teacher create interesting words like PIV to help us remember equations (PIV is for equation P = IV, meaning power is equal to current times voltage).

Regarding practical skill development, teachers still play one important role that AI cannot fully replace, which is telling student what to learn. For example, in ICS 314 we are told to learn JavaScript, then TypeScript, then HTML and CSS, and lastly frameworks like Bootstrap and React. This is a nice plan for gradually learning stuff required for developing a web-based final project at the end of the semester. AIs can provide plans, too, but requires one to ask for one. Therefore the traditional teaching method serves as an important initiative for students to start develop practical skills that they need. Both teachers and AIs can help on furthering educations of the practical skill development.

Future considerations

AI will likely remain part of software engineering education because it can support students anytime. The main challenge is preventing over-reliance-students might use AI for answers instead of learning.

The future should focus on teaching students how to use AI responsibly: use it as a tool for learning, not a replacement for thinking.

Conclusion

In conclusion, AI is a powerful tool for saving time and a convenient tutor for learning, especially in software engineering courses like ICS 314. I recommend AI as a support for learning for future students, but with one warning: if AI is used as a shortcut to avoid thinking, it stops being educational and becomes a danger.

Note: ChatGPT was used to help illustrating the image above and editing this essay. AI prompts above was reformatted for better reading experience, but still nearly the same as the original text.