Introduction

Before I answer that question, I’d like to speak a bit about what qualities help me confirm that I’ve learned a given subject. For that purpose, I want to contrast two modes of learning, that I’m calling profound learning and superficial learning.

I consider profound learning to involve a deep(ish) understanding of the subject under study. When you have learned something profoundly, it belongs to you. You don’t have to be an expert in the subject, but you have done more than memorized some facts. You have an understanding of how the facts interact with each other, meaning you can extend the knowledge into a slightly different context and apply it to new situations.

On the other hand, superficial learning might provide some of the right answers, but doesn’t give you either a depth or breadth of understanding of the subject. It might allow you to respond correctly on a test, but won’t enable you to adapt and use the knowledge in a slightly different context.

Can you guess which kind of learning I’m going for when I embark on studying a new concept? Profound learning for sure.

However, how will I know that I’m doing profound learning, and not superficial? I’ve noticed four qualities that stand out when I engage in profound learning. If I notice these qualities arising when I am learning something new, I can be pretty sure I’m engaging in the profound kind of learning. When I do that, I’m…

1. not taking the easy route

2. seeking out and weighing all available information

3. slowing down

4. in the end strengthening my confidence

So if we want to know whether AI is useful in learning, we’ll want to see whether it benefits profound learning. We’ll see in the course of understanding these four qualities, whether in my experience using AI has done that.

A quick disclaimer, because Internet:

This article is based on my personal observations about what has or has not worked for me in my learning process, followed by my guesses about why that might be—it’s not meant to be a scientific statement of fact or a declaration about what is the best mode of learning for each individual to pursue. If my experience resonates with you, I hope you will find these words uplifting and benefit from my observations, but if it doesn’t that’s fine too.

Without further ado, I’ve observed the following behaviors when I engage in what I’m calling profound learning.

I don’t take the easy route

Have you ever asked a teacher, parent, team lead a question and gotten dragged into the trap of figuring the answer out for yourself instead of the immediate satisfaction of the pristine reply you were hoping for?

Why does a good teacher strive to make you think for yourself like that? What are they aiming for you to achieve? Rather than giving you a ready-made answer, they enable you to answer your own questions. But what’s so bad about getting a ready-made answer? It removes the work of thinking from you and renders you a passive listener. These sorts of teachers understand that they don’t want to rob you of the thing that makes you learn—doing the work.

A teacher can demonstrate how to do the thing you are learning, but if you’re really going to become proficient in it it’s on you to do the actual work, to get your hands dirty. I think this need is more obvious in disciplines that involve the body. Things where you’re mastering motor skills or developing specific muscles, like sports, dancing, or playing an instrument. You can’t offload this responsibility to another person. If you want biceps, you need to be the one lifting the weights. This being the case with our bodies, I wonder why we think there won’t be consequences to offloading the responsibility of using our brains to someone else?

Over the past year, I’ve been encouraged to incorporate AI code writing and learning solutions into my workflow. Resistant at first, I eventually gave AI tools an honest try. I have spent close to a year relying significantly on AI to assist in writing code, understand code, and explain unfamiliar concepts on a daily basis. My team leads have supported me in this, and have challenged me and helped me learn to use AI tools more efficiently.

After all that, what is my opinion about using AI in my workflow and learning endeavors? My main concern with consistently using AI as a replacement for the work of discovery is that I simply do not learn the topic as well—an easier process seems to correlate to less depth of understanding. This isn’t to say that I ought to make my learning process needlessly difficult, just that—as a general rule—the more effort I put into understanding a concept, the better and longer it sticks with me. When I immediately get my answer from AI in a neatly wrapped package, I read it, I comprehend it on a surface level (“that makes sense!”)...and I quickly forget it.

In work this year, I have persistently felt that I am not developing my knowledge base or skills as much as I would like. I have noticed that I learn best in a hands-on way. The difficulty with asking AI to answer my questions and letting it do the coding work for me is that I lose the opportunity to code it myself, which is the best way that I’ve found to retain domain-specific knowledge.

When I use AI to learn new things, I am taking the easy route. I’m not doing the work I need to develop profound understanding. In general, I consider using AI in my learning process a part of superficial learning—it provides a cursory overview of a topic with some quick answers, but it’s not sufficient for learning the subject meaningfully. It doesn’t help me make the information my own.

I seek alllllll the information

I lately enjoyed reading The AI Con by Emily Bender and Alex Hanna. In it they contrast using answers from a chat bot such as ChatGPT to obtaining the information contained in those answers from a search engine, such as Google.

They point out “friction in information access is actually not only beneficial, but critically important” in the imagined case of looking up a health-related search in a search engine such as Google (Bender/Hanna, p. 171). The searcher gets back a collection of results that might include a page from a medical center, a WebMD article, a Dr. Oz article, and an online forum where people with similar health concerns interact. You will not interpret the information from each of these sources in the same way. Some sources you may trust more, and some less, and the information about the source is what allows you to develop this trust.

They go on to consider the case of asking the same health-related question of a chat bot. It may come back with similar information in its answer, as it draws on all the above sources to fulfill the searcher’s query. However, the searcher is not seeing the information in the original source context as in the first case. They are thus unable to use this context to build trust, or lack thereof, in the sources that the information was derived from.

In other words, you’re actually gaining more information from the search engine than you are from the chat bot. That’s because the extra details surrounding the actual answers to your question, such as the source of the answer (WebMD vs online forum), provide meta information that helps you make an informed decision about how much weight you should give to the answer—how much should you trust the source?

One thing I felt AI was useful for was helping me to develop learning plans and strategies. After considering the above I feel less inclined to use AI to develop a learning plan for me, for the reasons stated—I don’t know where the information is coming from, or if it has been tried and tested. When you are learning a new concept or area of study, you are particularly vulnerable to misinformation and bad advice. I would prefer to know the resources that I am trusting my learning experience to, whether they come from a knowledgeable source, and whether or not they have been used to success by others before me.

Beyond that I’d like to suggest that if you discover answers to your questions through a search engine, rather than a chat bot, you are likely to retain the information better. Why? Because you’ve made more connections (meta info). You’ve spent longer thinking about it by having to find the information manually.

I also think it’s interesting to notice that the process of using a search engine involves certain steps within which learning thrives. You engage in a (most often repeated) process where you:

1. search

2. discover an answer

3. consider the source

4. integrate the new information with your current understanding, if necessary by refining that understanding based on new information

Using a chat bot does not replicate this entire process, but merely replaces the first and second steps—question and answer. Using the chat bot creates an appearance of learning the answer that you were searching for. However, using the chat bot is no guarantee that you have actually learned or will retain any information contained in the answer.

I slow down

AI is touted as being faster than humans, which is one of the reasons using it in software development is emphasized as a benefit—if not an outright need! I’d like to suggest that, in the realm of learning, and most likely coding (which is a skill highly dependent on continual learning), this may actually be AI’s weakness.

When you force yourself to slow down you tend to strengthen the concepts in your mind. What do you do when you don’t understand a paragraph in a book? You go back and re-read it slowly. Musicians, what do you do when you are learning a new song? You play through it slowly. In the gym, when you are learning a new exercise, what do you do? Your trainer probably demonstrates it slowly, and encourages you to do the exercise slowly until you have attained the proper form.

To use an example from the world of software development, lets talk about rubber ducking. i.e. Detailing your code logic from start to finish as though you were explaining to a rubber duck and thereby hopefully noticing the bugs in your logic. What is the point of rubber ducking? It’s not to explain your code to a rubber duck, or a fellow developer as the case may be, so that they will understand—it’s to force yourself to slow down enough so that you will understand what your code is actually doing, not just what you think it is doing.

It seems humans have a need to slow down to notice the details that are essential for profound understanding. I’d suggest that AI encourages us to move too quickly—by receiving those details instantaneously, we tend to gloss over them. When was the last time you thought you understood a problem, only to discover down the coding rabbit hole that you had a fundamental misunderstanding from the beginning? My concern is that in using AI, we are encouraging such misunderstandings to linger, because AI usage divorces us from the process of discovering our mistakes. And what’s a simple way to notice mistakes—slowing the f*** down!

I strengthen my confidence

Throughout my life I’ve experienced a strong sense of satisfaction as I reflect back on my achievements that I felt I earned through hard work. Ownership of skills leads to confidence in skills. Experience helps me to build confidence in my skills and knowledge base. In my workplace, I believe my efforts are not just about what product I am able to spit out in a short amount of time—they’re about what new details and deeper understanding I gain along the way, making me a stronger developer. After all, I’m an entire human, not a code-making robot. What matters about my work is as much what I learn along the way as it is the code I produce.

I encourage myself, and I encourage you, to treat ourselves with the care and consideration deserving of human beings: we must take care to nurture our knowledge, develop our skills, and give ourselves time to learn things from scratch.

Conclusion

I’m still in the process of evaluating and adjusting my behavior with respect to AI, but over the course of about a year of using it seriously, I am less inclined to involve it in my learning process than I was when I started. I haven’t noticed significant gains in my domain knowledge or skills in my job since incorporating AI into my coding workflow—although I have produced some pretty fancy, definitely usable, and even kind of elegant code (that I nevertheless don’t understand fully!). In the long-term, I have not benefited from any knowledge that I “learned” from AI. My retention was brief because I didn’t solidify the learning with hands-on practice.

Using AI in my learning makes things too easy for me, obscures available information, speeds me up so that I miss key details, and in the end has severely weakened my confidence—for even with what facts I might retain I don’t understand the surrounding context. And in far more cases I forget the answers to my questions almost as soon as I read the chat bot’s response.

So to answer the question posed by this article's title—How do I use AI in my learning? Well, actually, I’d rather not.

Notes/Sources

*Artificial Intelligence is not really intelligent. It doesn’t “know” things in the sense that you or I know them (by drawing on some internal map of knowledge). The AI used in chat bots for instance is a speedy and specialized prediction engine, that spits out what a likely next word in a sentence should be. I am using the term AI throughout this article for ease of communication, since it is how people tend to refer to this technology.

• The AI Con book: https://thecon.ai/