Another summer vacation has passed in the blink of an eye. As my daughter began her third year of middle school (ninth grade), she also AirDropped me a Colab program file and some data files.
Yes, she completed a project she had zero concept of before the summer began: an emotion recognition model based on machine learning.
As I ran her code line by line and checked the results, I couldn't suppress the excitement of her "greatly exceeding expectations." Before summer started, when she told me she had signed up for this competition, she was still asking me what machine learning was and how to do emotion recognition. My answer then was simple: a few years ago, this was at least the difficulty of an undergraduate thesis; I didn't expect her to finish it, but just hoped she'd find the answers to her questions over the summer.
Now, in less than a month of actual work, she delivered a model that produces results and, in turn, explained to me what Naive Bayes, Decision Trees, Random Forests, and Support Vector Machines are...
I think I can hire her as my research assistant now.
In fact, I barely participated in her project throughout the summer. Although I occasionally saw a look in her eyes hoping I would help her debug, I basically just pointed her to a function library or a tutorial keyword and let her look it up herself.
I know she got stuck on dataframe data alignment; I know her eyes got blurry while cleaning data; I know that to understand simple matrix calculations, she actually had the patience to go through the basics of linear algebra, even if she still only half-understands it; and I know she actually drew neural network structure diagrams herself, cramming on activation function input/output shapes and a mountain of model terminology...
There were also frequent "calls for help" to GPT-4 and Gemini 1.5: sometimes saved, sometimes misled...
I told her the result doesn't matter at all; what matters is the process of completing a full project and understanding those once-obscure terms.
She said that as a reward, I needed to "show off her results" on my WeChat official account because she has a bigger goal: to start an AI practice extracurricular course at school.
So, this was the initial motivation for writing the words above, and the source of inspiration for something I've been thinking about for a long time but suddenly seem to have found a glimmer of an answer for.
After countless discussions about the massive social changes AI will bring, what to do ourselves and how to educate our children has naturally become a more realistic and urgent question. As mentioned, it may take me a long time to find the final answer, but this past summer, I began to feel glad about some of the choices I made in the past:
- When she was seven or eight, I chose Scratch for her and helped her with a set of basic courses. So, when the school officially started that course a few years later, she knew she was ahead and tried her best to maintain that lead;
- When she started learning Python, I told her not to memorize any functions but to read the code in tutorials frequently; it's okay if you don't understand it now, you will naturally understand it later;
- While walking, we would discuss binary, encryption/decryption, greedy algorithms, backtracking, queues, dictionaries...
- I also told her that after the emergence of ChatGPT, rote memorization of knowledge is no longer important, and programming languages aren't either, but we must know how to think through problems because that determines how we talk to AI;
- I also said that the foundation of models is mathematics, and the end is logic;
- This summer, I showed her a YouTube channel called 3blue1brown, and now many of her classmates know about it too...
- I told her that models like Naive Bayes and Decision Trees are actually obsolete now, but the thinking behind them is much more beautiful than neural networks, and perhaps in the future, people will put "old wine in new bottles" again...
My generation is lucky; the knowledge we learned became outdated and obsolete slowly over the past decade or two. The generation currently growing up is unlucky because much of the knowledge in their textbooks is already a "historical relic" by the time they learn it.
How to help children grow freely among "knowledge relics" is something none of us has a definitive answer to; we can only take it one step at a time.
So, I wrote this article because I support and encourage my daughter's idea to start an AI extracurricular course. The answer, in the end, must be found by her. Being alone is ultimately too lonely; having more partners along the way will always be better.