ai-camp
Published in

ai-camp

AI On Thumbs: Using the Socratic Method to Teach Machine Learning

In its simplest form, the mission statement of this project is to interest students in our summer and after-school programs by teaching them AI concepts on their phones. From the previous iteration of AI On Thumbs, we’ve established our target audience as teenagers interested in learning about AI, assuming no technical experience. Any lesson that cannot be understood without previous technical experience, assuming it isn’t covered in another lesson, is a failure on our part. As Einstein said to the French physicist de Broglie, “if you can’t explain it simply, you don’t understand it well enough” (Clark, 1972).

Einstein catching student illegally recording lecture (circa 1935)

One of the core philosophies of AI On Thumbs is its mobile platform; it must be completed on your phone or tablet. No desktops. No exceptions. In addition, lessons must be bite-sized. We’re not creating problem sets or writing a textbook. Lessons need to be something you can complete in an Uber or during a lunch break. This means each screen must have a maximum of 140 letters, one button, and one picture. If you can get the idea across with fewer words, then more power to you.

AI On Thumbs presents a unique opportunity to AI Camp. As of now, when it comes to short, online STEM courses, Brilliant comes to mind. However, Brilliant has two weaknesses we can use to drive attention to AI Camp. For one, have you seen Brilliant’s content? Thorough? Yes. Well-explained? Yes. Interactive? Not so much. Knuth’s magnum opus The Art of Computer Programming is thorough and well-explained, so why not just crack open a book? As Bill Gates said, “you should definitely send me a resume if you can read the whole thing” (Weinberger, 2016). All of your problems are solved with one decision. Teenagers want interaction. They want to feel heard. YouTube and Spotify’s recommendation algorithms can read their minds, so why hasn’t education caught up? To engage users, the educator must do it by interacting and conversing with the student. Question them to reconsider their past notions. After all, in The Meno, Socrates taught a servant the solution to a geometry problem using this method. Throwing albeit well-written information at them and asking comprehension questions doesn’t cut it.

Socrates and The Slave Boy Experiment (2022, colorized)

In addition, Brilliant’s machine learning content is incredibly limited without a premium subscription, which is $10 a month. What else is at most $10 a month? A Spotify subscription. A Netflix subscription. A Hulu subscription. If a teenager could choose any of the four, can you guess what they’re least likely to pick? It’s like asking a teenager to pick between cookies and turnips. If you’re wanting to learn AI, you don’t need a $10 Brilliant subscription. You can find a good textbook for free. However, teenagers are bombarded with textbooks♩eight days a week♩and groan at the thought of more homework.

Self-Portrait (circa 2018, 8:25 AM)

Our opportunity is the antithesis of Brilliant’s problem. We aren’t offering homework. We are offering the Socrates experience and Brilliant is Euthyphro. The goal of an AI On Thumbs lesson is straightforward: start broad and end with domain knowledge. For example, our two-part Facial Recognition course starts with an idea as simple as pixels and ends with explaining how computers use context to detect faces. We want to start with a question, move to the building blocks of the answer, and ultimately expand the user’s knowledge of a simple concept to an AI algorithm. That’s the vision of AI On Thumbs. It’s a preview of the teaching style students can expect with our courses. As instructors and team members, we are here to guide them to ask the right questions, but ultimately they are their teachers — and that’s priceless. Better yet, with our project, we are offering it at no cost to the user.

Who Are We After?

59% of teenagers don’t have a debit card. (Wakefield Research, 2019)

In recent years, fewer than 20% of U.S. teenagers report reading a book, magazine, or newspaper daily for pleasure. (Twenge et al., 2018)

Around 95% of teenagers have access to a smartphone (Schaeffer, 2019)

Our research on our target market, which in a broad sense is teenagers, gives us a fighting chance. A majority of teenagers don’t own a debit card, so they couldn’t purchase Brilliant premium even if they wanted to. A majority of teenagers don’t read books or magazines daily, eliminating competition with other mediums of information. Last but not least, the vast majority of teenagers have access to a smartphone, giving them a perfect opportunity to learn AI with their thumbs.

The odds are on our side. Teenagers want a modern way to learn but aren’t able to pay subscription fees to do so. If we offer a better product than our competition, at no cost to the user other than the publicity it brings to AI Camp, this presents a huge opportunity for us. In addition, imagine the satisfaction it would bring to a parent to hear that their child is using their phone for something other than social media. What parent wouldn’t approve of their child getting a head start in a market projected to grow at a CAGR of 38.1 percent? (Precedence Research, 2022)

Success Criteria

Our criteria for success is retention. If we measure our success by our number of downloads, big deal. There are YouTube videos with hundreds of thousands of subscribers that explain AI models. Our edge is our ability to retain users. Ultimately, the success of our project is reached with statistically significant evidence elucidating that our approach to teaching AI has a higher retention rate than other platforms such as Brilliant.

Road To Success

What we’re after is statistically significant evidence that our approach to teaching has a higher retention rate than Brilliant and related platforms. For a valid comparison, we need to establish a standard method of computing retention rates. Considering lessons in AI On Thumbs and Brilliant are structured similarly, we will define retention rate as the decimal value of the ratio of screens completed before interest is lost to the total number of screens. In addition, another control variable would need to be the topic of the lesson: the K-Nearest Neighbors classification algorithm.

“The First Day of School”

With enough samples, we can use standard inferential statistical methods to make an inference about the population of our target market, which if it’s the desired outcome — that our application has a higher retention rate than other platforms — could turn into a selling point that we use for marketing. To establish our North Star metric, it’s our mean retention rate. Our success criterion is whether there is significant evidence to support that the mean retention rate of our product is higher than similar products by Brilliant and other related platforms.

Milestones Along The Way

There are several important milestones in this project. Firstly, we need a script — a dialogue — to simulate a conversation between the teacher and the user. This is inspired by The Meno and will allow us to follow Socrates’ style of teaching. After all, what better way to make it a conversation between the teacher and the user than to adapt the screens from an actual conversation? We plan to complete this step a week before July.

After that comes development. Luckily, we aren’t starting from scratch. Our talented team of student developers has granted us access to the GitHub repository maintaining the existing code. Without a background in React Native, this is a lifesaver for us. We will plan to have a prototype of our lesson done as soon as possible and recruit users to constitute our sample.

In the next stage comes testing the users. As data scientists, this is the fun part. Our control group will go through a Brilliant course on K-Nearest Neighbors and our experimental group will go through an AI On Thumbs course on the same topic. We’ll ask the users two questions: where did you lose interest and what feedback do you have? The first question will allow us to measure whether we met our success criteria and in the second question, for the experimental group, we will consider the feedback in another iteration to improve the course.

Avoiding Failure

Our product is not a textbook. Turning our product into a textbook would be a failure on our part. We couldn’t compete. There are so many quality resources that cover machine learning. Our domain knowledge of the field probably came from one. The best example to describe our vision is again, the conversation between Socrates and the servant.

Socrates: And this space is of how many feet?

Boy: Of eight feet.

Socrates: And from what line do you get this figure?

Boy: From this. (points at figure)

Socrates: That is, from the line which extends from corner to corner of the figure of four feet?

Boy: Yes.

Socrates: And that is the line which the learned call the diagonal. And if this is the proper name, then you, Meno’s slave, are prepared to affirm that the double space is the square of the diagonal.

Boy: Certainly, Socrates.

Meno: Yes, they were all his own.

The flow of our course follows the same path Socrates used to educate the servant. Teach the user a machine learning algorithm, but make them feel as if it’s their idea. On the contrary, consider this example taken from Brilliant.

Many datasets have an approximately linear relationship between variables. In these cases, we can predict one variable using a known value for another using a best-fit line that follows the trends in the data as closely as possible.

Here, x is called the predictor variable because it will be used to predict y, while y is often called the response variable.

This technique is known as linear regression, and although it is one of the simplest machine learning techniques, it is often surprisingly powerful.

There are a few differences between this approach and Socrates’ method. One approach treats the brain as a hard drive — encoding information and making use of the hippocampus. Another approach treats the brain as a compression tree. We are given a piece of information in the form of a question and we must use our creativity to decompress the information into its original form, making use of the frontal cortex. Do we want the user to use their hippocampus or their frontal cortex as they complete the lesson? Studies show that “learning increases when students generate their own contexts for meaning” (Slamecka & Graf, 1978).

Acknowledgements

We’d like to acknowledge the original developers of AI On Thumbs: Mitch Cutts, Zac Brammer, Alexandra Fry, Jackson Choyce, Axel Mora, Alexander Zhou, Rohan Joshi, Sanjay Manoj, and Advay Aravind. The road to success starts with a solid base, and that is exactly what our team of developers provided for us with their foundational work.

In addition, we’d like to acknowledge our founder and CEO, @Michael Ke Zhang, for his vision on what AI On Thumbs should and should not be. His words of guidance, encouraging us to make it a conversation between the user and the content, inspired the Socratic style that we are attempting to capture in this second release.

If you haven’t already, check out our website. AI Camp’s Summer Program pushes teenagers to go from zero to hero with their data science and machine learning skillset. In addition, for high-performing students, AI Camp offers paid internships through our Team Tomorrow program. The original build of AI On Thumbs was developed by Team Tomorrow developers. To view projects created by students at AI Camp, click here. Currently, as of June 2022, you must be 13–18 to be eligible for our summer program. Scholarships are available and we select the best students to train for tech internships at top technology companies.

References

Clark, R.W. (1972). Einstein: His Life and Times. William Morrow and Company.

Introduction To Linear Regression. Brilliant. https://brilliant.org/practice/linear-regression-introduction/

Plato, & Grube, G.M.A. (1976). Plato’s meno. Hackett Pub.

Precedence Research. (2022). Artificial Intelligence Market Size to Surpass Around US$ 1,597.1 Bn By 2030. Globe News Wire. https://www.globenewswire.com/news-release/2022/04/19/2424179/0/en/Artificial-Intelligence-Market-Size-to-Surpass-Around-US-1-597-1-Bn-By-2030.html

Schaeffer, K. (2019). Most U.S. teens who use cellphones do it to pass time, connect with others, learn new things. Pew Research Center. https://www.pewresearch.org/fact-tank/2019/08/23/most-u-s-teens-who-use-cellphones-do-it-to-pass-time-connect-with-others-learn-new-things/

Slamecka, N.J., & Graf, P. (1978). The generation effect: Delineation of a phenomenon. Journal of Experimental Psychology: Human Learning and Memory, 4(6), 592–604. https://doi.org/10.1037/0278-7393.4.6.592

Twenge, J.M., Martin, G.N., & Spitzberg, B.H. (2019). Trends in U.S. adolescents’ media use, 1976–2016: The rise of digital media, the decline of TV, and the (near) demise of print. Psychology of Popular Media Culture, 8, 329–345.

Wakefield Research. (2022). The Junior Achievement Fintech Survey. Junior Achievement.

Weinberger, M. (2016, April 26). Bill Gates once said ‘definitely send me a resume’ if you finish this fiendishly difficult book. Business Insider. https://www.businessinsider.com/bill-gates-loves-donald-knuth-the-art-of-computer-programming-2016-4

--

--

Get the Medium app

A button that says 'Download on the App Store', and if clicked it will lead you to the iOS App store
A button that says 'Get it on, Google Play', and if clicked it will lead you to the Google Play store