Peterson Fellow Chronicles #2: NLP-42-Towards Meta-Rules for Life

The Original 42 Rules For Life Quora Post

In this article, I'm going to take a good look at the 42 rules that Dr. Peterson first posted on quora. This is the genesis of his recent book, 12 Rules For Life, as well as his new book, 12 More Rules For Life (https://www.youtube.com/watch?v=RRLt_cVueTw). You can read the 42 rules at the link below, or, even better, just listen to the song about it...

Akira the Don & MeaningWave

If you're wondering what was going on with that 42 Rules for Life song, I'd encourage you to take a listen the artist who created it himself, Akira the Don, who made a song explaining how he was making songs like that (sort of a "meta" song about his songs). It draws from a conversation he he had on the Jordan B Peterson Podcast in 2019 about how he's used his unique skills and gifts to come up with this interesting style of music production.

While Akira the Don originally called what he was doing "Lo-Fi" to describe how he was using low fidelity and analog content along with his typical high quality hip-hop-indie-dance vibe electronic digital content, the form of music has taken on a life of it's own at this point. It's called MeaningWave, and he's got songs that he's put together using content from a ton of different people. Check it out here:

Research Question: Is There a More Fundamental Set of Rules for Life than the Original 42?

So the way that Dr. Peterson is pulling these rules down from his original Quora post, writing books, and constantly lecturing and tying them all together, while keeping them grounded in his central ideas about meaning and the architecture of belief has had me intrigued for a few years now. How is there such an infinite depth and richness in this list that he continuously draws from? Could there be a way to extract a more fundamental set of rules from the 42?

I mean, we should take a moment to acknowledge that everyone knows the Answer to the Ultimate Question of Life, the Universe, and Everything is, quite obviously, 42. But still, I've been wondering if I might be able to break things down even more concisely. To that end, I've decided to apply some Machine Learning (ML) tools for Natural Language Processing (NLP) and to visualize the space of 42 rules to see what I could glean.

Analyzing 42 Rules For Life With ML Tools For NLP

In order to do some interesting analysis, I put the entire list into a Jupyter-Notebook in Python, as shown below:

We should note here that for some reason the original list on Quora really has only 40 rules listed. There are also 40 rules in the song. So...there seem to be 40 rules, thus the analysis rolls on with 40. I think that technically you could break two of them down into two each, but I'm also not completely sure about that and certainly open to other opinions on that point.

My interest was initially to see if I could use NLP tools to take a look at any substantive words that show up with relative frequency in the set of 42 rules. As a first cut, I chose to use the simplest possible technique to create word embeddings (i.e., the "numbers versions" of the rules or encoded representations of each rule in vector form as a list of numbers, instead of as a list of words); namely, I chose to simply count the most frequent words. In Sci-Kit Learn, there's a tool called CountVectorizer that you can used to accomplish this.

Before creating word embeddings, however, I had to run the 42 Rules of Life through the standard order of operations to cleanse the text of any non-standard inconsistencies. In NLP, generally this includes the following steps:

1. Stripping html tags
2. Removing accented characters
3. Expanding contractions
4. Removing special characters
5. Lemmatizing the text (see https://nlp.stanford.edu/IR-book/html/htmledition/stemming-and-lemmatization-1.html)
6. Removing stopwords (see https://en.wikipedia.org/wiki/Stop_words)
7. Removing non-words

After removing all of the above, I created a custom corpus of words from the 42 Rules. The most popular words, along with how many times they (or their variant) appeared in the rules are shown below:

CountVectorizer Word Counts

These findings were relatively robust to various techniques that can be used to create embeddings. When I used a Term Frequency Inverse Document Frequency approach to creating embeddings, for example, which is a bit of a gold standard in the industry (see for example, https://towardsdatascience.com/tf-term-frequency-idf-inverse-document-frequency-from-scratch-in-python-6c2b61b78558 for more detail), I also saw similar results. Note that the counts are normalized in the figure below.

TfidfVectorizer Word Counts

So we're seeing that not is clearly killing it. Followed in a distant second by thing and one, followed in turn by someone, and make.

By combining my knowledge of Dr. Peterson and his ideas with associated content that others have created in a "meta-way" for him (like Akira the Don) and with my knowledge of Artificial Intelligence and Machine Learning tools, I'm choosing to creatively combine my biological ability to detect patterns with my technical ability to develop computational pattern detectors using open-source coding techniques.

For me, the point of my search for "Meta-Rules For Life" is to stretch myself out into the unknown and see what I can learn. Who knows - it might be useful for me as I begin the path of my Next Great Adventure, and it might even be useful to some of you in my network.

Proposed "Meta-Rules For Life," A First Cut

Below I've taken the most popular words that were used to create embeddings, and I've used them to craft new rules that I think help to concisely deliver what I consider to be Dr. Peterson's most important and useful messages. Further, I've paired each "Meta-Rule" to multiple MeaningWave songs that think help to understand why I chose the words that I chose. In order to get a feel for these rules, it's highly recommended that you listen to each song. Therefore, the four "Meta-Rules" that I'm proposing as a starting point for this series are:

1. Figure out what not to do first. Then aim at something.
2. Focus on one thing at a time, from the bottom up. Then you will know what to do next.
3. Focus on someone. First, make that someone yourself.
4. Get to know something. Develop yourself.

and the songs are below...

Figure out what not to do first. Then aim at something.

Focus on one thing at a time, from the bottom up. Then you will know what to do next.

Focus on someone. First, make that someone yourself.

Get to know something. Develop yourself.

I plan to investigate Dr. Peterson's 42 rules for life more in the coming weeks and months, and eventually to do analyses larger chunks of text derived from Dr. Peterson's books and lectures.

In my next article, I'll be reviewing the word embedding spaces created by various embedding techniques, including CountVectorizer, TfidfVectorizer, and also by leveraging GloVe embeddings (https://nlp.stanford.edu/projects/glove/) so as to try to capture semantics between words as well. It appears that in all cases there is a clear "what to do" design space of rules for life, as well as a "what not to do" design space, which I find quite interesting. This aligns with the analysis done so far in creating the first "Meta-Rule," which is focused on what not to do.

Until next time!