Principal Component Analysis

Years ago when I was helping my Grandma with her computer, she would always ask me “Ethan, how do you know how to do this? You’ve never done it before!” and at one point I responded with “I have a computer in my brain, so I try it there first, then I can do it with you”. That wasn’t (and still isn’t) true, but at the time it was the easiest way to explain something that felt so natural to me.

Now as I reflect, I realize that I was simply good enough at making predictions about how the software would act, that to an outside party it would look like I was already familiar with the specific behavior. This is something we all do every day. If an app has a new popup or interface style, we have enough knowledge based on what we’ve seen before to be able to use it fairly well (a prime example of this is correlating the floppy disk icon with a “save” action). And as there is more and more exposure to the norms of a situation, guesses can get even better.

I took a class on the fundamentals of machine learning while I was in Denmark. Machine learning is a complex subject that continues to evolve on a daily basis with new breakthroughs and impressive demos (look at ChatGPT). However, at its core, it boils down to statistics, linear regressions and a smattering of other math concepts.

One of the more fundamental techniques within machine learning is something called Principal Component Analysis. Essentially, the goal is to take a ton of data and reduce it to something more directly useful but slightly less accurate. The power of this is you can choose the number of components you want to use. If 85% accuracy is enough you might only need to take the first two components instead of all eight data values. This can save significant time, space and energy.

To reiterate my point in the second paragraph, we do the same thing every day for everything we see and interact with. One of the most powerful differences between humans and machines is our ability to treat things generically while also using specialized knowledge (this isn’t strictly true, but given the data available to us, humans tend to do a better job right now). We know how to use a crosswalk, so if it’s a crosswalk in Copenhagen, London or Portland, our intuition will still work. And if something is totally different (an edge case), then we still have the skills to quickly adapt and give it our best effort to resolve in a reasonable fashion.

One of the more perplexing things I came across while abroad is how this breaks down when it comes to analyzing or trying to understand people. I feel sometimes my EQ leaves much to be desired so I crave any insight in the area. Unfortunately principal components are a rather outdated crutch in the social arena.

At what averaged out to twice a week I’d had a conversation like the following with a fellow student while abroad:
Them: Wow, <insert thing that was slightly unexpected such as the weather>
Me: Yeah, that is interesting!
Them: In <insert country they are from> I’m used to <insert what they expect> so here in Denmark it’s hard to get used to <unexpected thing>
Me: Yeah that’s a good point. In the US, <insert my experience with unexpected thing>
Them: Oh really? That’s very interesting! I would have expected <insert their expectations of the US>
Me: Yeah! Although it does depend on where you are in the US. I would have expected <insert expectation based on country they are from>
Them: Nope!
Me: Huh you learn something new every day!
Them: Yeah same here! Wow Ethan has anyone ever told you how good you are at fictional conversations?
Me: Yep, apparently I’m the best.

That’s approximation at work. Most of the time it’s all going as expected, but sometimes there are edge cases. This is where prejudice and bias come into play and can be very harmful. This leads to some fascinating contradictions regarding people you know really well and people you don’t. If you are like me, every time you talk with someone, you are always trying to predict what their reaction will be. Generally for people I know well (my family or close friends), my predictions are generally decent. In the same way, celebrities or people who have established personalities are also mostly easy to predict. Because each person/culture/society is special (source: Mister Rogers), we are all edge cases and it’s impossible to pin down exactly how someone thinks or feels.

Originally, I was going to include a brief exercise where I would list several attributes of a person in my life and if you knew them you would immediately know who they were. As I was trying to come up with these attributes, I realized how difficult it was to find a set that wasn’t too simplistic. People, by their nature, are complex beings with unpredictability being a continually presenting trait.

One of my goals for this year is to look past the principal components of the people in my life. It’s easy to fall into the trap of looking at a person’s social media presence, Spotify playlists, text messages and responses to “how are you?” as a complete picture of who they are. Those are all easy mediums to warp and shape into the persona people feel obligated to show. I’m doing it right now by publishing this!

None of these thoughts are groundbreaking. There are far more qualified professionals particularly regarding the topic of prejudice which can give much better insight. For me, this is simply another exercise in demonstrating that the world is complicated and connected in impressive ways.