I'm enjoying this interview between Holly Erin Copeland and Steffan Iverson:
Tracking meditation using Meditation Monitor with Steffan IversonSteffan
has collaborated with Daniel on an AI-Daniel. He also collaborates with
Kaio Shimanski. I don't know much about Holly (this is the first time I've seen her).
The opinions of Holly and Steffan are informed by their experience as real-world data scientists. I felt a camaraderie with them based on my own technical background. For what it's worth, I have an MS in Statistics (officially "Statistical Data Science", but this is really just branding), and I've worked professionally as a freelance data analyst. Despite all the hype about AI and such, data science is mostly difficult, unglamorous work. It requires some math and programming skills, and some domain expertise. It also requires a lot of patience, healthy skepticism, and honesty. You must be willing to put basic assumptions to the test, over and over again. Also, as Steffan says, you must be open to accepting what the data tells you, even if it's now what you want to believe.
To give a small example from my own professional life, I spent two years analyzing a big table of user practice data for a music education software company. The app team had basic questions about content (song popularity, amount of time spent practicing drills and exercises, etc.) and user practice habits (practice time per day, rate of progress through content, practice accuracy, etc.). It generally took several attempts to get meaningful metrics for each of these questions. The technical challenges were substantial: sometimes we didn't standardize our data correctly; sometimes we misunderstood a subtlety of data collection that invalidated a metric; sometimes we had subtle mistakes in the code that messed up an aggregation step; etc. Beyond the technical challenges, there were many interpretive challenges: often a metric didn't actually indicate what we thought it indicated. As analysis progressed, we honed in on a few basic insights about how the software was being used, which turned out to be quite different from what the team had imagined. In this case, data science produced a clearer picture of a complex system.
Steffan makes a number of points that are useful to keep in mind when thinking about practical data science for meditation:
- The practical value of the equipment involved (hardware, software, mathematics) is inversely related to its expense (money, time, energy). The Muse turns out to be cheap enough, reliable enough, and pleasant enough to use for daily meditation tracking.
- Iteration is key: collect some data; analyze; repeat. It takes a while to get to know your equipment, your data, and your metrics.
- Simplicity is key. There are steep prices to pay for complexity, including interpretability of results. Simple, well-understood tools (indeed, the math tools one actually studies in a stats program, imagine that!) are often sufficient for the analysis task at hand, but it can take a while to realize that.
- For analysis of similarity, style, etc., what you're measuring matters much less than the consistency of measurement.
After exploring Steffan's work and playing around with his
Meditation Monitor app, I've been motivated to buy a Muse and start analyzing my own data more seriously. Already, this has been a productive exercise, getting me unstuck from some frustrating technical and conceptual problems. I'll write about my discoveries more in a separate post. I would recommend the Muse to anyone who's interested in this sort of work and who can spare a few hundred dollars.
