Weekly review through February 26

The daily review

The last couple weeks I tried to reinvent Shuff from the philosophical ground up. During this week I came upon a much simpler solution that works fine for me right now. I didn’t even have to code anything!

To recap: Shuff allowed me to track the amount of time spent doing some things, but that doesn’t actually reflect my progress in various dimensions that I consider important. I struggled to think about how to assess that progress in other ways. The solution I came up with is to make a quick assessment of the state of things once per day. I created a Shuff task called “daily review” that I make sure to run through every day before bed and record in a spreadsheet. So far it hasn’t been a problem to follow through with that. I’ll explain some of my choices for what to track, though they’ll probably be evolving over time:

  • Learning Chinese: Right now my biggest concern is getting through the Heisig characters, so I’m tracking my number of written characters learned on Skritter, which is easily found on my progress page. I could do something like beeminder with this data since it should be increasing at a constant rate.
  • Eating healthy: By eating mostly paleo/primal, I don’t care so much about calories, or even tracking weight. Instead, I’m just trying to get myself to eat more home-cooked meals, so I put smiley faces for breakfast, lunch, and dinner when I do that. In addition, I’m trying to make sure I eat enough protein, so I have a quick calculator spreadsheet and record an estimate. This usually results in me scarfing down something right before bed, but I’m becoming more aware how much I need to eat during the day.
  • Getting crap done: A huge problem using Shuff is that my generic to-do list doesn’t get emptied. Meanwhile, I do pretty well about clearing my inbox, but since the two go hand-in-hand, I record both my number of inbox items and my to-do list count each day. So far the to-do list has only gone up, but now at least the problem is more salient. Getting both down to zero might make it more tempting to keep them there.
  • Keeping my house in shape: At first I tried record the cleanliness state of each of the main rooms (kitchen, living room, bedroom), but that was depressing. Instead, I’m taking a smaller step: recording whether I’ve done at least one “clean up” Shuff task that day.
  • Being a good researcher: I’m still thinking about this one. I haven’t found a good way to quantify anything. So for now, I’m doing some simple qualitative tracking: an idea, an accomplishment, and something random from each day.
  • Getting sleep/getting away from the computer/not abusing the daily review: Something that already happens is to start doing the daily review before bed and then try to improve some of the categories. Which is good. Except that it’s bedtime. So I also have an “time I went offline” column to put some pressure against doing that too much.

It remains to be seen how much this tracking can improve the issues I had with Shuff, and whether I can continue to do it consistently. There are other things I care about that aren’t being tracked, but I can continue thinking about them.

Living by a principle

I’ve mentioned Bret Victor several times already on this blog. He has a great talk about the principle he lives by and why he encourages others to think about living by one of their own. His principle is to give creators an immediate connection to what they’re building. He goes on with amazing demo after amazing demo, showing ways to make changes in code (or circuit diagrams, or animation) instantly visible and arguing that some discoveries that could never happen before are enabled by this connection. His principle applied to math would be that manipulating algebraic symbols does not give one an immediate connection to the result desired. I’d have to agree with that much!

The way he describes this idea of living by a principle is that one sees a problem in the world, perhaps one that is taken for granted by most people. He says he found his principle by looking for a theme in some of the work he’d done. In other words, it’s not something that comes right away, or comes easily.

I don’t know whether I’m ready for a principle of my own yet. In academia we call this a research statement, though some are more specific than others. One theme that I see in my work is the idea of making processes that are hidden more visible. Shuff is an embodiment of the process of choosing what to do. Some of my ideas for learning tools are about exposing the problem solving process in math problems, or revealing the patterns and vocabulary being learned while studying a video. By materializing these processes, we can better understand and manipulate them.

One commonality between Victor’s principle and the one I describe here is that they are controversial according to what I’ve been reading for a course I’m taking, Design Perspectives in HCI. Design typically recognizes a strict separation between the design process and the implementation of the artifact (Goel & Pirolli, 1992). But Victor is saying the designer should be able to immediately see the result of his whims! Maybe he is talking bringing computer tools closer to sketching, what some designers seem to consider the best we can do (e.g. Purgathofer, 2006). But I think we need to start thinking about a shift, where designers can see the real impact of their design instantly — a merge of sketch and implementation.

My principle relies on the notion that determining a concrete process for activities is both possible and worthwhile. The field of design has struggled with this issue for a long time, with Simon, 1996 attempting to define design as searching a state space, and many retaliating against the idea that design can possibly be captured be any formal model (see Schon, 1983, Jones, 1992). My thinking is: a sufficiently abstract model could be conceived that would describe any [design] activity. More specific ones are more interesting and useful but decrease the number of instances they describe. Searching for the model that strikes the right balance is challenging but can be rewarding (as illustrated by my musing on Shuff).

Short thoughts

  • I’m learning how to do qualitative research for my new research project. Creswell, 2007 describes five types of qualitative research: biography, phenomenology, grounded theory, ethnography, and case study. Our research methods course uses contextual inquiry (holtzblatt1993contextual), and I think that fits best with my research. The book has a strongly business-consulting — as opposed to scientific — perspective, so I’m attempting to find and apply of the scientific perspectives from other fields. I also realized that there may be a much easier source of data: online communities. Apparently this is called netnography in the marketing world (Kozinets, 2002).
  • Sean “Day [9]” Plott’s episode on eliminating assumptions is not Starcraft-centered and thus worth watching for anyone. In spirit of my principle, I’m still thinking about a good process for applying his advice.
  • The future of citation management, reference sharing, journal clubs, research as we know it… it’s coming!

Weekly review through February 19

Shuff: New perspectives

Task selection

Getting a random task to do was the beginning of Shuff, but it’s always been questionable whether this is a useful thing. I have a new perspective on Shuff that suggests not random tasks but something similar is useful.

At any given time, we are in some state of being in our environment. The desirability of this state can be assessed by questions about various aspects of the state. For example, “Do I have to go to the bathroom?” (or “Am I going to the bathroom?”) are questions one can ask about their current state. Many queries are handled non-intrusively and automatically by mechanisms that may be physiological (the feeling of having to pee), social (hearing your name called), or visual (seeing a dangerous animal approaching you).

As our lives are complicated and externalized onto the digital world, many built-in mechanisms fail to suffice. Nothing in the physical environment outside my computer tells me that I should catch up on my emails. Or that there is a bug in my codebase. Or that I want to learn twenty new Chinese characters today. It’s clear why people are so addicted to their computers when we realize that much of our primal instinct about fleeing from danger, gathering food, or finding a mate has to be channeled through our screen and keyboard.

Abstractly, we can think of all of the assessments-of-state as streams of digital information, and the job of processing these streams inspires the consumption model of Shuff that I described last week (and presented at Quantified Self Pittsburgh on Wednesday).

The new perspective is that we can process the assessments directly and then, perhaps, decide on a task to do. Shuff can be a tool to supplement our digital environment with cues about our state of being that we can act on. Because there may be more potential things to assess than we can get to, the assessments may still be shuffled. Some of the current tasks can be quickly converted to assessments (Clean room -> Is your room clean?, Empty inbox -> Is your inbox empty?). Many will not come from a task but could suggest actions depending on the response (Are you focused? (if not) -> Talk a walk, remove distractions, or set a timebox). Meanwhile, everything is being tracked, so you can look for trends between assessment responses and what you actually did.

Still, if our natural feedback loop is non-intrusive and automatic, “consuming streams of digital assessments” leaves something to be desired. Unless that digital information actually is bright and delicious. Someone at the QS meetup mentioned an app that shows a tree where the leafs wither and fall when you neglect categories of tasks. The metaphor doesn’t have to be quite that blunt, but thinking through it may allow Shuff to truly be “an environment where natural [productivity] flourishes”.

Task performance and the graph

The tasks on Shuff are bad. Shuff aims to make bad tasks more appealing. They are more exciting because they must be done within a time limit. They are more rewarding because they add points and make the graph bigger.

A good task would not be on Shuff. To use a familiar example, going to the bathroom is not on Shuff because I know exactly when to do it, I know exactly how to do it, and it’s intrinsically valuable. Sometimes tasks are removed from Shuff because they become similarly automatic, and this is a good thing. However it means the graph gets smaller over time, and this is sad.

Need to think more about this…


Am I not reading anything or just forgetting to clip it? I honestly have no idea… Anyway, Metcalfe & Kornell, 2005 is a nice paper on behavioral choices in studying that presents the Region of Proximal Learning model: people will not choose to study an item they think they already know (making a JOL – judgement of learning); they’ll stop studying when they think their learning is progressing too slowly (making a jROL – judgement of the rate of learning). Interestingly, this can mean they stop either when they’ve learned everything or when they are stuck with less knowledge.

Short thoughts

  • I’ve been arguing relentlessly in my Scientific Research in Education course that science is not just about A vs. B randomized controlled trials. I would hope not, as they first appeared in 1948.  Some people should do them, more people should do things that are not RCTs, like thinking. Dan Meyer posts a similar point this week in reference to Khan Academy’s use of A/B testing.
  • It weirds me out a little whenever people bring up mathematical models for stuff the brain does. If my brain is really doing all those integrals automatically, why does my math homework take so long? Marr, 1982, talking about visual processing, finally addresses me! “[T]he reader…should not be put off by the notion of a coordinate frame…[which] does not literally imply that a Cartesian coordinate system…is some how laid out across the striate cortex, and that whatever some line or edge is noticed it is somehow associated with its particular x- and y- coordinates, whose values are somehow carried around by neural machinery.” Glad to clear that up! He continues with an example of a potential physically-realizable solution: “an (implicit) anatomical mapping that roughly preserves the spatial organization of the retina together with a representation that makes local relations explicit…” I’m sure it would get tedious to carry out that explanation every time a mathematical model is used, but there are some cases (e.g. theory learning based on Bayesian models such as in Schulz et al., 2009) where I couldn’t figure out where to look for any hint of a physical mechanism after reading several papers.