The article explores the difference between the distributions of doing (normally distributed) and learning (log-normally distributed) processes. It discusses how traditional project management techniques fail to account for the log-normal distribution of learning tasks, which leads to underestimation of the time and resources required for software development and other learning-intensive tasks. It also examines the implications of this discrepancy on employment practices, particularly in software development, where specific knowledge and tooling experience are highly valued.
Main Points
Doing vs. Learning: Distribution Discrepancies
The concept of doing is commonly perceived as a normally distributed process, whereas learning curves, exemplified by the ‘leaky pipeline’ model, follow a log-normal distribution. This challenges conventional project management and estimation techniques, illustrating the inherent unpredictability and risk in learning-based tasks.
Impact on Software Development and Job Market
The prevalence of non-normally distributed learning processes impacts software development practices, job market expectations, and the overall approach to tackling new tasks. It underscores the necessity for realistic planning, acknowledgment of learning curves, and the strategic value of prior experience.
Insights
learning is at least non-normally distributed
Everything we know about traditional project management, from Waterfall to Gantt charts to estimation practices, are on some level based around the idea that each individual step in the chain is bell-shaped: A process taking twice as long as it normally does might be 2 standard deviations out, aka in the bottom 2.5% of outcomes. But in a log-normal distribution, processes taking two, three, or five times as long are much more common, and this throws things into disarray.
business processes that are normally-distributed are the exception, not the norm.
Every new process a person has to undertake will have at least some phase which is dominated by learning. That’s as true of web dev as it is of learning to operate the drive-thru at McDonald’s. It takes a special kind of rigor to transform that into an activity routine and repeatable enough that you get a long tail of normally-distributed, profitable activity.
Links
- gwern.net
- his essay on “leaky pipelines”
- just-in-time learning is at least non-normally distributed
- efficient markets hypothesis
- just-in-case learning is less this way
URL
https://hiandrewquinn.github.io/til-site/posts/doing-is-normally-distributed-learning-is-log-normal/