- In certain fields, age is a proxy for experience - typically people gather more experience and knowledge over time.
- However, in highly empirical fields, an efficient experimentation methodology allows you to highly compress the amount of time it would take to gain more experience.
- One such method is running many concurrent experiments at the same time.
- There are a few examples of companies that do this in machine learning: Google Brain, Deep Mind and OpenAI.
- In an article called Understanding is a Poor Substitute for Convexity, Taleb explains that under conditions of extreme uncertainty, rather than gaining more knowledge, you are better off improving the payoff function. ^1nStrategyAndConcurrentExp
- Same value as the 1/N strategy in convex tinkering
- As a company you can literally run experiments in parallel - multiple
- As an individual you are limited to experimenting concurrently - cooperative multitasking.
- Unless, you "chord" together multiple experiments / activities.
- A fine-grained parallel system is able to carry out what Douglas Hofstadter has called a “parallel terraced scan.” This refers to a simultaneous exploration of many possibilities or pathways, in which the resources given to each exploration at a given time depend on the perceived success of that exploration at that time.
- Douglas Hofstadter: Parallel Terraced Scan: The search is parallel in that many different possibilities are explored simultaneously, but is “terraced” in that not all possibilities are explored at the same speeds or to the same depth. Information is used as it is gained to continually reassess what is important to explore.
- How can I apply concurrent experimentation in learning?
- What other effective knowledge seeking entities use concurrent experimentation? (DARPA?)
- Should I change the name to concurrent tinkering?