In November 2009, I became interested in the economic value of information (EVOI) after reading David Lawrence’s book The Economic Value of Information (1999). I decided to clarify my thinking on EVOI by reducing some parts of the theory to practice. I performed systems engineering in EVOI until July 2010, and then began developing an EVOI software testbed. I worked on the EVOI testbed and associated use cases until November 2011. During this period, I gained experience designing and implementing decision making algorithms and the context in which they operate.
In November 2011, I suspended the EVOI project and began working on decision basis evaluation (DBE). The abrupt change in research direction was due in large part to growing public interest in artificial intelligence safety. I performed system engineering in DBE until October 2013, and then began developing the decision basis evaluation testbed (DBET). I was able to re-use much of the EVOI testbed software for the DBET software development project. I worked on DBET and the associated DBE use cases until June 2015. The DBE book was published on October 29, 2015.
It may be surprising that I spent almost two years doing DBE system engineering and then another almost two years doing DBE software engineering in order to produce the DBE book. There doesn’t seem to be anything in the DBE book that requires such a significant investment of time and energy. The four-year development period can be explained by noting that:
- I only worked part-time on DBE throughout the period.
- I performed a substantial amount of DBE system and software engineering on topics that do not appear in the DBE book. This material was excluded from the book either because it is immature or too complex for an introductory book on DBE. The largest body of work omitted from the book involves performing DBE for a recursive decision problem using a first-person decision model.
- What seems obvious in retrospect is often not at all obvious in prospect. Looking back over my DBE development notes, it is remarkable how many alternative DBE designs I conceived, developed, implemented, tested, and then discarded. Implementation is an excellent way to thoroughly understand a design; unfortunately, it can be time-consuming.