A failure to transfer knowledge in a timely and accurate way can lead to a host of consequences for an agency or program. For example, the U.S. House of Representatives' Budget Committee based certain fiscal policies on a spreadsheet model that was later found to contain errors.
An emerging technology can help us to use social networking to avoid such errors. It supports the writing of models in executable English web pages. When a model is run, there are English explanations of the results, showing step-by-step calculations and the supporting data. The explanations can serve as audit trails.
In a 2010 paper "Growth in a Time of Debt," Harvard economists Carmen Reinhart and Kenneth Rogoff (RR) analyzed data from many countries over more than a century. Their spreadsheet model led them to conclude that growth slows once debt exceeds 90 percent of gross domestic product (GDP).
The paper attracted considerable attention, and the chairman of the U.S. House of Representatives' Budget Committee cited RR's paper as support for an austerity policy.
However in 2013, Michael Ash, professor of economics and public policy chair of economics at the University of Massachusetts, Amherst, graduate student Thomas Herndon, and Professor Robert Pollin published "Does High Public Debt Consistently Stifle Economic Growth," questioning RR's result. They pointed to an error in a spreadsheet formula, and to an unusual way of calculating an average.
Social Media Can Confirm Knowledge
This can be viewed as a knowledge transfer failure, one that happened to have a significant national policy aspect. Non-economists could easily understand RR's result, but only economists were able to test the spreadsheet formulas and method used to do the calculation.
Instead of using spreadsheet formulas, RR's model could be written as an executable English web page. It could then be refined by a social process similar to Wikipedia's.
Since the English is executable, the model can be run as a program. It can then explain its reasoning process in detail, showing the supporting data. So, non-economists can be first-class contributors to the knowledge transfer process.
Executable English as a Web Program
Here is a small example showing how to write and run an executable English model. It uses part of RR's historical data for U.S. debt and GDP. The example finds out how the ratio growth/debt changes with debt for some recent five-year periods.
An application written in executable English consists of tables of data with English headings, and syllogism-like rules. In writing rules and table headings, there is no restriction on vocabulary.
Here is part of an RR data table with an added English heading:
And here is a simple rule that uses the table:
We can use the conclusion of the simple rule twice to find percentage growth:
The next rule finds a percent growth for a sequence of years we will want to use for an average:
Next, we find an average growth:
After typing in a few more similar rules we can write:
We can then run the rules to get an analysis result table like this:
In the table, debt as a percent of GDP falls from 1959 to 1979, and growth/debt rises, which is presumably a good outcome, and is consistent with RR's main finding. From 1984 onward, however, growth/debt is low, and it does not seem to respond to changes in debt.
Explanations as Audit Trails
So far, we have some executable English rules, a table of data, and a result of running the rules using the data. In transferring knowledge from one organization to another, such as from RR to the Budget Committee, we would want the recipient to be able to check that the result is correct.
Arguably, it's easier to find errors in executable English knowledge than by inspecting spreadsheet formulas, but it's still a challenging task. To help with this task, we can pick a row in the analysis result table and ask for an explanation that shows each calculation step in detail. If we pick the 1979 row, the explanation starts out like this.
In an actual explanation, the first line of this step is a hypertext link that we can use to drill down into details. Following some more links we get to steps such as
Continuing to drill down finally reaches the supporting data. So, an explanation can serve as a kind of audit trail that shows how a result was computed and also shows the data used.
Executable English is a new way of representing knowledge that can help with social-network-style knowledge capture, with accurate knowledge transfer, and with explanatory audit trails showing how results are computed.
The system that supports this is live on the web, shared use is free, and there are no advertisements. You can find the system via a Google search for Executable English, and you can also use Google or other search engines to find knowledge that is written in executable English.