Two kinds of software: It’s time to take world modeling seriously
It’s time to get serious about world modeling. Here’s why:
- There are two kinds of software in the world. (1) Tools (like Microsoft word) and (2) software that depends on a model of the world as its core capability
- Examples of the second category are fraud detection systems, churn analysis, google ads, recommender systems like amazon and Netflix, and more
These systems depend on the degree to which their internal, “cartoon” model reflects the world
In recent years, we’ve learned how to build these models automatically from a stream of data (e.g. fraud data, churn data, click data)
- But if you think about these cartoon models (we might call them “world models”), there are more ways to obtain them than from just data
- When we do interactive data visualization, what we’re doing is substituting a human brain, *informed* by data, for the “world modeling” capability.
- Essentially, we’re substituting ourselves for this capability, in situations / use cases where we recognize that the data isn’t perfect.
- There is an unmet need here, because as Mark Zangari describes in his talk at SFU last fall, it’s asking too much of our brains to do effectively in many complex situations.
- In particular, this is the source of a lot of cognitive bias, and leads to unintended consequences in many situations
- So, we need to think beyond the use of historical data for “world modeling”.
In particular, we need to combine a) the data; b) good interactive data visualization (as with Tableau, Excel, or Qlikview; and c) a good systems model.
- (c) has been missing until recently. This is what World Modeler™ software does, and Decision Intelligence helps us to do, and what Decision Engineers build for a living.