Coding the markets

I posted earlier on Bret Victor’s Magic Ink. Now I’ve finished reading this quite lengthy paper. In the second half Victor goes into more detail on how information applications – which includes most financial apps – should be implemented. Info apps need to learn from the history of user behaviour: Victor uses the example of his own implementation in the BART scheduler app for this. He points out that decades of research have been done on machine learning, but we still don’t have neatly packaged abstractions around the results of that research that would make it usable for Joe Average app developer. This shouldn’t be the case, as we have nice usuable abstractions for all sorts of other areas of comp sci research: file systems, sort algorithms, GUIs etc.

The other major pillar of info apps in the Victor scheme is context sensitivity. He makes a compelling case for achieving this via a dynamically bound component model that sounds similar to that advocated by Brad Cox 20 years ago. Plus ca change !

Finally, Victor discusses the kind of device that info apps should run on. It reads like he’s describing the iPad. Bear in mind that the iPad launched in 2010, and Victor wrote Magic Ink in 2006. So he was remarkably prescient.