Algorithms, transparency and informational efficiency

I studied economics from 1991-1998 when the Internet and World Wide Web were in their infancy. 1991 was the year that is usually considered the final year of the Cold War. Of course there was a lot of talk about secrecy and spying and asymmetric information. However, most economic theories assumed perfect information as a key to achieve equilibrium and maximize value.

Having worked in IT (information technology) for most of my career I have always been extremely interested in the power and value of information. In my world, data is gold and information is power and transparent information is a power for the good. I spent a great deal of time looking into how to estimate the value of open data and shared information in my doctoral dissertation. However, I realize that I haven´t really taken a critical stance towards the technology that is used to create information from the data, but rather focused on the (in my opinion) vast potential for using IT to create a better world. I view technology as a tool and the assumption I make is that we are in control of how well we adopt the available tools and how they are used (governance and regulations), and that people in general want to do good deeds but sometimes lack the opportunity, ability or motivation (or all three) to reach the desired end.

That said, I am going to take a slightly different stance here, in a very short post, where I am just flexing my thinking muscles a bit. The reason for these thoughts is that I am writing about a company called Zillow. Zillow is a real estate platform and they collect a lot of data, and develop machine learning algorithms that crunch these data and come up with an rough estimate of the value of your house. You can check out their book here. There is a similar site here in Denmark called Bolighed.dk (actually, they do not have the property listings per se but they do provide an estimate of a reasonable price for a given property). Well, as I will sell my house in a few months and I want to get a good price (obviously) I have a personal interest in how house prices are developing and how I can estimate when is a good time to sell. And of course, I really love to get more information, to be able to browse the web and look at prices and pictures of the property in my neighborhood. I cannot even imagine how to sell a house without all this information to build on. As usually, I have been very positive towards these developments where we have more data and more information.

But then I started thinking. What if, for some unknown reason, the price estimate for my property would be way off as compared to my own expectations? As the estimate is public it would obviously influence the buyer, if he or she would look at this information (here they estimate that 80% of all property in US have been "zillowed" which means that people have looked this information up). So if my own price estimate and this estimate are way different, what is going on? (this is a hypothetical example, and not the case in my personal situation). Are my expectations inflated? Is there something there that I missed? Do they know things about my property I do not know? Nowadays there is a lot of interesting information available (see great use examples here) and your property price can depend on so many things: How close you are to transport, the quality of the schools in the neighborhood, closeness to other services, the socio-economic status of your neighborhood, closeness to water, view, age of property, who was the architect of the property, is the kitchen new., what kind of floors etc. Many of these things you can do nothing about - others you can influence.

So - while I appreciate that it is good to calculate a reasonable price based on many parameters I do worry what the effect these algorithms have if they are not taking in the right parameters, especially as the algorithms are completely nontransparent to you as the user.

Continuing with this line of thought, if your bank uses much more data to determine if you are creditworthy or not, if your insurance company does the same, if you municipality does the same, what then? While I am still completely pro use of more data (as I said, perfect information was always presented as the key to efficiency in my formative years), I do worry what happens to efficiency if these algorithms start to influence us so much that we start depending on them to provide some kind of a "truth", but if they fail or are unfair or even corrupt we would not know as they are a black box. Kind of like the Google flue experiment but harder to compare to facts afterwards, as who knows the "correct" price of a house or your "correct" creditworthiness? 

The danger is that the journey that started towards more information transparency will backtrack to even more opaque markets. Therefore I think that algorithms should be well documented and that some people should know what goes in and how the data are processed. As algorithms are generally very complex, making the algorithm itself public might  not improve our understanding at all, and might also ruin the financial motivations of companies developing such algorithms. But there could be some kind of "algorithm auditors" and some documentation about what parameters/data are being used. Then we have the potential to understand at least the inputs and some smart person would have to make sure that there are no errors, just as auditors do for accounting.

As usually, all thoughts and ideas are welcome. Should we extend our quest for  open data to transparent algorithms?

p.s.  did you know that according to this article in the Economist, a huge chunk of all published research papers actually have an error in the Excel spreadsheet used for the underlying data and calculations? I think with horror of all the linkages and VLOOKUP statements I have in mine. This is why it is a good idea for society to open research data, although it might make you as an individual a bit vulnerable...