I recently finished reading Shaping the Next One Hundred Years by Lampert, Popper and Bankes; published by the RAND corporation. The book presents a new technique in analyzing a policy's effects on the future which they call Long Term Policy Analysis or LTPA. By long term they mean roughly 100 years out. This technique involves generating large sets of data and creating an environment for human-computer interaction to reduce the dataset. Unfortunately the writing is very academic, which will limit its audience. But, the full text is available online at RAND's website.

In traditional policy analysis a model is made that predicts what happens in the future. This model is very complex as the designers try to incorporate every possible effect and interaction that is understood today. LTPA is different, while it uses many of the same understandings, it uses them to create a large set of competing scenarios on how the future will turn out. One thing that was never stated in the book, but I think is important to understand LTPA is: (fortunately I was given the book over coffee, which gave me this framework to begin with)

LTPA does not predict the future. It allows people to see how their policy decisions effect the future.
The difference here is subtle, but important. No one can predict the future, but what we can do is analyze how a particular decision might play out in future scenarios. We all have done this, and we've been in meetings where different groups of people had different opinions how how a decision today will play out in the future. LTPA attempts to take all of these different opinions on the importance of particular stimulus and combine them into one model. Then we can optimize our policy decisions based on all these different scenarios for the future.

One of the things that concerns me about LTPA is that while it allows the person working with the model to gain great insight, how often is that the person making the final decision? Yes, several times they mention using teams of decision makers as part of the process, but how often is that going to be the entire set. It's those late night meeting in congress where the final compromises are made. If the person making those compromises doesn't really understand, they're likely to sabotage the entire policy through ignorance. This doesn't solve the problem of communicating the results, and in fact I think it compounds it. When discussing potential policy people want to know what "the answer" is. They want to know that policy A causes children to suffer while policy B saves the world. But LTPA doesn't provide that. I'm not sure that isn't a critical flaw in the communications aspect of LTPA. But, perhaps those working with the model will be able to provide "the answer" in a report, even though they don't know it.

I'm curious if, in future uses of LTPA, they will have what I'll call "the Wikipedia problem." Wikipedia always tries to maintain a impartial point of view on topics in the encyclopedia. This means that they will present competing theories on events, explaining them as theories. But, as no one is willing to say that one theory is wrong (they're being impartial) some theories get more weight than they deserve. I'm curious if that won't occur with trying to involve large groups of people working on the model. If you have oil companies and Greenpeace sitting in a room, they are probably going to have different views of the future. The oil guys might want to have the scenario included that takes into account the possibility that we discover the entire center of the earth is oil. In this case, the regret of spending any money on green power may be high, but it really isn't a reasonable scenario. In some ways, not every scenario is valid, but how that will work with the teams working on the LTPA analysis is unclear.

I think one thing that will make an LTPA exercise successful or not involves how to visualize the data. Unfortunately in the example there is only simple 2-d colored graphs as examples. Obviously more complex visualizations will be needed as users will be expected to balance multi-variable scenarios to achieve the best answer. I think this is the most interesting off shoot of the discussion in this book. While many of the uses of LTPA are likely to be on a global scale, visualizing them may be difficult. It is likely that in the next 100 year borders will change, and perhaps even geography. I think this is an interesting problem, and one that will provide researchers material for the next hundred years.

The funniest thing about the book is that I felt that in many places the authors probably wanted to put "and then you just Google the data." Because that is what they're doing, providing an effective, quantitative search though a dataset. I'm sure the editors took all the references to Google out and replaced them with "analysis of data using computer algorithms." I still found it fun to think about when reading through the book whether or not it was part of the author's manuscripts.

I really enjoyed the book, and think it would be an important read for anyone looking at policies designed to last through the long term future. Predicting the future will always be an imperfect science, but making a reasonable guess is important as we do effect the future with our decisions today. It will be interesting to watch the development of LTPA as they continue to work with it and start to build more complex scenario generators.


posted Mar 3, 2006 | permanent link