I was reading Klaus Oesch's book about the "digital revolution". The book was written in 1993, and I found fascinating the many predictions, which are now manifesting themselves and seemed to have been very much on target (digital TV, wireless LANs, PDAs, etc.). It's probably obvious that you can predict the future (at least the major trends) through the past.
But could you automate the process? Is it possible to create a computer program to find those trends which extend from the past to the future?
Detailed records about the past are important, but they do not appear from thin air. The most important thing about predicting the future is the now. Keep records of the state of now, and after a moment the records themselves become a part of the past. After a longer moment you have a lot of these records about historical states, and then you can divine the directions the future might take. Memory and imagination, observation, these are the keys to predicting the future.
For instance, suppose you have data about the climate over a period of 100 years. It's a small period of time, but let's say the data shows that the average temperature has risen steadily over time. What's the conclusion about this? The Earth is warming up, it seems. Now, think about causality, i.e. cause and effect. Let's consider the Earth warming up as the cause. (We're not interested on the Earth warming up being the effect, i.e. we don't care why it's warming up) What could be the effect of the Earth warming up?
It might be that (1) gradually the vegetation zones will shift more north. This likely means (2) animals too, as they follow their food sources. Mosquitoes and other pests will also move more north (3), and (4) grow in number in their original habitat. So you need either (5) more pesticides or plants more capable of withstanding the bugs (genetically altered, perhaps) which means that in the longer term you might want to invest in chemical companies or biotechnology companies.
So, from "the Earth is warming up" we eventually got into "invest money into chemical companies and/or biotechnology companies". Quite a non-obvious leap if you disregard the parts in between. Now, the million dollar question: could this building of the causality chains be automated? Would it be possible to write a program which, given some fact as input, would use its massive database of interrelations (or dependencies) between various things and dig out the nth level prediction about the future?
The program as described above does have one obvious limitation; its prediction capabilities would only be as good as the database it has. If the database is of a narrow scope, it might miss out on the many possible outcomes of the causality chain exploration. In the example above, a database missing the knowledge about ways the pests can be eradicated would also have missed the suggestion of investing to chemical/biotechnology companies.
Also, the further you spread out from the original fact, the more "volatile" or "imprecise" the prediction will become. In the global warming example above, the value of n was 5. Very high values of n would probably be no better than taking a (remotely plausible) wild guess. The program would likely not be suitable for very dynamic situations like analyzing the short-term political events of a region, like the DARPA FutureMAP program was.
Despite these limitations, such a program would be interesting to devise and use, even if it were within some special niche. I don't know if there already exists such a software tool, or if such a tool is being planned somewhere. Some Artificial Intelligence researchers might be interested in making this kind of program.
What was the DARPA FutureMAP program? In case you have not heard of it, the Defence Advanced Research Projects Agency (DARPA) had experimented with something called FutureMAP ("Futures Markets Applied to Prediction") as part of their Total Information Awareness (TIA) program. The FutureMAP was a software tool which used a marked-based approach to determine the probability of various events in the Middle East of interest to the US military, such as the assassination of the King of Jordan. The FutureMAP differed from the Artificial Intelligence-like approach presented above, as the FutureMAP got its intelligence from the aggregation of the actions of many individuals who aimed to maximize their profits through "betting" their money correctly. If there were a rumor about a bomb strike or a strong gut feeling about an event going to happen, people would place money on it (buy the futures), it was reasoned. If a lot of people did this, then there would be seen a certain "push" towards this event, an indication that something was about to happen, which the FutureMAP system would show.
Notice that I used past tense when talking about the FutureMAP. This is because (at least officially) the program has been cancelled due to certain issues with the market-based approach: would anyone really use it due to the heavy US government involvement? Could the "bad guys" manipulate the markets to feed wrong information to the system?
Software-based prediction of the future seems a very interesting subject. Such a tool would definitely be useful for people who have to make decisions which have no immediate impact, but which have consequences that are seen 10 or 15 years from now.