In reading Chris Anderson’s The End of Theory, I found myself constantly swinging beyond agreement and reaction. In it, Anderson writes that as the Internet becomes  an enormous corpus of human history, modelling is becoming irrelevant. The reason for my mixed feelings is that while I disagree with the conclusion (or the extent of it), I very much understand the points Anderson makes to get there.

Anderson is certainly more qualified to speak on the scientific method than I am. He was a scientist for many years, before becoming a writer for the very well-respected Nature and Science, eventually settling in as Editor-in-Chief of Wired. Yet, his argument appears to  be unnecessarily broad. The scientific method won’t die off, as there are still many uses in which it will reveal knowledge in traditional ways. However, as we reach problems that the scientific method cannot help with, the digital age’s gradual progress toward a corpus of human data may help. (A sidenote: perhaps Anderson intentionally exhaggerated his arguments to spur discussion. If this is the case, given the reactionary comments an the Wired page, it was not particularly successful.)

First reading the article, I kept thinking: Does having more data not allow you to create more relevant models? If a large corpus of data reveals something, do we need to retread the steps every time we intend to built on that knowledge? Contrary to my first reading, though, I think this is the exact point Anderson is making. If we have a large set of data (one nearly incomprehensible in scale, encompassing our present and our past), we are safer in assuming that knowledge derived from that data is sound. In other words, as our corpus of tangible human knowledge grows, it makes fallibilism (the bane of my existence) increasingly irrelevant as uncertainty decreases. If something is correct a thousand times out of a thousand, yes it could still be incorrect on that 1001st time, but it’s more reliable than if it had simply been right ten times out of ten. As we reach problems that we simply cannot test for absolute certainty, correllation will have to do. What Anderson appears to suggest is that, given the size of the data, this isn’t as bad as it sounds.