The historic method

History is an intriguing subject of study. In opposition to the tenants of the scientific method, it attempts to derive conclusions on the basis of analyzing existing evidence, where various historical artifacts, chronicles and so forth are used to piece together events using, essentially, simple point interpolation. Attempting slightly more sophisticated methods as linear regression would be a worthwhile exercise.

While the experiments are not repeatable per se, one may make use of the ever heeded saying “history repeats itself” to perform repeated trials in history, seeking cases where a similar conflation of events or “variables”, occurs, and measuring whether the outcomes are predictable as such. Naturally we cannot even hope to identify, let alone control, the control variables involved. Another important aspect is falsifiability, which in principle is possible on the basis of newly uncovered historical evidence, though I suspect this does not work reliably in practice.

Despite these objections and others, I still find history to be both an important and useful tool for advancing human knowledge. I also find it to be interesting, but that clearly does not constitute as an argument in this context. I think the missing key here is a scientific-method like revelation for the humanities, one that is capable of dealing with the type of data available and the limitations on the experiments that can be performed. It can be argued that Bayesian statistics may constitute as such, but I feel that it is too general a tool to be effective and influential in this domain. It was Newton’s discovery of the calculus framework for physics that sparked the revelations that he and scientists that followed him have uncovered, and I wish to see the Newton of the humanities in my lifetime.

Employing machine learning, and unleashing it onto the sheer volume of historical data available, may turn out to be fruitful venture. Envision building a generative model of history, breaking up clusters of interconnected events into causal links and weighted contributing factors. Once these are identified, we may be able to predict events given the existence of their contributing factors. We can also decide the best course of action to take in order to achieve, or avoid, certain outcomes. I appreciate the lack of rigour associated with this type of modelling exercise, and this underpins my disappointment in our progress lately as a humanity. It appears that our methods are starting to scrape against the ceiling of low hanging scientific fruit; we need better tools and processes. With that said, it appears to me that the big data of history remains a, to this day, a ripe low hanging fruit waiting to be picked.