In his Foundation universe Isaac Asimov introduced the fictional scientific field of psychohistory. In this science fiction setting, it could predict the future by analyzing data and making inductive inferences using various algorithms and formulas. The predictions are not about specific individuals, but broad events. For example, it could predict the fall of the Empire but could not predict who would be the emperor presiding over the fall.
Not surprisingly, people have been trying to make such predictions and there has been some success at statistical predictions involving large numbers of people. For example, the number of traffic accidents a year can be predicted with a high degree of accuracy as can the number of births. However, making the sort of predictions seen in the Foundation series has been beyond the reach of current science. However, this might change.
Psychohistory is, in many ways, would like weather prediction: data must be collected, analyzed and used to create mathematical models. Ideally, the model would be a perfect duplicate of reality and time could be accelerated in the model to see what will happen. Of course, making a good model is challenging.
One limit has been data. After all, an ideal model would be a perfect reconstruction of the world and to the degree the available data falls short, the model becomes less than accurate.
While humans have been gathering and storing information since the advent of writing, we are gathering and storing more information than ever before. In fact, the practice of gathering, storing and analyzing data is a standard business practice that was called “Big Data.” Google was one of the pioneers of modern Big Data, although other companies have started hoovering up data that would be useful in predicting human behavior whether the goal is to sell more baby products or fight terrorism. People contribute (often unknowingly) by handing over data via social networking sites and other ways, such as trading private information for “free” stuff. As such, there is a massive amount of data that would be useful in modeling the future.
The data will always be incomplete. In addition to practical limits, there is also the problem of limited omniscience—knowing everything that is and was. Unlimited omniscience would include knowing everything, including what will be. Given human limitations, we will never have complete information, although we might be able to create something that is better at knowing things than us. But our epistemic limits will prevent a perfect model because there will always be past things we do not know (and perhaps there are unknowable things) and hence they will not be in the data.
But perhaps there is a way around this. If a suitably awesome machine could be built, perhaps it could know everything from a single certain truth—a Cartesian machine. This leads to the second limit.
A second limiting factor is logic. In practical terms, this is the problem of creating the software to analyze the data and to run the model(s). Much of this involves inductive reasoning. After all, the goal is to make an inference from what is known (the sample) to what is not known (the target). This sort of reasoning is essentially philosophical and it is hardly surprising that Leibniz was one of the first to explicitly propose creating a model of reality using symbols. Hobbes also believed that the social sciences could be “real” sciences and took geometry as his model.
While the software we have today is not up to psychohistory standards, there have been some impressive results in the field of predictive analysis. Of course, some of these successes have created moral concerns such as Target’s infamous pregnancy predictions.
As might be imagined, metaphysics is a factor in assessing the limits of predictive software. One key issue is whether humans have free will. If humans have free will in the classic sense, then predicting human behavior will always be limited by this. But it can be argued that even if people do have free will, people still behave in ways subject to statistical analysis. So, X% of people will freely do Y, while Z% of people will freely not. Though they are all free, the general patterns of behavior could remain predictable. After all, we already engage in effective statistical predictions and if these are compatible with our (alleged) free will, then the same should apply to other large-scale predictions. As such, psychohistory could be consistent with free will. That said, perhaps free will could be a factor that could “break” some predictions, perhaps in important ways. The “breakage” caused by free will would seem to depend on how much impact individual choice has on events in general.
A second issue is, obviously enough, whether reality is deterministic. If we live in a deterministic world, this would seem to make predictions easier (if that even makes sense to say in a deterministic universe). After all, there would be no random chance or free will to complicate matters. Of course, even if we live in a random universe then predictions would still be possible. They would lack the certainly that would be theoretically possible in a deterministic universe, but such is life in a random universe.
A third issue is whether reality can be adequately modeled. This involves concerns about the nature of reality as well as the capability of humans to develop a means of modeling reality. It seems reasonable to believe that our models will always fall short of reality, thus ensuring that predictions will always potentially be in error. There is no doubt some clever argument that shows that modeling the entire universe within the universe itself is logically impossible. It probably involves an infinite regress on modeling.
A third limit is processing power. Before computers, data analysis was done by humans, and this placed a limit on the volume of data processed and the speed at which it could be done. While modern computers lack human intelligence, they are suited to some forms of data analysis—at least once they have been properly programmed. While the industry is starting to run into the limits imposed by physics when it comes to improvements in processors, creating massive networks provides a means to work around this, at least for a while. We can always build more compute up to the point we cannot.
It is also probably impossible to build a machine with enough processing power to recreate the universe (even if it is assumed that the data is complete and accurate) even in a virtual way. As such, this will also limit the efficacy of predictions.
But perhaps someday we will be able to predict the future well enough to know whether we need to wear shades.
A Philosopher’s Blog is Now on Substack!
You can subscribe and read for free.