Framework for Predictive Systems
Predictive systems can be classified based on their types of inputs. The Framework breaks down these inputs into four broad categories: “random,” “randomized,” “human,” and “deterministic.” Systems described in Diviner’s Guide use a variety of these inputs, but they rarely go beyond “make prediction,” in the cycle of improvement shown to the right. The ideas of “evaluating accuracy,” and “making changes” to a predictive system are relatively modern concepts that many historians associate with the origins of "modern" science.
Inputs stem from naturally-occurring random processes, without any systematic predictability and without any intervention from human beings; sometimes also called “spontaneous.”
Example: Roman Bird Augury
The first step in augury begins with observing the flight and/or behavior of birds. Augurs need to wait patiently for spontaneous flight or cries of the birds being watched before they can begin the process of making a prediction.
Comments and Caveats:
Diviners must wait for unscheduled events before making a prediction about the future.
One can always debate just how “random” any system really is. For example, birds’ flight patterns are well known to correlate with weather patterns, magnetic fields, and other natural phenomena. For the purposes of this Framework, though, we classify systems based on how they were or are implemented by their users, e.g. from the perspective of Romans, the flight and calls of birds were entirely random
Input is the result of a process that a human initiates for the express purpose of producing a random outcome.
Example: Casting Lots (e.g throwing dice)
Tossing objects that roll to rest on a random face. Human action is needed to initiate the process, but the side of the bone where the bone (die) comes to rest should be random.
Comments and Caveats:
There can be fraud or biased outcomes if input devices do not actually produce random outcomes (e.g. weighted dice).
Dice-like systems have been used for millennia, e.g. “astragaloi” made from the astragalus bones of various animals, favored by the Greeks.
Systems like “Ifa” (which involved small objects being moved around on a shaking tray) are similar to throwing dice in that essentially unpredictable physical processes intervene to randomize a diviner’s selection.
Inputs that come directly from the diviner or, at the very least, inputs that come from a source not discernible by anyone besides the diviner.
Example: Utterings of Possessed Persons (e.g. The Oracle of Delphi)
A woman, purportedly possessed by supernatural forces, responds often ambiguously, to a question.
Comments and Caveats:
Prediction systems based on human data can be conceptualized as the diviner being a sort of “spokesperson” of the divine.
Inputs come from observations of any process thought to be repeating, predictable, or in some way knowable in a consistent fashion.
Example: Measuring the Positions of Planets
While ancient astronomers may not have known the cause of planetary motion, they knew the positions of planets followed standard, repeating, paths with respect to the background of the stars.
Comments and Caveats:
One must be very careful to distinguish “non-random” inputs from an end-to-end “deterministic” predictive system. The latter (a fully deterministic predictive system) is exemplified by physicists’ modern understanding of the theory of gravity, which today is so good as to be able to predict the position of a distant comet well enough to land a spacecraft on it. Meanwhile, astrology has, and still does, rely on deterministic inputs (e.g. positions of planets) to make non-deterministic predictions, based on humans’ reading of mystical portents.
We think of most modern science as being deterministic (and, indeed, modern science is based on the premise of being able to prove claims falsifiable or not based on repeatedly doing the same experiments and getting the same results). However, as we see in this course, most of the systems in the Diviner's Guide do not fall within the category of deterministic, with a few notable exceptions: astrology serves as a bridge in that it was not originally deterministic, but which astrology did contribute to the growth of the field of astronomy and, fascinatingly, Babylonian haruspicy also eventually developed into a deterministic system, as we will explain in more detail in that section.
The box in the diagram labeled “predictive system” can be thought of as the core instruction set (or “algorithm”) for making predictions using a particular system. The simplest possible example of such an algorithm would read: “If you observe ‘A,’ predict ‘B.’” As in, “if you roll a four on a die, predict the querent will meet her future spouse tomorrow.” Almost no predictive system is quite that simple--and many rely heavily on the expertise and interpretive skills of the diviner.
We discuss several different prediction systems in Diviner's Guide, but hundreds(!) more can be discovered on the Methods of Divination Wikipedia page.
In more “scientific” predictive systems, we can clearly understand how observed inputs come together to create a predictive system. For example, with Newton’s famous equation, F = ma is a predictive system: if we can observe mass and acceleration, then we can predict what force will be.
Unlike Newton’s equation, Diviner's Guide systems are sometimes a bit unknowable at this stage -- someone observing from the outside looking in cannot always know how exactly the predictive system really “works.” In some cases, there is mysticism inherent to the prediction system (such as with human-based prediction systems like Egyptian Diviners). Sometimes, the system appears unknowable because it has a high-level of complexity that requires years of training to fully understand it (such as with Ifa). Finally, we might not know the details of how exactly the prediction system works because that information has been lost to history (such as with Chinese oracle bones, where we don’t even know for sure if the diviners were interpreting the shape or the sound of the crack in the bone).
Observation of data is the first step to making and analyzing a prediction. We can think of this as the “input” that comes from the prediction system and is fed into the “algorithm” of the framework. In ancient prediction systems, this could range from the flight of birds to how a bone cracks, while in modern prediction systems, this can range from the observation of a planet’s location to the results of a survey asking someone who they intend to vote for in an upcoming election. In theory, the more observations made, the stronger a prediction system can become, as is discussed in the section Make Changes.
Once the observed inputs are processed through the predictive system, the practitioner will make a prediction. The definitiveness and clarity of the prediction varies from system to system. In the case of Newton's F=ma, there is no “wiggle room” in the prediction: the value of force will be as precise as the observed precision of mass and acceleration. In contrast, many prediction systems in Diviner's Guide incorporate ambiguity, intentional or otherwise, such as the (possibly apocryphal) stories about the Oracle of Delphi making predictions that can be ironically misinterpreted by the Oracle’s petitioner.
The next step of the framework is evaluating the accuracy of the prediction. The question here appears simple: was the prediction correct or not? Often, however, there can be a deeper level of complexity since often there can be confusion when the prediction was accurate, but the process (e.g. the prediction system itself) was not based on accurate facts. For example, when studying the cholera outbreaks in the 19th century the miasma theory appeared to be accurate and even had convincing data that showed a correlation between elevation and cholera. Unfortunately, they did not know they were confusing correlation with causation: individuals at lower elevations were getting cholera not because of miasma, but because water at lower elevations was generally more contaminated with fecal matter from the Thames.
The dividing line between Part 1, Oracles, Omens, and Prophecies and Part 2, The Rise of Theory prediction systems seems to be whether or not they evaluated accuracy. Once prediction systems become more “scientific” we see an increased emphasis on empirically measuring whether or not the prediction was accurate. Again, using John Snow as an example, we can see where Snow’s predictions about cholera being spread by water were later supported by evidence found during the Broad Street cholera outbreak and also by Snow’s larger study of London’s water companies. Today, in many fields, this process of evaluating accuracy has been standardized and is a central component of the prediction systems in Part 3, Modern Simulations.
(There are some rare exceptions in Part 1 prediction systems that seem to incorporate evaluating accuracy: for example, inscriptions on Chinese oracle bones sometimes include if the prediction was accurate or not. What we don’t know if this then flowed into the final step, Make Changes.)
The “final” step in the framework is to make changes to the prediction system based on the accuracy evaluation. This is more of a recursive step than an ultimate goal -- an idealized prediction system is never finalized, but is constantly making changes based on any observations that show imprecision in the prediction.
Similar to evaluating accuracy, making changes to prediction systems is not normally a feature of Diviner's Guide systems. However, it should be noted the historical evaluation of almost all of these systems is undocumented (e.g. there is not a clear written record covering hundreds of years that describes how Aztec rituals came to be in their final form prior to the Spanish conquest), thus it is possible many of these systems changed and evolved throughout their existence, though not necessarily by evaluating the accuracy of their predictions. This is in contrast to many modern day simulations which are based on constant, never-ending iterative learning and changing in order to constantly get a more accurate prediction.
One final note: we do not mean to be dismissive or condescending to systems in this part of PredictionX because they do not go through the stages of Evaluate Accuracy and Make Changes. While it is true prediction systems have progressively improved in validity and at making more accurate predictions (though not always in a perfectly chronologically linear way), there are other purposes to prediction systems besides making accurate predictions. As seen when we go in-depth on each system in the Diviner's Guide, prediction systems serve many crucial purposes: the passing of specialized knowledge from mentor to apprentice, the building of a sense of community, and the legitimation of power are just a few of the non-predictive purposes of the systems we discuss in Diviner's Guide.