Following the correlation and isolation model for general phenomena, the contributors to the future based on human behavior are:
– The forced behaviors caused by active discipline, confinement, and other constraints (usually > 90%; most insensitive to situation/least isolated)
– The daily routines within the situation (usually > 90% correlation; some situation dependence)
– Rational behaviors (e.g. following clinical protocols, investment decisions made by professionals) (depends on the situation but when well executed, well over 50% correlation; high isolation/situation dependence)
– Frequent behaviors (e.g. hobbies, usually > 50% and in many cases more correlated than rational behaviors and less isolated, however x+1 transitivity is much more limited than with conforming rational behaviors)
– Infrequent behaviors (usually when trying things out or making big purchases, contributing at low correlations to exact time, potentially a lot less isolated, but x+n is poorly predicted)
– One-off and random behaviors
In the less-correlated behaviors, for “predictably signaling individuals”, the signals then predict x+n; how big is n, and how correlated/isolated, heavily depend on the individual and are not generalizable. Such an estimate is fundamentally (metaphysically and otherwise) uncertain (explanation below), however the below points provide some limited guidance for the bounding of the CEP to approximate the future in most cases to the extent that it is feasible/effective:
– Assuming sufficient means of course, n signals accurate in m result close to 100% of n, predict n+1 signal will result in m+1 correctly predicted at some squishy percentage above 50% (note if the actions really are frequent, they would be binned with behaviors and not in signaling, they could not be functionally distinguished to increase prediction accuracy since any contrary claims would be incredible). The most common case: if someone lives up to their word or doesn’t live up to their word, the prediction is in that direction.
– The likelihood of the action following individual’s previous performance decreases with the situation change, complexity, cost and the difficulty increase from previous situations in the evaluation
– The correct estimation methodology would use the performance of previous individuals who displayed similar signaling in similar situations, however the issue with the less frequent behaviors is that signaling often is weak (not high correlation anyway even if accurate) and the situations are not comparable
– Any statements about influencing others are subject to the failure rates of the persuasion in question; therefore an individual’s signaling only credibly refers to the type, number, and quality of attempts performed. Claims beyond that, lacking coercion or other violent action, make an assumption about the success of the persuasion and so must be discounted.
– Over time, the predictive value of the past signals and performance decreases
Hence, the effective decision tree associated with an individual either is to deem them as “predictably signaling”, a construct that only has validity in relatively simple, unstressful situations, or to monitor their actions and be prepared to remove them from your service.
There is one significant exception: the class of situations where you are highly resourced, and making the correct decision has a large payoff. In this case, using “conventional attribution” (popular theory of intent) is a worthwhile supplemental prediction approach.
—
Macro Explanation:
In some senses human beings have very predictable behaviors; they grow and decay, consume and expel, and follow a series of common behaviors. The animals also migrate and have daily routines, but the humans also migrate and have daily routines; as pastoralists, humans migrate in the same way as many animals. Likewise, when humans are confined and forced to work, they work in the same way as the work animals, occasionally resisting, but on the whole generally doing what they are forced to do.
We also should consider many phenomena that don’t require anything other than a frequentist approach, i.e. “it happens often in the past, so it will continue happening in the future” such as:
– People going to work, eating, sleeping, and otherwise going about their daily business
– Young people falling in love and starting families
– The charity of both young and old
– People following the biological path of youthful vigor and ignorance, followed by middle-aged and early elderly relative wisdom, followed by physical and mental decline, followed by death at a population level, and which for a particular individual also apply most of the time, assuming there is some past data.
These two (of course there is some overlap between the two) make up the first two categories, which is the bulk of the human-related future.
In regards to the scope of rational behavior:
We should consider situations and disciplines where some humans exhibit high degrees of conventionally-modeled rational behavior, and where differences in the level of analytic and predictive capability can decide the outcome, such as:
– Poker games
– Wall Street financial trading, and investment activity generally
– War
– Politics
– Local optimization where the participants are highly experienced
We see these practiced on a regular basis by certain individuals.
Likewise, we should consider situations where humans exhibit conventionally-modeled irrational behavior at a much higher rate:
– Under physical stress like hunger and thirst
– Being tortured or threatened
– Poverty, or heavily indebted
– Extremely high levels of activity and task-switching
– Confronted with large amounts of information in situations where the interpretations are complex
– Inexperience
The rational behaviors exist in high-isolation, controlled environments, both because the regular phenomena the rational actors wish to exploit exists in those situations, and because the work to behave rationally has prerequisites and support requirements that effectively confine its application to the controlled environment.
I will not repeat the many situational studies on specific instances of these; there are numerous market research, historical, and economic papers and studies detailing the various gradations and extents of rational behavior.
As for the classification of the remainder of behaviors, since no high correlation is claimed, there isn’t much to refute, but they would be considered as suboptimal characterizations of the historical record and possibly there is some room to improve them via subdivision and reclassification.
—
Explanation Of Predictably Signaling Individuals
As per the metaphysical discussion (on leifpowers.com) of attribution and measurement of my own mental state, I reached the conclusion that directly correlating my personal feelings of intention, my optimizing feelings, my words, actually optimizing behavior, and actual actions was fundamentally difficult. I realized that such correlations were nowhere near 100%.
When predicting the behavior of other humans, I don’t have the advantage of being able to measure feelings; I only have their actual words and actions. Therefore, with less data, reliable prediction naturally becomes more difficult in the perceived average case.
Against that, we have to consider the ordinary conduct of general managers, miscellaneous business, personal assistants, and family members who help out around the house: they perform a wide variety of tasks, some of them infrequent, and they also signal their intent to complete. Clearly there is some characterizable phenomenon here.
To say our conclusions are not generalizable across humans: on what basis would you generalize: sex; race; social class; wealth and income; geographic region? Even nominal religion can’t be relied upon as a signal: note the alcoholic Muslims and the pro-choice Catholics. Normally we consider an affiliation with certain groups such as a military or social work organization as indicative; but then there are mutinies, fleeing the battle, molester priests and CEOs. That affiliation does not suffice to permit the type of trust we demand from a predictable signal, since we still require many or all of the accountability mechanisms as if we did not label an individual in this way.
Likewise, betrayal is the realization that such a designation is not 100% correct.
One example of the decrease in predictability is promotion or job change generally: a fully reliable signal would have the similar intent would have similar results, but that’s not what we see.
One example of the lack of correlation of signal vs. performance over time is the increase in competence/skill with practice; another is the decline in competence associated with lack of practice and/or increasing old age.
—
Explanation of Exception to Writing Off Predicting Individuals Who Aren’t Predictably Signaling
All I mean by “conventional attribution” is the popular theory of intent, which is a Bayesian model where:
– The states of a person’s mind are an unknown quantity
– Through the observation of external variables related to a person’s words and actions, this quantity can either be directly mapped, or at least confined to a small number of possible states
– The states of a person’s mind predict their future behavior with high accuracy
This model works extremely well in situations where frequentist interpretations also work well. It works well in situations where the parties are not intellectual and do not have the resources necessary to invest effort into communications strategies, or where their actions are easily visible and verifiable.
Naturally, this model is not useful in situations where:
– People just lie
– People are aware that others are using this model, and are willing to modify their behavior to manipulate their conclusions
– There is not enough observation of the variables (evidence) to narrow down the possible conclusions
– There is enough mass of evidence, but the evidence contradicts itself
In practice, conventional attribution doesn’t work in any negotiation, complicated interactions with individuals, any situation of war or conflict, or any situation where a social expectation of a certain standard of behavior has been set.
In negotiations, each party has the interest to misrepresent the situation in order to convince the other party that they have extracted the maximum.
With complicated interactions, mistakes and communications confusion alone will cause conventional attribution to fail.
In situations of war, the parties are willing to kill each other, so if they consider their situation rationally, they will realize that in comparison, lying isn’t a big deal. In many of these cases, their words and actions are so negatively correlated that it is easier to predict the future via inversion. However, the more skilled actors seek to frustrate conventional attribution through a mixed record of honesty, selective transparency, and deceit.
When there is not enough evidence, numerous interpretations are valid, so conventional attribution can’t work regardless of whether the particular techniques used are otherwise valid.
Likewise, with contradictory evidence, both the affirmative and negative hypotheses are only consistent with a subset of the evidence, so conventional attribution cannot settle upon a rough state of mind.
Depending on the society, these could constitute the majority of all events and interactions between people.
Practically, you consider the matter uncertain, hedge and hope except in this type of situation:
– I would still recommend the use of a conventional attribution model as a supplement to these strategies in situations where you have a large amount of resources to invest in influencing others’ behavior towards an end that will pay off if your manipulation succeeds. For example, influencing the German Third Reich armies to station their forces more towards the wrong landing site on D-Day was correct not merely because it worked, but because the Allied forces had large amounts of money to invest in attacking the German intelligence efforts, in decoys, and in propaganda, towards an objective that could ultimately save hundreds of thousands or millions of lives if were successful. Hence, even a 10% improvement in success rate would easily justify even a large up-front investment based upon the conventional attribution model.