Recent Paradigms in Predicting and Controlling Human Behaviour
Confusion Worse Counfounded
Those of us- as humans or scientists (the conditions may be- or may not be — mutually exclusive) who are interested in identifying and predictively modelling the behaviour of human beings as a path to so-called Artificial Intelligence and a ‘better’ society seem to have a choice between working back from painful personal experience or establishing correspondences between some type of neural activity profile as a series of cascading neural ‘firing’ events- and an observed ‘behavioural’ event. This latter is a correspondence argument which seeks to establish a predictive relationship between two sets of event-based variables.
The first of these event sets is the firing of individual neurons and neuronal layers in either an organic or mechanical brain and their relationships. This is tricky for a variety of reasons, some of which I will set out below and in following papers.
The second is a set of physical and social ‘behaviours’. At perhaps its mechanically most common we see it in the elicitation of pre-recorded phrases in the Cloud which can be accessed by Cortana, Siri, Alexa or Google Assistant and used to respond to questions. Or in the form of GPT3 which can assemble phrases and sentences, identified from some type of corpus, to create documents which read as if they have been written by a human being.
Although this paradigm is a fairly recent one, it is dogged by an inability to understand and reproduce the neural states that both generate most mental “events”, in humans and how these relate to the behaviours that may follow. This frustrates people working in the field and disappoints those who support them.
Of course, people have been studying, explaining and predicting human behaviour for as long as humans have existed. In all fields of human endeavour, the ability to carry out these two tasks is a good predictor for both survival and growth. Humans are- much as we may deny it- a social and physical predator. And the ability to tell which way a predator will attack supports avoidance or effective response. While the ability to lie provides a countermeasure against this strategy.
This paper considers some of the reasons why understanding and predicting human behaviour is so difficult.
The first lies in the nature of the tools we use. Tools such as cognition, affect, conation often operate in ways and at levels of which we are not consciously aware. We connect data based on rules regarding properties such as similarity, contiguity, salience and vividness This means that we are often ‘data-driven’ rather than interpretation driven.
The second lies in the nature of the source of our access. For much of human history a person’s beliefs about the properties of nature, society, and the self were based on personal experiences, conversations with others, and pronouncements by ‘experts’- often figures of power. The first source of information necessarily occurred in a specific context; the latter two typically implied a context. A human telling another human about a risk or reward usually contextualizes the function of sentences such as, “I saw a big wolf near the river bend” or “There is a good crop of tasty mushrooms in the far field.”
The third lies in the general- but increasing- fluidity of the world, and the necessity to process this fluidity accurately and usefully. One approach can be seen in the use of jargon which excludes with a special vocabulary. Such vocabularies challenge the validity of traditional beliefs based on experience and conversation. The fragmentation of pronouncements by respected authorities on topics ranging from health and crime to the age of the universe and the origin of life, is dependent on particular classes of evidence. But the validity of many if not most inferences is limited to the conditions under which the observations were gathered, until someone demonstrates their generality.
The fourth lies in the fact that context includes not only the actual- physical- perceived features of the situation in which data are collected, but also the procedure, the collection of incentives presented, and a variety of properties inherent in both the subjects and the observer. When the subjects are humans, the language used to communicate with them is part of the context. Each context is associated with a set of probabilities assigned to the collection and attribution of probable outcomes. The context also selects one response from a larger set of possible alternatives to that incentive.
These reinforce or mute the effects produced by internal context in a looping structure which drives evolution. And stagnation.
There are a number of ways in which we ‘handle’ the issue -reconciliation-of internal and external contexts. One of these is to suggest the existence of mechanisms that predictively ‘model’ the external context that enable us to -say- intercept a baseball which has been hit over our heads. To complete this task we have to assign values to the baseball in terms of velocity, direction, drag, and other factors. We have to forecast these factors to arrive at the probable point that the baseball will reach, whilst coordinating our own activities to reach that point and catch the baseball. This is no mean feat of mathematical calculation, performed by people who often have no mathematical training.
And whilst prediction is clearly critical in the management of risk -avoidance is generally much less energy expensive than engagement- there is no reason why we should not creat other types of model such as control, explanatory or attack-defence models.
And while the neural system continually primes the neurons that ‘normally’ respond to the event that is expected to occur in the next moment. This preparation facilitates its detection. Usually, the event anticipated is the one that occurs. On the less frequent occasions when the expectation is violated, the mind overlay responds. Every mind profile, therefore, is a hybrid response to the event as perceived and the response to the event that was anticipated. If the event does not occur, the mind’s response to the violation of expectation becomes part of the reaction to the event that occurs. This fact constrains inferences about psychological states that are based on mind reactions to unexpected events gathered on human subjects in BOLD MRI scanners.
Prediction affects many phenomena. Humans displaying traits or performing actions that violate community expectations (‘norms’ which offer predictive support) are likely to be rejected or harmed. Those who are ‘just like us’ are welcomed and rewarded with our self/other confirming attention.
Humans thus occupy a narrow niche defined in the gap between certainty and uncertainty. This ranges from the pleasure we experience in reading books or watching videos to ‘confirm’ our predictive models that the ‘butler did it’ to the terror of chronic unpredictability regarding whether we might be infected in a pandemic. Or, if infected, whether we will die horribly.
And of course, this niche can be manipulated in fields other than entertainment and social relationships. Politics is another area in which our need for certainty may overwhelm our ability to cope with uncertainty. Populism- seemingly on the rise again, seems to provide a much higher level of predictability in which strategies that lead to civil unrest can be replaced with greater certainty. Many humans living in chaotic communities may be willing to give up their personal freedom in exchange for more predictability and- the illusion of- control.
A large number of investigators studying brain–mind- behaviour relations resist Niels Bohr’s insight that the validity of every conclusion is depends on its source of evidence. Two statements referring to the same observation may have dissimilar validities if they originated in different kinds of evidence. The validity of estimates of the heritability of intelligence based on equations whose values were the degree of behavioural similarity among the biologically related members of a family does not correspond to the validity of the considerably smaller estimates derived from similarity in genomes.
And, of course the concept of ‘big data’ is a windfall which data starved scientists — and businesses- have turned joyously towards after years of not knowing their research subjects and customers very well at all. Now, ‘guided guessing’ or big data analytics has become a major industry in which people’s online ‘behaviour’ is also predictive of their offline or aggregated behaviour. Of course, how indicative ‘pointing and clicking’ — or visits to particular web sites — are reliable predictors of human behaviour online or even- in some form of ‘real’ world — is open to question.
Clearly, website selection, pointing and clicking are quite a small subset of human behaviours. And although the data gathering methods used can also be developed to -allegedly -detect human “feelings” through “sentiment analysis”. This leads us to another problem with both ontological and epistemological effects which reinforce the challenge of human gullibility. Because new theories involve both theory producers and theory consumers. And the critical symbiotic relationship between these two parties may well be broken.
We hope to explore the issues that arise as a result of this paper in more detail in the next papers we publish. We also hope to set out a systemic approach to addressing these issues.
Thank you for your attention. Oh, and don’t worry about the Blake image. I am not the Angel of Revelation. You are ;-)