More is less, and machine driven human behaviour modelling

“Sigh no more, ladies, sigh no more; Men were deceivers ever;One foot in sea, and one on shore; To one thing constant never.” William Shakespeare.

People have a tendency to attribute immutable character to a person from the slightest of situations were we can relate to each other. Our brains automatically process and assign a rating to someone we interact with, and whether we know the person or not, an impression is already with us. An acquaintance that happens to jump the queue in front of us for the train ticket is characterised with a flaw that is transferred when we later have to provide an opinion about his reliability as a trade in the business of plumbing. Surely the rush hour, or the fact that he could have been waiting to get a train ticket much longer than us —without our knowledge— can be argued in his favour. But, in my quick assessment of him—unless prompted to the aforementioned possibilities— I will just see a queue jumper. Someone who might choose to fake a quick fix of my friends toilet pipe for it to breakdown later. Even if he just plainly jump the queue that time without any justification it is not reasonable to qualify him as not trustworthy for a complete different task and situation. However, our brain does it automatically, and when we process behaviour of people, what we are really doing is creating a quick model of character that allow us to assign a permanent category to them. That in case we’ll need in the future, a sort of mental library.

In one hand, our assessments not always remain permanent. Our mind, change them with more information and we keep adjusting our perception based on it. In another hand, even for people we know and trust for long time —friends or family— a behaviour, an act or just simply a gesture might condition our response to subsequent behaviour that we routinely face with the same people in the same situation, every day. Our reaction is adjusted according to the events that happened minutes or seconds before, and is never the same. Although, we like to assign people to a character or a regular form of behaviour —”he is an extrovert”, “she is a good Samaritan”— these traits might reside less in the person and more in the situation, and even in great measure, on the surrounding group of people who influenced, directly or indirectly, the conduct. There seems to be not —even with the situations and people we know most— human behaviour consistency. Indeed, a long winding challenge for social science.

It appears to be, that the person and the situation needs to be decoupled, and even more than the definition of attributes of a person —kind, generous, evil, arrogant— is not portable from situation to situation. A general perception on the contrary is prevalent in culture “he who steals and egg, will still an ox”, “he who lies will also steal”. This attribution bias of permanency of character across domains and situations is deeply ingrained in mainstream culture psychology, and in assessments we made of public groups when referring to a situation and how an individual will be judged by it . A politician knows the value its people confers to this consistency, and hence recognises how his personal conduct in managing everyday situations—like marital affairs, management of personal assets, and even his own children personal issues— although not related and neither relevant for their qualification as a public servant, is of utmost importance by its electorate when casting their votes.

Even when imagining our behaviour in future situations —which we know we can not based in the present— people want to carry on thinking our actions will be consistent to our current individuality, morals, and values. We have total blindness in forecasting situational induced change by the interaction with a new setting, social encounter or just simply the exposure to a never seen artefact or machine. This is probably why Steve Jobs, Henry Ford and many other entrepreneurs regularly discounted —to zero— the value  of asking people how they rated the idea of products or services not yet in the market, or even interaction with prototypes. As only actual interaction counts in showing the real use, fit and problems with a product, and the influence of machine to human behaviour.

So, if our individuality is not consistent, what are the implications when designing complex automated systems that assume user predictable pattern behaviour?

Well, in a way there is no other alternative as a program has to have a beginning, and end and a “defined” middle, and this is something machines will struggle without. Still worth the question,  are we be transferring our human biases into the code that make the machine perform more like what we think our behaviour should be, rather than what it really is?

Pausing for a second. If that is true, all the efforts to mine data at the individual user level to tailor marketing messages, preferences, and even user interface to improve engagement and relevance are completely misdirected. They should not be collecting information at the individual level —it might be even irrelevant— they should be more looking at information of the setting where an interaction happens, and how changing parameters of that setting alters the human response. In a very simple example, if provided a choice between a set of personal data of your users (like gender, age, location) and another set of data that only includes changing attributes of the store or website where the interactions happen (visual cues, colours, design), the latter value should trump the former most of the times.

The actual choice, data point, or any attribute associated to an individual interaction with a product that does not also link it to the context where that result was driven from is almost useless, or just noise that distorts the real signal driving the behaviour. Many Big data solutions will be just be doing exactly that. Incrementing the data points in a vacuum from the setting they were derived, therefore just making harder to draw conclusions about the real patterns driving human behaviour.

Now, let’s imagine trying to capture, for every data point, the context where it was derived. Most aspects of it. Mood, lightning, smell, noise, recent argument with your girlfriend, etc. The data sets become gigantic pretty quickly and one can almost argue it is not closer to be more meaningful. The more you have, the less you know. More is less, when machines try to model us. 

Maybe, human behaviour might be beyond the realm of what the machine is able to compute or predict. Actually, because it was designed by a human —not even fully aware of its limitations and bias— we should not expect any different. The machine was born as mirror image of its creator, and carries at the least the same limitations— or even more. This problem is small as long as people designing the systems are aware of it, but could be big and ugly when forgotten.

Only time will tell where the future go, but chasing our tails might not be the most promising action.

Send to Kindle

Leave a Reply