Have you ever received a video recommendation that does not fit with you? What about the chat autocorrection feature that very often guess the wrong word? Did you know that in websites like LinkedIn some job postings are filtered out from the results depending on your profile (location, job preferences, etc.)? Why nowadays a random articule in the internet or a small video in YouTube have almost zero views while other content is viralized and almost everyone seems to have watched it?
I still remember how internet use to look like 20 years ago. Typically, there was a single version of a global website, same content visible for every page visitor. Instead of getting recommended content, you usually got the most visited places, last uploaded article or the most watched videos, none seemed to be linked to your browsing history.
That internet era is very far now. It was much more refreshing and liberating. Digital content for me it was a synonym of exploring that world that I did not know, to expand my interests, to go beyond my current environment. It was much more likely that the web was very different to my own world and stumble upon unimaginable things that did not happen in my daily life. Nowadays, sadly for me, the content displayed changes depending on predictions based on your location, language, browsing history, etc. Internet once was an open window to the immense world, but it has just become a mirror to see ourselves.
Many of the tools that enable the new recommendation systems and algorithms rely heavily on machine learning (ML) which is built on statistics and probability, which in turn are based on important principles such as law or large numbers (LLN) and central limit theorem (CLT). Avoiding technicalities, the LLN says that a random event repeated a large number of times will converge to the expected value. These principles are great and they have enabled an amazing world with ever more accurate predictions from traffic, delays, weather, etc. However, when the subject of the prediction are human decisions, it is necessary to understand the profound impact that they may have.
In humans, I would like to take the assumptions of law of large numbers (LLN) with a pitch of salt. First and foremost, homo sapiens has an incredible genetic variability between individuals, not even monozygotic twins will share the same genome. On top of that, individuals also differentiate by the epigenome, selected parts of the genome will be suppressed or activated due to the environmental factors. Secondly, the myriad of circumstances that affect the life of each person is subject of the theory of caos: a ridiculous tiny change in the initial conditions will affect massively the outcome of the human action. Therefore, the assumption of the law of large numbers (LLN) that the same experiment will be repeated is highly disputable. Tossing a coin several times could be assumed to be a repetition of the same experiment, but observing the decisions that an individual is taking will never be the same experiment, nor we will know all the factors that led to the decision (both deliberate and/or subconscious).
Even it free will did not exist, we are still very far away of understanding how our brain takes decisions and it is doubtful that we will be capable of tracking the factors to predict human actions purely link to a single individual (considering a person an independent event with respect to other people) with an acceptable level of error
That was not an argument to destroy the usefulness of machine learning (ML) applied to human decisions, but rather a caveat of how complex this science could be when applied to humans and how prone to error are the predictions of much more complex phenomena.
Julian Assange holds that while human freedom is at risk by mechanism of control through vigilance and data collection, the freedom is being still preserved by more complex and unpredictable societies, which have become much more impredecible than ever, thanks to internet and smartphones despite the fact that we are more trackable than ever. Julian Assange points to the Trump’s victory in the US elections and Brexit referendum among the major surprising cases.
Nonetheless, predictive error is a small part of the story. I believe that the predictions themselves are a dependent variable, they also influence us and have the potential to change our lives. Then, another major concern is the loss of variability and diversity.
If everything becomes a prediction based on the average or historic records, then for the future actions, recommendations or predictions there will always be a bias towards repetition and gregariousness (replication what others do). This is equivalent to the application of a tax to the uniqueness and novelty. For example, even during the composition of this article, I had to fight against the autocorrection feature whenever I decided to use a less common word (i.e. I had to reconfirm that I wanted to use the word). Moreover, this also reduces the amount of unrelatable input (content that prima facie is not correlated to our preferences) that we receive in the digital world. This rather than expanding our knowledge, reaffirm and reinforce our preferences. Therefore, it is less less likely to develop creative, unique or novel content, for the reason it is likely to have a similar output from similar inputs.
The individuality factor is suffering in exchange for more traceability and enhancement of advertising. Not only our world is becoming increasingly digital, but also the virtual world is filtered out to match our “predicted profile” with abilities to block the content that we don’t want to see or hear. What are the consequences of this for patience, tolerance, empathy or curiosity?
In the physical world, the most enriching experiences I ever had are these that I could have never imagine, beyond my expectations, that helped to modify to a certain degree the person I am. Human reason is incapable of comprehend the richness of the world in every sense of the term, therefore life is full of surprises. In this sense, accommodating the world to our reason will be a reinforcement of our prejudices and limiting our life to an ever more limited life.
I believe this is a major factor behind the upward trend of narcissistic personality disorder (NPD), which rose to 80% from 12%,1 as it is more difficult than ever to exit from your individual self. In addition, I would expect all gregariousness to be also on the rise, since being unique seems to be more taxed than ever before.
These predictions of human actions are broad brush strokes. Not be the right way of painting a realistic masterpiece, but with the aggravating factor that these broad strokes are also shaping the nature of our society. If this trend accelerates even more with AI, I worry about our future society and individual freedom.
Rincón del logos 