DO U KNOW, WHAT U DON'T KNOW?
Let's talk about neuroscience, Bayes and fairy tales. Let me tell you a tale of unexpected uncertainty and unaccountable joy of learning. The term 'unexpected uncertainty' is purely scientific and coined relatively recently by neuroscientists who share the concept of Bayesian brain (that is, in essence, the brain that is trying to predict what to do next based on a generative model of the world which it is constantly updating).
Brain Named after Minister Bayes
Thomas Bayes was a Presbyterian minister and believed in the divine order, not in unpredictability of the world. He also believed that it was not possible for a human to grasp the divine providence in its entirety simply because a human is not God. Brain does not have enough computational power, as they say now. Therefore, a human is forced to live in constant uncertainty, and in order to better navigate it, the brain must constantly calculate the probability of everything that may happen… and that may not.
Bayes’s revolution was in making probabilities subjective. He proposed to calculate the degree of a person’s (or an agent’s) belief in the probability of an event to occur instead of calculating the probability of this event’s occurrence in the outer world (as old-school frequentist statisticians do). Bayesians may base and update their degree of belief using data from frequentist statistics. In fact, they often do. Bayesians also apply standard frequentist rules of manipulating probabilities to manipulate degrees of belief.
Yet the theorem proposed by Bayes and, independently of him, a little later by the French scholar Pierre-Simon Laplace, a supporter of not divine, but scientific determinism, made it possible to perform new important manipulations by making probabilities dependent on other probabilities as conditions which may influence them. Degree of belief becomes a hypothesis that can be made conditional and verified, adjusted or even rejected with the advent of new information.
Because people in real life don’t consciously perform all the frequentist and Bayesian calculations pertaining their degrees of belief, the notion of Bayesian Angel (or Bayesian ideal observer) was introduced. The ideal observer can’t always be right. In fact, it can be entirely wrong, but it derives most accurate probabilities from the incomplete and ambiguous information about the world which is available to us. The more information Bayesian Angel gets, the better predictions for the future it makes.
Alan Turing used Bayes’s theorem to crack ciphers of German encryption machines including renowned Enigma and saved millions of lives during the World War II. The backbone of AI - machine learning is also based on Bayesian reasoning. Yet the swap to subjectivity has turned out to be so counterintuitive that some statisticians even now, more than 200 years after Bayes, don’t accept it.
Nevertheless, computational neuroscientists, having received MRI and computer simulation at their disposal, were able to establish that the human brain spontaneously (that is, without the participation of reason) uses probabilistic statistics to estimate a hierarchical model of the world just like the ideal Bayesian observer. For the fitness of such world model not to decrease, but to increase with time, the brain must constantly be fed with new knowledge. Our decisions and actions are guided by this model through the uncertainty of the world.
Calling and Unaccountable Joy of Learning
And here comes the turn of unexpected uncertainty. There are questions where to look and what to look for. “I don’t know what I don’t know!” Says the brain, “I have to know what I don’t know to find it out!” This is unexpected uncertainty. “We need to find out what the rustling in this dark room is about, what has happened there that I don’t know!” Says the brain and produces an overdose of dopamine to overcome anxiety, simultaneously being stimulated by another neurotransmitter - noradrenaline. The more unexpected the uncertainty and the larger the scale of the anomaly are, the stronger the work of both systems becomes until one overcomes the other.
The brain forces us to engage in a constant search for and harvesting of information with the help of dopamine - the universal neurotransmitter of curiosity, and not at all of pleasure, as it was considered for a very long time. Dopaminergic system is involved in so many processes in human brain that it was hard to pinpoint its unified role. Our brain derives knowledge from errors and accurate predictions, rewards and punishments alike. Yet the informational value of uncertainty is much higher than of almost certainty. The information value of full certainly is zero. The information value of unexpected uncertainty is maximum.
Our brain desires uncertainty and this lust isn't always pleasant. The “die of curiosity” metaphor conveys very well the feeling of dopamine in its purest form. Endogenous opiates are responsible for pure pleasure. If dopamine is ‘to want’, then opiates are ‘to like’. But opiates synthesized in the body, are only enough for a fleeting joy of a Facebook or Instagram like.
Yet, dopamine is also a hormone of joy because we really experience the strongest euphoria, only when dopamine and endogenous opiates act together. This is the explanation why by removing unexpected uncertainty from life and art, we are depriving ourselves of the opportunity to experience euphoria in a natural way. Not knowing the reasons why unaccountable joy has disappeared from our life, we begin to look for substitutes, surrogates, exogenous opiates, among other.
Have you heard of the opioid epidemic in the US? A couple of days ago, Congress adopted another plan to combat it. In 2015, 33,000 Americans died of overdose having taken opioid prescription drugs s alone (not including heroin and illegal synthetic drugs ). In 2016 33 thousand died. In 2017 - already 42 thousand. The damage to the US economy exceeded 78 billion dollars. In 2017, the CDC announced the opioid epidemic. Same year in Russia some 14,000 people died from alcohol poisoning. I suspect that the Russian people drink themselves to death for the same reason that Americans are swallowing handfuls of pills - trying to escape from the incomprehensible bleakness of life. You can't feel the unaccountable joy from all sorts of likes, payouts and benefits, but you can feel it from... What did you say?
How a Fairy Tale Has Become the Boring Reality
“Go there, I don’t know where, bring it, I don’t know what,” says the king to the archer in a fairy tale. And the archer leaves to look for adventures on his own a... far away. “Give me what you don’t know at home!” Demands the water monster from the soldier. The soldier agrees, comes home and finds out that his wife gave birth to a daughter while he was away. And going gets tough. We began the Humanness Learning project by searching for hidden morality in fairy tales. Looking for the secret of how a fairy tale makes us more humane, and found ... unexpected uncertainty. Not only in the above examples, although they are indicative.
Unexpected uncertainty is hidden in the sequence of fairy tale events, where each episode provides an update of the fairy tale world: real life, real life, … tenth time real life. And what is next, magic? Magic again! Why? The brain begins to update its model and explodes with dopamine with the addition of endogenous opiates as it does every time when it extracts new significant knowledge from unexpected uncertainty. The hippocampus, weared with the boredom of predictable routine, comes to life and gladly clutches at solving moral puzzles that fairy tales throws up. The characters of fairy tales are programmed as robots for binary judgment in accordance with the rudimentary baby morality, and they leave to those, who tell and who listen to the tale, to decide how they would apply moral values flexibly in the character's shoes given all the uncertainty and ambiguity.
Previously, a fairy tale had a strong influence also because it, like a virus, came to life, penetrating the living tissue of brains of the narrator and listener. Unfortunately, Disney has successfully vaccinated us against this virus. It seems to us that we know everything about fairy tales and they are not interesting for us. This is certainty. So, we need to develop new forms of artificial life, which will fill the shortage of unexpected uncertainty and return to the lives of millions of people the natural incomprehensible joy of being (of satisfying spontaneous lust for knowledge), which does not depend on the level of welfare, nor on the position, nor on the number of likes on Instagram.
We can’t program unexpected uncertainty. It can be only produced by real life (uncomfortable and dangerous) or by real art (expensive and hard to scale). Therefore, artificial real life needs to be created in order to produce plenty of comfortable and safe, affordable and scalable unexpected uncertainty to return the unaccountable joy of life to millions.
However, it takes much less time to tell a tale, than to get the job done.
I Took Chaos out of My Head! Put It Back!
After all, reinforcement learning is considered complete only when from punishment the knowledge is obtained that will allow us to avoid this punishment in the future. If the lesson is not learned, the organism or the whole species loses its fitness. Evolution is not interested in the amount of wealth accumulated by mankind, be it gold, or knowledge, because according to Bayes, knowledge must be constantly replenished, so that its updates can be used to predict the future with the accuracy necessary for survival. The fittest survive, not the richest, nor the most knowledgeable, nor the most powerful.
Fitness of a human is determined by the flexibility of mind, which makes a human capable of finding answers to the most diverse challenges emanating from the chaos of nature and from other people with snippets of chaos in their heads. Fitness of humanity is determined by its ability to provide the requisite variety of responses to the challenges of Universe, overcoming the thermodynamic limit of the computational power of the human brain by the close cooperation of diverse people, again, in a cybernetic sense. Robots do not need to cooperate. They need only instructions from an app like Uber. Automation and depersonalization of people is a legacy of the past industrial epoch that we need to overcome in order to survive as a species. It is as clear as day. But clarity is not enough. Fitness beats truth. Always. Knowing is not enough. Action is required.