Game theory is experiencing a renaissance driven by the evolution of AI. What are some classic and new ideas that data scientists should be aware of.
Game theory is one of the most fascinating areas of mathematics that have influenced diverse fields such as economics, social sciences, biology and, obviously, computer science. There are many ways to think about game theory but one that I find really helpful, although overly simplistic, is:
game theory is probabilities with incentives
Games are playing a key role in the evolution of artificial intelligence(AI). For starters, game environments are becoming a popular training mechanism in areas such as reinforcement learning or imitation learning. In theory, any multi-agent AI system can be subjected to gamified interactions between its participants. The branch of mathematics that formulates the principles of games is known as game theory. In the context of artificial intelligence(AI) and deep learning systems, game theory is essential to enable some of the key capabilities required in multi-agent environments in which different AI programs need to interact or compete in order to accomplish a goal. (A Crash Course in Game Theory for Machine Learning: Classic and New Ideas)
Both emotions and facial expressions are ancient instincts.
Human language and cognition have grown well-beyond ancient instincts. Grown beyond but also still affected by.
We have become more complex.
Today, we not only read instincts, we also read instincts into other people’s cognition through what they say, how they say it, how it sounds, how their faces move.
Which micro-expression is the right one?
The truth is we don’t know. Our readings of facial expressions in real-time, real-world situations are often wrong, often tragically.
Our cognition has advanced beyond our instincts but generally speaking it has not advanced far enough for us to generally recognize this fact.
Cultures and social groups deal with the ambiguity of facial expressions by being formal, wearing masks, emphasizing “face” or “saving face,” promoting respect or strong egos that can sell themselves through assertion of meaning, Botox, makeup, boobs, etc.
I think it is arguable that many/most/all people take on and use religion or philosophy in order to provide themselves with a generalizable set of emotions and facial expressions that can be employed in many situations. In this we can see how the architecture of our cognition (our philosophy/religion) is connected to our emotions and facial expressions.
Obviously, our reading of other people’s faces and emotions is not always wrong. If it were we wouldn’t do it at all. But our readings are wrong often enough that tragic mistakes are frequently made.
It is a pity that these truths are not more widely recognized. Browse almost any psychological forum and you will find many comments concerning the anguish people feel at having a condition that is widely misunderstood or misread.
At least they know what is going on.
This morning I saw this article: NEVER trust a person’s face: Scientists say it is ‘completely baloney’ that you can read people’s emotions from their expressions.
And that led me to search for this paper: Emotional Expressions Reconsidered: Challenges to Inferring Emotion From Human Facial Movements.
I am sure most, if not all, psychologists recognize the basic problem of our poor abilities at reading emotions, tone of voice, gesture, and even what we mean at all when we speak and act.
Does anyone know what to do about it?
Global workspace theory is a description of how our minds work. The word global refers to the whole mind or brain, not the world.
The central feature of this theory—the global workspace—is conscious working memory, or working memory that could be made conscious with minimal effort.
This global workspace is also what a great deal of Buddhist mindfulness attends to. If we focus our attention on what is coming in and out of our global workspace, we will gain many insights into how our minds operate.
The Buddha’s five skandha explanation of consciousness can be understood as a form (or percepta) entering the global workspace.
Consciousness is the fifth skandha in the chain of skandhas. It is very important to recognize that whatever we become conscious of is not necessarily right.
With this in mind, we can see that being mindful of what is entering and leaving our global workspace can help us forestall errors from forming and growing in our minds.
In the Buddhist tradition, ignorance (a kind of error) is the deep source of all delusion.
But how do I know if the percepta or bits of information entering my awareness are right or wrong?
Well, there is science and Bayesian thought processes to help us, and they are both very good, but is there anything else?
What about my actual mind? My psychology? My understanding of my being in the world? How do I become mindful and more right about these?
Besides science and Bayes, I can ask an honest friend who knows me well if the percepta I think I just received from them is right or wrong.
If my friend knows the game, they will be ready to answer me before my global workspace changes too much. If my friend confirms my interpretation of what they just did or said, I will know that my interpretation (or consciousness) is correct.
If they disconfirm, I will know that my interpretation was incorrect, a mistake.
This kind of information is wonderful!
We calibrate fine instruments to be sure we are getting accurate readings from them. Why not our own minds?
This kind of calibration can be done in a general way, but you will get a general answer in that case. If you want a precise reading, a mindfulness answer, you need to play the FIML communication game.
The game linked below explains some basics of game theory and also some basics of why FIML practice works so well.
The game can be found at this link: The Evolution of Trust.
I highly recommend playing this game. It takes about thirty minutes to finish.
For the first part of it, I was only mildly interested though the game is reasonably engaging.
When it got a point where communication mistakes are factored in, I sat up and took notice.
The game is a very simple computer model of some very simple basic choices human beings make all the time. Without giving away too much, even this simple model shows something I bet most of us can already see.
And that is: zero-sum games do not give rise to trust. Win-win games do.
What was most interesting to me is the game also shows that communication mistakes foster trust if there are not too many of them.
Accepting mistakes in communication requires trust. Mistakes happen. When two people accept that in each other and in themselves, trust grows.
This is a very important point and a foundation of FIML practice.
In fact, I would say that mistakes foster trust even more in FIML than other communication games. This happens because in FIML mistakes are isolated in such a way that they can be fully recognized and understood for what they are.
This provides a method for solving immediate problems while also building a foundation for the inevitable occurrence of future ones. Moreover, the kinds of mistakes people make become less stupid.
In many respects, the game of FIML is largely one of recognizing communication mistakes or potential mistakes as soon as they arise, within seconds of their onset.
By doing that FIML shows us how our deep psychology is actually functioning in real-life. Multiple insights into this aspect of psychology are transformational.
To be very brief, Karl Friston’s “free energy principle” says that the brain is an “inference machine” or “prediction machine” that uses Bayesian probability reasoning and is motivated to act by an inference seeming not true or “surprising” to it.
The free energy principle is a straightforward way to explain what FIML practice does, how it does it, and why it works differently than any other form of psychotherapy and in many significant ways why it works better.
A psychological “complex,” “neurosis,” “personality disorder,” or “persistent thought,” call it what you will, affects human behavior by being or having become a nexus of thoughts, ideas, perceptions, feelings, interconnected neurons and chemistry.
The same is true for any personality trait or skill, including very positive ones.
In Friston’s free energy terms, the psychological elements described above are surrounded by Markov blankets.
That means they are isolated or protected systems with their own variables. These protected systems (protected by Markov blankets) are hard to change because they have their own sets of rules and habitual inputs and outputs.
And that makes them stubborn candidates for most forms of psychotherapy, especially psychotherapy that requires a therapist. One reason for this is time & expense. A second reason is it is difficult for the patient to change without therapeutically experiencing for themself the complex or trait in real-world situations.
The key here is therapeutic experience in the real-world of the unwanted trait or complex that requires change.
The third reason most psychotherapies are ineffective is very subtle incisiveness in real-time is needed to penetrate psychological Markov blankets.
What FIML does is penetrate the Markov blanket enshrouding a complex with a series of small pricks. Each prick in the blanket is small, but each prick also allows some of the valence (gas) inside the blanket to escape.
FIML slowly punctures the Markov blanket with many small pricks, eventually causing it to collapse.
Once it has collapsed, the energies that were trapped inside it can be used for other things. In this way FIML optimizes even non-neurotic psychology by removing pockets of inefficiency held within psychological Markov blankets.
By using only small pricks to penetrate Markov blankets, FIML allows people to gradually and painlessly see what needs to be changed, why, and how to do it. Since FIML works in real-time real-world situations, even very small insights can bring about large changes.
The following are some basic rules for a practical behaviorist approach to speech.
Use real speech in Real-World, Real-Time (RWRT) situations.
Keep it simple by using only two people. Make it deep by using the same two people for years. No third person is needed or wanted.
Use only good data that both speaker and listener can agree on. For RWRT speech, this means only speech that is/just was in the working memory of both partners.
That is, both partners must agree on what was said and heard. If the listener heard “boo” and the speaker agrees they said “boo”, that is good data.
Partners must reach that agreement while keeping intact the contents of their working memories when the word “boo” was said/heard in RWRT. (This takes a little practice but is not that hard to do.)
Both speaker and listener can now analyze that data by discovering (through speech) what was in their working memories when they heard or spoke “boo”.
These simple rules bypass predetermined thoughts about what we believe we are saying or said, believe we are hearing or heard.
In so doing, these simple rules lead us gradually but very significantly away from surface speech (see: Time pressure encourages socially acceptable speech) to much deeper and more accurate communication between partners.
And this deepens partners’ sense of who or what they are across all domains.
That such simple rules can deeply change how we speak and hear and how we think about ourselves and others and how we understand the entire enterprise of human psychology, shows that—you might say—God exists, or the Buddha Mind exists, or profound other realms are available to human beings, that a deep sense of karma is real, that what we say matters, that not doing our best to speak the truth is what the Buddha meant by “frivolous speech” and he did so for a good reason because enlightenment itself lies thataway.
I think it’s delightfully paradoxical that simple behaviorist rules can lead us to having religious experiences.
Can we restate or add to Cogito ergo sum by saying: Recte loquendo Deum esse demonstramus? (By speaking properly we demonstrate that God exists)
Private language—what we say to ourselves, how we cogitate while alone—is greatly dependent on public language, that which is readily understood by many.
In fact, private language is so dependent on public language, it can be argued that a private language completely divorced from public language cannot exist.
It is obvious that anyone wanting to influence or control large numbers of people will address them in public language.
It is less obvious, that those same people frequently will also seek to change the public language itself.
Sometimes this language changing is a good thing as that is how civilizations adapt and grow. It is probably best, or usually best, when civilizational changes arise organically from the whole society or from important parts of society that are behaving honestly.
Sometimes, however, the changing of public language is done dishonestly by small numbers of people who have seized positions of power, sometimes precisely for that purpose.
They change public language to further their positions, ideas, or programs; to seize control of public topics; to seize or secure power over the public.
It is not as easy to parse this as it may seem. Who is restricting honest organic input into public language? Or when is organic input into public language itself but a ruse to falsely commandeer that language?
After Lenin and Stalin seized control of the public languages of the Soviet Union, we can see a clear-cut example of bad actors creating a basis for indoctrination. Before they seized power, we can see an example of a dishonest “organic” group seeking to commandeer control of public language.
And how do we see that today, through the lens of “history”?
Firstly, whose history? The same problem with public language arises.
Secondly, maybe we can never know. Maybe only societal laws or rules of governance can help us determine what’s right or best. But then the same problem arises.
Whose laws, whose rules?
In this sense both public and private languages have enormous problems basing themselves on anything.