The Lone Thinker Who Worked Out a Theory of Human Mind and Society (1)
From the "One Person Game" Theory of Mind to a Theory of Everything: A Glimpse into Peter Putnam's Unread Work.
This post is part of my exploratory journey inspired by a story that I recently read, a true story about three vastly different lives - an investment genius from a wealthy family who donated $40 million, a brilliant neuroscientist and philosopher whose grand theory of the human mind was decades ahead of their time, and a janitor and night watcher who lived in a one-bedroom apartment.
You might wonder how these three vastly different lives were connected. The answer is simple: they didn’t belong to three different people. They were all lived by the same man, and his name was Peter Putnam.
Peter’s triple-life ended tragically in 1987 at the age of 60 when he was struck by a drunk driver on the way to his nightshift. Amanda Gefter, who wrote the story that I read, spent more than a decade interviewing Putnam’s friends, students, and reading through piles and piles of his unpublished work. Her article details her journey to discover Peter, the high praise that Peter received from his advisor and coworkers, what his grand theory of the mind looks like, and how money and his mother eventually overshadowed his academic career.
But the story also left me with profound unanswered questions. Did Peter really have a theory of the mind that was decades ahead of his time, or is it just a journalist’s exaggeration? If he was a true genius, what would be his opinions on today’s artificial intelligence? What other insights are there in his work waiting for us to discover? And why has he decided to write all this stuff just for himself?
Driven by these haunting questions, I set out to read Peter Putnam myself, scouring samples of his largely unpublished work online. Reading his work demanded lots of persistence, as his unusual, idiosyncratic writing style often left me in frustration. However, the curiosity to explore a genius’s mind and uncover a piece of history kept me going, and eventually, I was able to get a delightful glimpse of his mind. Today, I invite you to join me on that journey.
The One Person Game
Is life - including the human mind - fundamentally computation, like some complex software running on a giant computer? Well, maybe, but such a statement, just like saying life is made of atoms, doesn’t provide much information. Historically, such a model degenerated to expert systems - a program with hard coded rules and knowledge - because that was the type of program that people could imagine at that time.
Putnam agreed with the computational nature of the human mind, but he saw it as a very special kind of program he called a “one person game” that runs on a large parallel digital computer - the human brain. The player of the game is the brain itself. A move in this game is taking some action. As a predicting machine of its own actions, the brain ’s goal function of this game is not “winning”, but “repetition”, to be able to make the same predictions of moves under the same situation again and again.
Through evolution, the brain is hard coded with some primitive heuristics, for example, moving away from source of pain, heuristics to find mother’s breasts for milk when we are hungry. As we grow, the biological drive gets us to explore a larger environment. But how does the brain learn how to act in the expanding, changing environment? Putnam argued, the role of interactions with a new environment, just like playing chess after you learn some basic heuristics, is to draw different scenarios where your existing, successful heuristics interact and contradict. The central role of the human brain is to resolve contradictions by refining the heuristics.
How does the brain resolve contradictions? Given a certain drive (thirst, hunger, etc), the brain has a set of contradicting next acts that it can choose from. The brain emits random acts to try, until the drive factor is satisfied. This is called external random search because acts are immediately emitted to the environment to try, but the brain can do internal random search as well, by simulating a sequence of acts without actually emitting them. He calls these chained simulation series elaboration. In either case, when the drive is satisfied, an act with temporary relative dominance is found and the brain will go back to update its contradicting heuristics - the act that lead to the winning path will be strengthened, while those that lead to losing paths will be suppressed, for the particular situation. As this process repeats in future similar situations, the brain will have less and less acts to try and a stable relative dominance will be established; in other words, it has found a path of “repetition”.
The successful resolution of contradictions opens up to new drives, new explorations and new contradictions to resolve; it is through this contradiction creation and resolution cycle, that our brains “repeat” - making sustainable and repeatable predictions in a dynamic world.
The above is a very broad stroke translation of Peter Putnam’s model of the human brain and the emergence of the mind, which was formally articulated as early as 1963. From artificial intelligence’s perspective, what he outlined is essentially an online, model based reinforcement learning system, using sparsely encoded neural networks, which is at the frontier of AI research today. From a neuroscience perspective, various components of his theory have direct counterparts in modern theories - including neural Darwinism, predictive coding, free energy principle and parallel distributed processing, but he was 10 to 40 years ahead.
If you want to know more what a model based reinforcement learning system is, check out my the other post:
In Putnam’s theory, the external random search is analogous to learning & training from interacting with the external world, while the internal random search islearning from the internal world model.
Putnam Quotes
There is no better way to understand a thinker’s mind by directly reading their words. In this section, I collected a couple of more readable paragraphs from his work (which is not very common by the way), from which you can get a glimpse of his profound insights.
The brain is a predicting machine that learns from contradictions.
The brain, as we have seen, may be usefully treated as a computer for predicting the ordering in the emission of its behaviors. In this way, all of life and knowledge are brought under the general forms of learning or education, and more concretely of self-model building. The center of attention is itself treatable as a function of latent inconsistencies or contradictions (X) in our self-model insights.
- Comments on the Origin of NS Model, 1966 [link]
The brain couldn’t have learned if the world doesn’t contain remarkable regularity. As Einstein said, “The most incomprehensible thing about the universe is that it is comprehensible”.
Thought (in its terms) is ultimately a property of the environment, or class of correlations fed into the brain, not of the brain itself. Were there not these latent harmonies in the data, the brain’s organization would rapidly fall apart.
- Comments on Functional Form of Life Game, 1968 [link]
On the relationship between learning from external interactions (external RS) and learning from internal “world model” (inter RS):
At first the internal RS is oriented as helping reconcile external RS ahead of time. Later, the external RS is oriented as helping fill in gaps in the internal RS. The internal RS becomes dominant, and the external RS is oriented as a relatively routine externalization process to help fill in regions where no through path can be found, which are then internalized.
- Mathematics of Brain Modeling, 1974, page 127 [link]
Modeling or simulating the human mind through natural languages / symbols is highly inefficient. This can be thought of as a critic of Symbolic AI (expert systems), and an advocate for embodied AI.
Automation [in digital computers] starts with a verbal or symbolic type of encoding, … , which is a very late emergent in living computers. As a result we find ourselves led into simulating non-verbal models with verbal ones, which can be very inefficient. There is no need to represent symbolically what is already available existentially in the analogue or digital parts of the human computer. [In a living computer, ] the act takes care of itself, so there is no need for the symbol-processing parts of the brain to provide a determination of acts in any general way.
- Mathematics of Brain Modeling, 1974, page 155 [link]
What’s Next
Putnam’s theory of the brain and the mind didn’t just come from vacuum. As he said in his 1963 paper:
The people studying the operation of the brain by experimental means have gone as far as it is possible by the process of direct abstraction from facts. The field is now ready for professional model builders to come in and make an overall synthesis.
But Putnam didn’t stop there. Since the brain is central to everything that we do and experience, he took a huge step forward to theorize the evolution of human society, and offered an explanation of everything from science, religion, culture, politics and war, to technicalization of human society and middle class anxiety. If the theory of the brain and the mind is Putnam’s theory of special relativity, then the theory of human society would be his theory of general relativity.
Obviously, it is impossible for an ordinary person like me to understand the depth and breadth of his insights. But even a glimpse into his theory may open us up to a new way of thinking of our own lives, and may help us understand his life trajectory as shaped by his unique experiences. These would be the topic for next post on this series.