Epistolution Musing №6: Causal Suspense

Dear Friends,

This letter is part of a weekly series of brief thoughts I would like to share with you, either because I’ve come across your related work in biology or because you’re a person I like. I discovered an interesting problem in 2019, a problem I can’t forget. Epistolution is the unknown biological mechanism necessary to explain purposive activity that can’t be accounted for by genetic influences.

Recap: with the Anna and Amanda twins thought experiment we saw that there is something about an organism other than its genes that controls learning. Since cells which have no experience with survival and reproduction (S/R) while choosing their learned behaviors, it means that S/R can’t be the primary aim of life. I conjectured that the primary aim must instead be the pursuit of a form of cellular knowledge. Now we are looking for a notion of “knowledge” that explains how cells choose between gene regulatory attractors to attain learned behaviors.

This theory of epistolution is now trying to unify epistemology entirely with biology, no small task. So I’d like to slow down and take this in little bites. I want to approach it from different angles, like the blind men describing an elephant. One man touches the tail and says its like a rope, one touches the leg and says its like a column, and yet another at the tusk describes a spear. Like this I’ve come up with four conditions that describe what this “prime mover” must be like from different points of view. I’ll lay them out in the next few musings, and then I’ll speculate more about the physical process that causes epistolution. Then in the final musings I’ll express a call to action for those of us in science who are aware of this problem to unite in groups to try and entertain proposals for its solution.

My four observations about cellular knowledge are that it is:

1. Causal

2. Conjectural

3. Information-determining

4. Open-ended

First, causality.

What motivates us as organisms? As we look around our environment, some things attract our attention and others go unnoticed. What motivates us to notice and interact with some things in our world rather than others?

As I drive familiar roads here in Vermont, I pass millions of trees, yet one particular tree always interests me. At a crossroads stands an enormous sugar maple, at least two hundred years old. Its giant branches stretch far above buildings, parked cars, a road, and a septic field. In the main stem near the ground, a rotten gash extends downward, hollowing the supportive base of the tree. I can’t resist checking every day for signs that this rotten trunk is failing.

Why am I interested in this tree? I have developed an expectation. Because of what I know about trees and gravity, I expect that one day I will look and this tree will have fallen, perhaps blocking the road or damaging someone’s property. This is normal, but I reflect that there is something peculiar about my expectation. I am waiting for something to happen that I have never experienced before. Not only have I never seen this tree fall; I have never actually witnessed a rotten tree anywhere falling on its own. I haven’t lived for long among sugar maples, and I haven’t any clue how exactly it will come apart when it finally cracks. If my knowledge about this tree were built by averaging my past experiences of similar trees in a Bayesian way, I would expect continued tiny increments of growth, not catastrophic failure. If my knowledge were built from examples I had experienced directly, I would feel no suspense at all, yet I am keenly fearful, interested, and curious. In a way, when I look around my world in general, suspense is all that I feel.

What creates the suspense, the drama, that captivates our attention? We all hold causal maps in our minds, maps that tell us how causation moves the world around us. They tell us “what wiggles what.” They aren’t based on averaging our experiences together at all, but based on forming concepts that make new predictions, explanations, in other words. In his wonderful book, The Beginning of Infinity, David Deutsch describes how these explanations are the special sauce that makes us creative. We seek to improve them. We seek out events that refute our explanations, and we spontaneously generate better ones when it happens. Imagine with me that our minds are nothing but a fabric of causal textures, one that expresses not what is likely based on what has happened before, but why things happen at all. It is this causal map that makes me intervene in the vicinity of the sugar maple, probing to discover its soundness so I can anticipate its catastrophic demise.

Holding a set of expectations that cover everything in my world is not a specific task. Yet it is a capability that could explain how I, as an organism, select the specific tasks that I am motivated to pursue. If the aim of life were to improve a causal map, then every surprise an organism experienced could provide motivational force, driving particular interactions. This capacity is of course dependent on what kinds of knowledge I can hold within my body. Depending on what you’re made of, you may not understand much, but perhaps you understand everything in a very unique way. When a deer stamps at a suspicious movement in the trees, when a fox stops to investigate a new smell, when a fish rises to a bait, when a snake searches for a den, when a fungus finds a new plant root partner, from an observer’s perspective it can appear that they are simply machines attending to their own survival. But from their own perspectives they are surely doing something else. Whether or not they have an interior life or an awareness, they have a volition, and they know something. Perhaps they are updating their expectations about the salient features of their worlds.

A new paper from Johannes Jaeger points out just how different AI is from the way organisms make sense of the world. He points out that the problems that are solved by today’s “smart” algorithms are problems that are specified by the programmers themselves. An AI cannot select its own aims; they are provided in the code. There is no global contextual understanding developed by the program. It doesn’t understand what it is and where it is, or how to do anything. It isn’t, in humanist terms, free. Large language models do not understand language, they simply piggyback on the way people who do have contextual knowledge have written about it, processing words as a form of mathematical statistical data. This is why it requires an astronomical set of data to train an LLM, yet a human child trains itself with access to only a small local physical experience. How many million cat pictures does an AI require to identify a cat? Yet a child can understand the cat-ness of a napkin or a ball of clay after seeing only one cat.

Artificial general intelligence (AGI) is sometimes defined as the ability to do with an algorithm everything a human can do with her mind, but I believe this misses a much deeper point. The superhuman ability to solve math problems has been a feature of technology since the abacus. It’s the ability to understand the world and find meaningful problems with one’s understanding that is the special thing about the intelligence that currently only organisms possess. I suppose that this ability to build maps of causation is the real “general intelligence.”

(Added 1/21/24: The way that creativity links arises from causal suspense is quite direct. When we have a causal knowledge about something, it allows us to repurpose that thing for other uses. Denis Noble once imagined this example in a conversation I had with him about this subject. Suppose that a mother monkey was teaching her child to crack nuts with a rock. The young monkey, in order to develop this complex imitation, is required to build a causal understanding of what rocks are and what they do to small things like nuts. This is logically necessary if the monkey is to find his own such rocks one day and use them likewise, because he needs to know that other suitable rocks will do for pounding, and that he can stand in the place of his mother, and many other such realistic physical facts that can’t be learned from correlations in the maternal scenes he is viewing. But if he has this understanding, Denis continued, he might well one day apply it to pound the nut into a finer powder, or combining this with his understanding of water, might guess that wetting this powder might result in a nice biscuit. Of course it is amusing to think of monkeys inventing gluten-free biscuits, but then again, we are such monkeys! )

Next week we look at why I believe that the purposive behavior that comes from contextual causal awareness with epistolution is synonymous with morality, and why this means it is imperative that we uncover its physical mechanism.

Be Kind, and Be Brave,

Love, Charlie Sent 1/2/24