Neurons as the Smallest Units of Free Will and Consciousness: The Coexistence of Determinism and Unpredictability in Future Forecasting
The Possibility of Predicting the Future
Generally, the difficulty of predicting the future is often explained from three perspectives.
The first is the lack of information. If there is insufficient information about the initial state or laws of the subject to be predicted, it is difficult to predict the future.
The second is the existence of complexity and errors. Even if we fully understand the initial state and laws of the subject to be predicted, complex systems are difficult to predict. In particular, in systems where errors are amplified, even a slight error in the initial state will lead to a significant discrepancy between prediction and result.
The third is uncertainty. If the laws include elements that are randomly determined afterwards, it is difficult to predict the future.
On the other hand, it is believed that future prediction is possible in deterministic systems that are simple, have no errors in the knowledge of the initial state, and have all the information, and are without uncertainty.
At first glance, it may seem that future predictions are possible under these conditions, but that is not necessarily the case.
Even if all the information is in place, the system is simple, there are no errors in the knowledge of the initial state, and it is deterministic, there are structures that cannot predict the future.
This is the case when there is a mechanism that changes behavior by referencing the results of future predictions.
From the outside, this structure seems predictable. However, from inside the structure, it is impossible to predict. This is because there is a possibility that a different future from the one predicted will be formed by the mechanism that changes behavior by looking at the results of future predictions.
The Reward Structure of Two Ropes
The promise of rewards
There are red and blue ropes, and if “OK” is displayed when you first pull the rope, you will get a reward.
The gimmick of two ropes
If you first pull the red rope, “NG” is displayed, and then when you pull the blue rope, “OK” is displayed.
If you first pull the blue rope, “NG” is displayed, and then when you pull the red rope, “OK” is displayed.
Operation rules of simple machines
Check the future monitor and pull the rope that displays “OK” first.
Future Monitor
Displays future predictions.
Can it display a future prediction that matches the actual future? The answer is “no”.
Even if you display the future where you pull the blue rope after pulling the red rope, the simple machine will pull the blue rope first, so the future and the actual do not match. Similarly, if you display a future where you pull the red rope after pulling the blue rope first, the simple machine will pull the red rope first, so the future and the actual do not match again.
Coexistence of Determinism and Unpredictability
In other words, despite such a simple mechanism and operation rules, it becomes clear that the future is impossible inside this structure just by adding an element that decides the operation by looking at the future monitor. There is neither uncertainty nor error there.
Therefore, even if it is deterministic and simple and there is complete information, if there is a mechanism that changes behavior by referring to the results of future prediction, it will not be possible to make future predictions that match the actual future.
There, you can see the coexistence of determinism and unpredictability. The key here is the structure that includes a mechanism that predicts the future and changes behavior based on it.
It should be noted that from the outside of the structure, future prediction becomes possible. This is because from the outside of the structure, there is uncertainty about what this future monitor will display, so it does not meet the condition of having all the information. Therefore, from the outside of the structure, if all the information is available after determining what the future monitor will display, the future can be said to be predictable.
Conditionals as the Essence of Free Will
At this time, this “mechanism that changes behavior by referencing the results of future prediction” is a mechanism that performs conditional branching. If the future monitor displays that it is “OK” when you pull the blue rope, pull the blue rope. Otherwise, pull the red rope.
This is a behavior that can be defined with a single conditional statement in programming. Conditional branching is called an if statement in typical programming languages.
Generally, we consider a subject who observes the surroundings, makes a final decision and takes action based on the observation results and the memory and experience inside as a subject with free will.
And the existence of this free will is sometimes raised as a factor that hinders future prediction. Even in a deterministic situation, the reason we think the future can change is because we are taking a view that the reason is because there is a subject with free will.
At this point, the “mechanism that changes behavior by referring to the results of future prediction” fits perfectly into the explanation of this free will. From this, I think that a single conditional statement might be able to be said to have free will. In other words, a single conditional statement is the basic and minimum unit of free will.
The simple machine in a system where you get a reward for pulling a rope is a very simple mechanism, and if it’s a program, it can be written with just one line of if statement. In other words, it is a simple mechanism that can be expressed with a single conditional branch. This simple mechanism seems to have the nature of free will in this system.
Prediction as the Essence of Consciousness
And for conditional branching to have the nature of free will, the structure of “observing the results of prediction” is necessary. This structure is a cognitive structure and may be something that can be said to be consciousness.
Therefore, you can also see the possibility that “prediction” is the essence of consciousness.
And the structure in which “prediction” becomes the input of “conditional branching” and the result works on the outside world is what I think is the structure of consciousness and free will.
Minimal Model of Consciousness and Free Will
Of course, this simple machine and future monitor themselves do not self-recognize. There may be a need for a more sophisticated mechanism for self-recognition.
However, if we assume that self-recognition is not an essential requirement for consciousness and free will, and that if we can recognize objects, consciousness and free will can exist, it is possible to think that such a model is a minimal model of consciousness and free will.
On the other hand, even if self-recognition is considered essential, the story is not over.
When you think about what self-recognition is, the story divides depending on whether you perceive object recognition and self-recognition as completely separate, or you think that there is a self as part of the object of object recognition.
If you take the position of perceiving object recognition and self-recognition as completely separate things, and think that self-recognition is an indivisible condition for consciousness and free will, the simple machine and future monitor are nothing more than toys.
However, if you consider that there is a self as part of the object of object recognition, even if you take the position that self-recognition is an indivisible condition for consciousness and free will, the structure of two rope rewards has meaning.
In other words, the simple machine and future monitor include themselves as objects, and you can think that
they just can’t separate and recognize the part of themselves and the part of the external environment well. In that case, the simple machine and future monitor still seem to be a minimal model of consciousness and free will.
Conditionals and Neurons
The argument that a single conditional, or in the case of programming, a single line of an if statement, is the smallest unit of free will may seem like a leap.
However, when I came up with the relationship between this conditional branching and free will, I had the intuition that a question I had long had about the neuron model, which underlies many artificial intelligences, would be dissolved.
A model of a single neuron is multiple arithmetic operations and a single conditional. Yes, the neuron model includes a conditional. And when compared to the discussion so far, there is a possibility that a neuron can be said to be the smallest unit of free will.
Artificial intelligence, AI. Its definition is broad, and there are various methods, but the AI system that is most expected and has achieved many astonishing successes in reality is the one that adopts a model called a neural network. Neural networks are constructed by connecting neurons that model the mechanism of nerve cells in the human brain as basic units.
Prediction and Neurons
I also mentioned the possibility that prediction is the essence of consciousness.
The future monitor presented in the “Two Ropes Reward Structure” was a concept prepared to present a discussion on whether complete future prediction is possible. However, it can also be reinterpreted as a device for future prediction as the best effort.
In that case, instead of mirroring the future from now on, this future monitor is interpreted as a device for predicting the future from experience.
Then, you can think of a mechanism where a simple machine challenges the two ropes once in the past, records it, and the future monitor displays it.
Thinking that the same thing will happen as the past challenge may not be a smart idea, but it is a rational strategy as long as there are no other judgement materials.
Then, the simple machine observes the result of the first challenge as a future prediction and pulls the rope according to the rules. At this time, the simple machine will pull a rope of a different color first than the first challenge.
With this, the structure has done two challenges, pulled different ropes each time, and got different results. What future predictions to display on the future monitor at this time will have various strategies. One method is to display a probabilistic future prediction that says it’s a 50/50 probability by treating the two challenges equally. Alternatively, you can consider a strategy to give strong priority to the most recent challenge and display it as the only future prediction.
In this way, I think that the mechanism of recalling the future of similar situations from the accumulation of past similar experiences is prediction and consciousness.
And the neuron, which is the basis of many artificial intelligences, is exactly such a basic unit that learns in this way. Therefore, it can be said that a neuron is likely a minimal model of consciousness.
Neurons as Minimal Models of Consciousness and Free Will
Many people may perceive the key to the emergence of intelligence in the scale and complexity of their network structure, but I have been pondering the smallest unit, the neuron.
If intelligence, consciousness, and will reside in AI based on neural nets, I thought that the essence must be in the neuron itself or in the structure that connects the two neurons. And if consciousness and will do not reside in the neural net, I thought that some decisive element was missing from the neuron.
Therefore, the discovery of the relationship between conditionals and free will was an interesting discovery that could be the answer to this question for me. Of course, I think there is a lot of room for discussion, but at least my intuition is that a neural net can have free will.
Similarly, the discovery of the relationship between prediction and consciousness also strongly pushes to resolve my question. It is clear that in neural networks, a neuron is the smallest unit that makes predictions. Therefore, if prediction is the essence of consciousness, a neural network can be said to have consciousness.
What remains is self-recognition. If that issue can be clarified, probably, you would be able to answer the question of whether a neural network can have consciousness and free will.