Novel Thought Duration Is Unknowable A Priori

:date: 2026-01-28 08:57 :tags:

I believe that it is impossible to know how long novel thoughts will take to think before one thinks them.

A corollary is that if you can predict the duration of thinking before it occurs, then such thinking is not novel.

This is a simple idea to me. Now. But a lot of novel thought went into it. It might not even be correct! What follows is an expanded explanation of the idea and why it might be correct and important.

Just like ones and zeros can do a lot in a computer, thoughts may have physiological mechanisms of the nervous system in common but they can be distinguished and classified by function. Many of our thoughts are autonomic — breathing, swallowing, Pavlovian body chemistry changes, etc. These thoughts are deeply internal and involuntary. They are safely set aside when considering discretionary thoughts.

Thoughts we have control over are quite interesting. They would seem to be the only non-reflexive mechanism the "me" of "myself" has to improve "my" life or even articulate what that could even mean. I propose that voluntary thoughts can be further divided into two major categories: primordial and serialized.

Primordial is Latin for "first + to begin". These are novel, original thoughts, ideas, and even feelings. I concede that it is possible that this kind of thinking may even be autonomic also. That is, you really may have no serious control over these kinds of thoughts. It is, however, a popular fiction that we do have some control over them so let's go with that. I believe that one of the properties of primordial thoughts is that how long they will take to arise/complete can not be predicted.

Currently my opinion is that almost all other discretionary thoughts can be associated with serializing. Serialization is a a computer nerd word that describes the process of organizing an internal data state to be packaged and made ready to transfer to other computers such that the state can be recreated there. Similarly, serialized thoughts are mental states which are packaged up with some kind of formalized semantics suitable for communication. Note that the communication could be internal, for example with your own memory faculties or inner monologue.

If you can think of a thought that you have not yet had, but trivially could, I believe it can be explained as a component of some kind of serialization function. For example, if a tailor has never made a bespoke suit for me, but is confident that a pattern for one could be designed in a specific amount of time, the suit may be new and seem novel but the mental processes involved in producing it are well serialized. There is a lot more detail that could be explored to elaborate on the distinction between serialized and primordial thinking, but for now let's definitionally assume the premise that primordial thoughts can not be timed and see how that fits our experience. Definitely send me (serialize) any challenging (primordial) counter-examples you think of.

Why are primordial thoughts not amenable to a priori duration predictions? The answer has to do with the concept of a solution space or feasible region. Mathematically this is a set of possible candidate solutions. Finding the optimal solution can be very hard. For example, the solution space for my safe's combination is a million possible combinations with only one being correct. Without a (serialized) protocol involving statistics or safe cracking or me just telling you the combination, how accurate could your prediction be for how long it would take you to open it? It really is like asking, when will you win the lottery?

An interesting way to understand this classification of thoughts is to consider a form of "thinking" that is alien yet still familiar to us: the cognition of ants. Ants evolved to solve problems just like the neurons in our central nervous systems evolved to solve problems. Many ants busy themselves internally within their nest managing food storage, reproduction, and other essential ant matters. These unseen ants would be analogous to autonomic thoughts.

Ants can also be quite visible with external activities, primarily foraging. Foraging activities are like our discretionary thoughts. Most ants engaged in foraging have access to a pheromone trail to some known previously discovered food. The pheromone trail is an ant's form of serialized thinking. When an ant sets out along a known trail, it is likely that the duration of the mission can be accurately predicted in advance based on how long it has taken other ants to make the nearly identical journey. It may also be possible to calculate the duration based on first principles of how long the trail is and how fast ants generally move.

However, if an ant will be going off piste to prospect for unknown food sources then the duration of its particular search is fundamentally unknowable. This kind of food discovery is the colony's primordial thinking.

I'm trying to roughly cover the main concept here but of course there are fine points and subtleties. Even the tick of a precision atomic clock — one of the most time predictable things we know about — has timing uncertainty. And likewise, primordial thinking may have some strong hints or even hard limits. For example, an ant scout searching for a novel food source is definitely not going to take longer than an ant can survive alone without finding food. If it takes ten seconds to try a safe combination, opening my safe should not take more than 116 days of full time attempts.

What can we say if there is some smooth gradient between thoughts whose duration can be predicted and those that can not? It can still be helpful to talk about the two distinctly with some arbitrary dividing point, just as it is sensible to talk about yellow and green being different colors and not two radiation sources with nearly identical wavelengths.

Ok, so why might we care about this distinction? As ants require new sources of food, humans require new solutions to problems. I have observed that the (serialized) protocols for producing primordial solutions are frequently somewhat muddled.

Imagine ant management talking to the foraging supervisors who are reporting that the bonanza main food source enriching the colony is starting to run thin; a new source is needed. The supervisors propose sending out scouts in search of novel sources. At that point ant middle management interrupts and asks,

"How long will that take?"

Ant management has been managing the steady flow of food from the main source for so long that the business of acquiring food seems possible to accurately forecast. Unfortunately, the correct answer to the question is effectively forbidden around a manager who can fire you. The correct answer is, "I don't know". Or better, "That is a degenerate question whose answer can not be known." (Followed by, "And here's a link to a rambling essay making that case.")

Imagine a technical solutions company in the human world. The previous scene is roughly replicated but a new technical solution is needed to solve some new technical problem that has arisen. If a time estimate for the discovery of a solution can be given, then that solution has in some way already been discovered. If it truly requires primordial thinking, then it is not going to be possible to predict how long that might take.

Maybe what the company does is very similar to solving a Rubik's Cube. One day a newly scrambled cube needs a solution. If the company has been just twisting identically scrambled cubes according to a static (serialized) recipe then a new cube likely will require some primordial thinking; how long that will take is unknown. However, if the company has already developed a generalized algorithm for solving cubes, then solving another starting position is really an exercise in serialized thinking which can also have very accurate predictions.

Note that the fact that accurate time projections for large solution space searches are routinely given does not make them valid. It makes them slightly incompetent fibs that muddle together with other projections because the distinction I'm outlining is not well appreciated.

If the distinction between primordial and serialized thinking can be so unclear it might be helpful to sample a non-exhaustive list of where primordial thinking tends to be required. What kind of thinking is primordial?

Pretty much any research by definition. Even following a very serialized procedure such as looking up words in a dictionary has a pretty big standard deviation for how long it takes me to understand what I need to understand.

The design of something that doesn't already have extant examples to emulate. That something could be a tiny subcomponent of a common well established thing.

Taking a turn at chess or Scrabble or completing a crossword or jigsaw puzzle. Finding a terrain feature in Minecraft or a monster trophy in Witcher 3 is almost too analogous to ant scouting. I don't know if it's true for everyone, but some people enjoy primordial thinking.

Coming up with jokes or a satisfying melody. Free form jazz and extemporaneous rap music are very interesting activities.

Debugging a problem with an unknown cause. (We don't like this thing happening; we don't know why it happens; make it stop.)

Science done properly — both originating interesting hypotheses and the methods with which they might be falsified. The actual falsification effort itself however might involve following a well-established serialized experiment design. Also devising strategies for discovering things like new species or protein structures or celestial objects, etc. In this case, even the serialized follow through activity of the actual search may not be forecastable.

Managing a dynamic group of people is pretty open ended. What improvements should the group undertake? What is the optimal allocation of the group's resources? Also related is crafting new laws, peace treaties, assistance programs, etc. Ironically, the people managing the activities of other thought-producing people are the ones most likely to ask for unknowable time projections for primordial thoughts.

Composition is interesting. Clearly coming up with ideas worth communicating requires some primordial thinking but even formulating extant ideas into a satisfying serialized form also requires primordial thinking. Once an ant finds a new source of food, a different primordial task begins as she tries to mark a serialized trail back to the nest.

Studying philosophy can be purely serialized, however practicing philosophy is primordial thinking almost by definition.

An example of primordial thinking I take a special interest in is software. Software is a computer's structured thinking. Writing software then is necessarily thinking about thinking. And thinking about thinking is philosophy, which pretty much is always primordial. Crafting good software which solves a problem that software has not solved before must be primordial. I've already reviewed Brooks' Mythical Man Month, a book filled with software development challenges whose title hints at the profound difficulty of time forecasting the art. To create software that makes progress is hard and requires primordial thinking. How long this will take to do properly is unknown despite a team's agility at making shit up.

Most of these examples cast an interesting shadow of counterexample. If my theory is correct, why is it then that AI can "think" about problems and provide satisfying solutions in a time span that can be predicted a priori? LLms can debug code, write poetry, play Dungeons & Dragons, etc. Even GOFAI image detection can find Waldo, if not quite instantly, in a very specific predicable amount of time.

I assert that the AIs are not doing primordial thinking. Mostly they're just running an input vector through a network of calculations, perhaps scoring the result, and repeating that over and over. They repeat it not until the highest possible score has been achieved but until the allotted time runs out at which point the serialized process is terminated and the best solution thus far is presented. This is how a simple chess engine might work for example. Chess may be primordial thinking for you, but not for the computer. And that "creative" LLM output is exactly like a "creative" move of a chess engine. It is a clever illusion.

Could an AI be infused with primordial thinking with a random number generator or some other kind of non-deterministic process that makes the operational time unknowable? Possibly. I'm not sure. Maybe there are fundamental limits to this kind of thing and humans (and other animals) are already stumbling through the vast search space of ideas at a pretty good clip near this limit. Also, don't forget that an LLM's real primordial thinking happened during training. How long will AI training take to get the performance you desire? Hard to say!

As a starting point though, I think it is reasonable to say that problems that truly require primordial thinking have the following property: It is not going to be possible to say how long the solution will take. If you can, you necessarily have somehow already in essence solved it. To waste a creative human thinker's energy with the forced fabrication of fantasy time estimates is distracting and robs mental resources that could be applied to the actual problem. If any such estimates are given, it should be well noted that the person giving them is comfortable "hallucinating" things.

I get it that middle managers really, really want to know in advance how long it will have taken once the clever idea is finally thought of. But wishing something were possible is not the same as it being possible. A more realistic strategy might be to try to figure out the point someone assigned to do some primordial thinking might become bored or frustrated by a lack of results. Allocate a shorter amount of time, but not frustratingly short either. Check back at that time and see how it's going and evaluate if more time should be spent or if losses should be cut. It's also important to note that if you are trying to extract creative ideas from people, giving them sufficient time to explore the idea is a kind of seed such that even when the project is restructured to not desperately require a particular solution, there is still a good chance that a creative thinker will think of something brilliant some time later in the shower.

Just remember if you are trying to extract a truly creative solution from a professional creative thinker and you are tempted to ask something like "How long will it take?" the answer must inevitably be, "Now, longer."