All logic is a prior
This is a rebooted version of all-logic-is-a-prior on substack.com.
There is an interesting discussion involving game theory and learning models at Updatelessness doesn't solve most problems (in the unavoidable context of AI training).
It’s written for the lay person. But it contains a lot of jargon, so it’s really for the lay person who knows what game theory is. The learning models (AI) part of it investigates that field of game theory with the terms updatelessness and updatefulness but using a strategic filter.
Off you go…
Since writing this over at Substack and the comments there I’ve updated my confusions here. Particularly by Apple Pie.
"I'll give you a loaf of bread for $5... OK, thank you for paying me $5 for this bread. Now that I have your money, I've updated on the situation and realize it makes sense to leave with the bread."
"Will you tutor me in mathematics for $30? ...OK thank you for tutoring me in mathematics today. Now that I know how to solve these problems, I've updated on the situation and realize that I won't need you again, so I don't think I'll pay."
There is a reason why evolution selected us to have integrity: we earn reputations. If every interaction we ever have is one-off, and can't be transmitted to others, then reputation doesn't exist, and short-sighted selfishness prevails.
He used the word integrity there, it helped delineate a model which updated me priors on this, and to have another look.
Anyway, I realised I was not using the exactly right frame. And it also made me realise that the awkward terms updatelessness and updatefulness, can be seen as versions of the old Nature/Nurture conundrum, interaction or gap.
In evolution the awkward terms updatelessness and updatefulness are replaced by behaviours hard-wired or less hardwired at birth.
Hard-wired means that for the duration of the life of the individual in a population it does not need to update those elements, even if it turns whole slabs of the genes off and on at times, tides and season. Any 'learning' is done in the selection of the population that shares genes and successfully reproduces.
We call this instinct. It is learning but not in a life-time as we know it.
Where it is less hardwired, learning and meta learning are required in development and maturation to develop routines and reactions that allow survival and reproduction.
All life does both to varying degrees. There is a trade-off delimited by lifespan. Learning takes time. It might be better to not learn the same lessons every time (father looks at teenagers, what is the benefit here exactly?).
In short-lived lives “instinctual” memory and its reputational unwelten would dominate.
Longer-lived 'lifes' (which would thus include cultures as trans-generation bets… —in game-theory each life is a bet) (a life is a bet placed on those genes a phenotype's body carries) would tend to use more updatefulness learning routines. They would create a world.
I also suspect the same trade-offs explain the potential for shorter-lived lives to develop a hive of social insects as it provides a longer life, not of the individual but the bet that is the hive's working memory. I doubt if there is much benefit for longer-lived lives to emulate this when culture is available. I.E. the economic advantage among longer-lived lives between ①non-breeding siblings with its hive memory, and ②non-sibling co-operation in the world is not in the hive memory's favour.
We need to develop a memory index like Kardashev's Scale did for a civilisation's capture of energy. No doubt they could be put in proportion.
It is important to note here that there is a pathway between less hard-wired learning in life and hardwired in the genes call Baldwinian evolution. All hardwired behaviours have to be learned before they can be selected for and be hardwired into the genetic code. In this context non-hardwired learning is a "meta-prior" for a later successful hardwired instinct. Evolution is constantly playing-off between what can be hardwired more and hardwaired less.
I suspect evolutionary theory if regarded as learning model will help more with issues raised in the less wrong article Updatelessness doesn't solve most problems .
And now back to the original text of my post. Which goes all metaphorical, poetical, if not, shamanically beserk.
— Meta-strategy structures iterate and thus possibly lead us into an impractical chaos, no chaos-surfing possible where their is no edge… —and even lesser chaos in the complexity in recursive doom spirals of madness-as-tactics become interference across levels, across governance and playmaking. Bookmaking.
Updating priors… if logic is a prior, how do you update that? (Is Logic updateless?) Has it got its own updateful prior, like learning for instinct? Is logic the end?
Is that a meta question? Is this a meta-meta question?
Priors are messy because they require a point of view (and so have a partiality if not a consciousness). This is the result of evolution, while evolution is a generalisable mechanic, any instant of it has a history. The laws logic provides are not the lived experience they describe or control. The territory is not the map.
So can logic itself be a partiality? Is logic just a point of view (POV)? And one born of another POV which may not itself be logical or available to logic. Are POVs always meta? Is the meta always prior to the iterational space. There is a confusion between the instance, the instantiation, the funcion, and the overloading of the function (update pending), the cycles are not clear, what will sort out the mess? Evolution? An emporer's favourite god? Winners are grinners said the wearer of the bear skin.
If you need a POV to develop logic rational objectivities, or even just rigourous logical forms and systems, but the POV itself does not require those to get started… Are logical forms updateless? If updateful are they then illogical, a logical, or just meta-metastasizing according to the tetralemma of enlightenment?
Just some emotional axioms to experience in the flow like a fear of being eaten? You know, a system can begin with any prior at all, or population of genes, and always develops into cute pictures of cats no matter the prior given enough iterations of monkey bashing on type-writers.
Can you have training without updating? Does revelation end when we go updateless.
If a POV is not required to be at all logical, but its survival or adaptation become a prior from which logic ascends, this explains why winners are grinners despite the pay-off gameplay that logic could provide.
Is all logic a post-rationalisation, a hindsight, that winners grin with? Evolution doesn’t care about logic per se.
“and there is a dependence on the observer’s ontology[8].”
I've removed the rational actor stuff because | evolution won out.
Also my comment
All logic is a prior.
received the reply:
Chan master Yunmon
And I had to look that up https://en.wikipedia.org/wiki/Yunmen_Wenyan
English: Yunmen Wenyan (830-902). Woodcut, Ching dynasty; caption:「四十世雲門文偃禪師」
This is a rebooted version of all-logic-is-a-prior on substack.com.