anecdotalisation

The sure sign for me that I’m coming around is when personal narratives begin again.

A. anecdotalisation

These can be fantasies, or self-narration, anything involving the construction “I”. Self-narration is such a force in my internal life that I have had many false awakenings in dreams, where I find a friend (still in the dream) and explain to them what had just happened in the previous part of the dream. I think the reason I rely so much on self-narration is because I read joke books as a somewhat autistic kid, my role model was Mr. Bean, I enjoyed making people laugh even if I was the butt of the joke, and my third grade teacher Ms. Stabb had this big box of gummy bears and she would give one to anyone who could make her laugh with a joke after school each day. This concludes the world’s first meta-shaggy dog¹.

I want to explore anecdotalisation as a process of self-construction, positive or remedial. The impetus for this post is the latter. A little bit of silicone earplug was stuck in my left ear today when I woke up. No physical pain, but very uncomfy. After some failed attempts to remove it, anger and frustration and panic, I checked the internet. I had managed to do just about everything that I was not supposed to do. I shouldn’t have ripped the earplug in half (I thought I was gaming the system, but I have a whole box of these things and zero reason to stick it to Big Earplug), I shouldn’t have used a tweezer (that just pushes it further in), I shouldn’t have used an ear irrigator (that just pushes it further in), I shouldn’t have dug my finger in there (basically just to push it further in). It was in so deep (still is) that I felt my sense of balance go off. The non-emergency line for the NHS is totally blocked off on account of the coronavirus, and now I have to measure up going to a hospital in the middle of a fucking pandemic and not making sense of Bohemian Rhapsody. I have grimly accepted the possibility of losing hearing in one ear (COVID-19 symptoms: deafness [1:8,000,000,000]), and I figure, Odin gave an eye for wisdom, I’m getting a bargain! Behold the wisdoms, which will also structure the remainder of this post.

1. Keep Calm and Think before you Act
2. Don’t use tweezers to remove a silicone earplug
3. It’s not time that heals all wounds, it’s the anecdotalisation over time

B. proverbs and superanecdotality

You know how sometimes some wise one drops a “stitch in time saves nine” or “early bird gets the worm” as if it’s sage shit? Most of the time it’s just pattern matching: see a situation, find thematically matching proverb, sprinkle into small talk, congratualations you are a native speaker of language X. What can we make of the semantics of proverbs? Too ‘vague’ to merit a truth-value, and not quite as pushy as an imperative or even a suggestion like an Aesop. Even if we look at the pragmatics, most of the time the only identifiable objective or principle a casually uttered proverb satisfies is “Pass the Turing Test”. What are they, and why is language full of them?

What we can say is that proverbs are strongly memetic and probably symbiotic with culture (we haven’t even gone into chinese proverbs; you could have entire conversations with them and they’re sometimes rhymy and catchy). It seems that proverbs come in two layers: an outer memetic shell, and an inner nugget of ‘insight’ condensed into metis (see James C. Scott’s “Seeing like a State” for the distinction between Metis and Episteme.) For example, “early bird gets the worm” evokes some visceral imagery, and there’s a kind of poetic semantic elision that invites the listener to fill in details. The ‘inner core’ is removed by a degree of abstraction: “early action means success”. In a decision maker’s head (i.e. anyone) a floating proverb is a tiny force upon the overall heuristic. We can see a proverb in the early stages of evolution in Minecraft’s “never dig straight down”; it’s memetic, but not per se as its quite literal, rather parasitic upon the memetic vehicles of the rest of Minecraft. But “never dig straight down” is an excellent example of what I think partially distinguishes superanecdotality from anecdotality: epistemic justification. “Never dig straight down” is pure metis/ritual. Epistemology comes in partially when we refine to: “Never dig straight down, because you might fall into lava and die”, and real, internalised knowledge comes when we ground the justification empirically: “This one time I dug straight down, fell into lava, and lost all my stuff, so do not dig straight down”.

I think Proverbs are very very abstract anecdotes, that require, in order, an almost pataphorical instantiation ‘sideways’ to unpack the memetic shell and get to the right context, and then an instantiation that tags empirical justification. Conversely, starting from many normal condensed experiences in the form of anecdotes, abducting/abstracting away specificity at the level of personal experience and context ought to arrive at metis, and then you get a little creative.

C. un-narrativised experience and subanecdotality

I’ve been bandying about the idea of an AI that learns how to play Minecraft “interestingly” for a while, long before the MineRL challenge (which had disappointing results IMO). This was something I have been thinking about with Philipp Koralus. What do I mean by “interesting”? At night (generally, in darkness) in Minecraft, monsters (mobs) spawn. Now think, what would a reinforcement learning agent do in such a sandbox-type environment where rewards must be user-specified? If the agent is penalised for dying, it might just get very good at avoiding mobs during nighttimes. If it’s particularly clever or lucky, it might decide to stand on a stack of blocks out of reach of the mobs, and simply do nothing half of all playtime, unless we further penalise it for doing nothing, but that’s like the most fun part of Minecraft. How would an agent come to the conclusion that the reasonable thing to do is build a lit shelter? I asked my sibling to do a blind let’s play of minecraft, which is quite instructive. Check out the segment from 15-17 minutes: they spend a while running from mobs in the nighttime, and then they get the bright idea to stand on a stack of blocks out of reach of the mobs. Sound familiar? And then they get the interesting idea to widen the stack of blocks at the same height so that they have space for a crafting table and some movement agency.

Me and Philipp think the following approach is a decent start. For context, Philipp’s big idea is the Erotetic Theory, which basically generalises Kahneman’s system 1 and system 2 thinking by placing them at two ends of a spectrum paramaterised by ‘inquisitiveness’. The big claim (backed by plenty of empirical data, I’ll post when the book is out) is that all failures in rationality are due to failure to ask the right questions at the right times, as raising alternatives in our mental representations is expensive, and our reasoning strategy defaults to greedy maximisation of answerhood potential for each new piece of information with respect to our current mental models. Here’s maybe how to have an AI play minecraft interestingly.

0. Make the agent is a cognitively lazy bastard like us.
1. Knowledge:
   1. Have the agent collect ‘knowledge’ into a logbook.
   2. Distill ‘knowledge’ into ‘insight’, to feed into future decision heuristics.
2. Use ‘knowledge’ to deduce methods to reduce cognitive load.
3. reward agent for idle cognition and discovering new things about the world.

We know that pressing constraints of the environment increase cognitive load. One gloss of civilisation is that it’s all there to reduce/transmute cognitive load. Think Dougla’s Adam’s take on civilisation: “…the first phase is characterized by the question ‘How can we eat?’ the second by the question ‘Why do we eat?’ and the third by the question ’Where shall we have lunch?”. “Where the food at” is a hard question laden with difficult cognitive avenues to explore for the hunter-gatherer, but it’s easier when you have a kitchen, or in the case of Minecraft, a farm to grow wheat to make into bread. The hope is that rewarding lesser cognitive effort expended on solutions rather than just the solutions themselves might lead an agent to rediscover some of the same things we take for granted, when the agent decides to simplify their environment so that the most pressing questions have a single, obvious answer.

The only tough point in that plan is the condensation of knowledge into insight. How to go from a collection of unlabelled and essentially factual experiences into a condensed and meaningful representation? I think purely logical methods don’t quite cut it; while we do have a formal framework to handle probabilistic first-order logic erotetically (paper in review), there’s an element of structure missing, and I think one way to get at this structure is to factor the process through anecdotalisation. The Queen gave the Jack two apples, the jack poisoned them both, and gave them to the King and Ace. How much more likely are you to remember that than the list {Queen,Jack,2,King,Ace}? It’s almost as if stories are a way to piggyback upon all the quasi-causal, relational and contextual knowledge you have in order to string together information. What’s crazy is that most of the time we learn all of these relations and contexts from stories in the first place! So Philipp and I thought that maybe a missing piece here is to give the agent a kind of bardic nature.

4. The agent keeps a Star-Trek style logbook, chronicling its adventures.

Then again, that may only feed back into the whole rational cognition process if there is some kind of society of these things to share the stories with, and be rewarded for. Though I’ve been talking about Minecraft in this section, I may as well have been talking about humans and Earth at large.

D. anecdotality as Maker’s Knowledge

I don’t know if narratives of self are universal (like Jordan Peterson would have us believe: “know what story your life is, so you can decide whether you like the ending”, or something along those lines), but they seem pretty powerful tools of self-construction, when used properly (proverbially, “fake it ’till you make it”.) I think that this idea is a good thematic bridge to so-called “Maker’s Knowledge”, a murky realm where certain modalities of thought don’t apply.

Here’s an example. I really like Hot Dogs, and we’ll skip the personal stories that make this true for me. So I am sitting at a bar that happens to have Hot Dogs, and I wonder to myself: “What is the probability that I will get a Hot Dog within half an hour?”. Statistically speaking, some frequentist or inductivist who is not me could have said “pretty likely”. Even the bayesians who knew me would have been justified saying so. But they’re answering a slightly different question than the one I’m facing: What do I think is the probability that I will get a Hot Dog within half an hour?“. Frequentism denies my own apparent will in the matter, and Bayesianism opens the door to the jungle of epistemic model logics. Do I believe that when I believe I can do X I will do X? Well, if I do, then I believe that when I can believe that when I believe I can do X I will do X, I will believe that I believe that when I believe I can do X I will do X. Damn, I’m hungry.

Beliefs about ones own decisions, past and future, are hard to build on neutral grounds. If you dig into the foundation that comes after “because”, it’s either Evo-Psych/Bio or more likely, anecdotes. Stories about what kind of person we were, are, will become. Actual change is painful and it’s easier and feels better to tell myself that I’m a certain kind of person than it is to make active efforts to be a certain kind of person. I don’t know about you, but I’m guilty of that all the time, and my admission that I’m that kind of person means I don’t have to do anything about it. #gamethesystem.

When we tell stories to ourselves, who is the audience? Whatever that thing is, the process creates the necessary discretisation and distancing from my own continuity of personality that I can actually become a different person. So that’s a new Gould-style take on Parfit’s theory of Identity as continuity of psychological experience; the psychology might be continuous, but the personality jumps are discrete, and their units are anecdotes.

E. conclusion

I’m a dumbass, and I need to see a doctor.

EDIT: the earplug is out

Footnotes

1. The editor notes that Robert K Merton’s On the Shoulders of Giants: A Shandean Postscript is earlier.