The Markov property sounds like a terrible stricture at first - "what do you mean, you're trying to solve problems without using a single bit of memory?" - but it gets the job done in many really complex domains.
And the harsh limit avoids some human pathologies. There is no sunk cost fallacy in RL; many people live as if a thing's history was the most important thing about it by far (cycles of retributive justice, national grudges, classism and its inverse), but no Markovian robot ever will.
"Temporally correlated behavioural noise"
- modelling term for irrationality (sleepiness, drunkenness, raginess)
Diagnosed my DQN algorithm with depression today.
Its memory was full of low-reward experiences, which it was constantly replaying, and its curiosity/impulsivity (epsilon) was way too low to break out of the loop and try new things. So it just kept literally hitting its head against a wall.
How many actuaries does it take to change a light bulb?
- The same number that it took last year, adjusted for trend.
- None: after credibility weighting, we have indications that the bulb is still lit.
- None, the underwriters are not allowing any modifications to the bulb at this time.
- Have any of our competitors changed bulbs yet?
- None, they prefer to leave us in the dark.
- Five: one to screw it in, and four more to estimate the length of its life before being screwed in.
BEGIN KEYBASE SALTPACK SIGNED MESSAGE. kXR7VktZdyH7rvq v5weRa0zkXmNnZo QnJSeOxKoAUXmIN gnA37CbYbAec1d7 NLyINzmfNzkNab2 YUyVcOKjopfQJDQ hQdjb5HyOcjcQMd PHB15K8ht4j9YPB jAswPaMHHG1ld2c uNsEPXm6lsRuiMx VIk2EUoTZEEB8xM OG6vqjWkKXH1OAj LB0FJhk6lLYNMb5 r1hvHZYOv1AdtBW hhKQw1bYtR9AHYm wENYo1CBLOc1LDz C. END KEYBASE SALTPACK SIGNED MESSAGE.
"You're a mentalist, goin fiji oer nuhin."
"RIGHT! Get tae fu-"
"Up yoors ye -"
"Mgonnae rip oot yer-"
"Ayaaaaaa nae the wings! Nae the wings! Peanuts!"
"Does this matter? Well, in the newspapers every week, there are dozens of stories about risks and rewards, epidemiology and politics, social trends and psychological differences, with serious public-policy implications, which you can't understand without understanding distribution-talk. And usually you won't just feel baffled -- instead, you'll think you understand, and draw the wrong conclusions."
- Mark Liberman, who often educates me far beyond linguistics
"I submit that even most intellectuals don't really know how to think about the evaluation and comparison of distributions -- not even simple univariate Gaussian distributions, much less more complex situations. And many people who do sort of understand this, at some level, generally fall back on thinking about properties of group prototypes rather than properties of distributions of individual characteristics."
"Until about a hundred years ago, our language and culture lacked the words and ideas needed to deal with the evaluation and comparison of sampled properties of groups. Of 300 million Americans, I doubt that as many as 500,000 grasp these ideas... The rest of the population is surprisingly uninterested in learning, and even actively resists them, though in their frequent dealings with scientists they have a practical need to evaluate the numerical properties of representative samples..."
Clinton was mocked for hedging on what the meaning of "is" is (when that's fair: it could be predication, or existence, or identity, or subsumption).
Rumsfeld was mocked for claiming that war was motivated by "unknown unknowns".
Sure, these were shifty men, but these are not optional or pedantic concepts. They are essential for not being fundamentally confused, and that we mocked the ideas (not just the self-service they were used for) just because an enemy used them is a dark indictment.
A good way to dodge responsibility for everything you ever say: put this at the bottom of all emails.
"Some archaeologists now say that when they dig up the remains of lost civilizations they are not just reconstructing objects but reconstructing minds. Some musicologists say that playing an instrument involves incorporating an object into thought and emotion, and that to listen to music is to enter into a larger cognitive system comprised of many objects and many people."
Beautiful, even if false.
Obvious objection: Is countersignalling sour grapes?
I don't think so; I could easily afford Fiji water or Apple gear or even the heavy upfront luxuries of idiot fashion.
It certainly feels like I am responding to the irrationality of wasting money in a needy world, but how can I know that I am not really jostling for status myself?
This is just the normal signalling / countersignalling account of human behaviour, sure. But I never noticed how strong and correlated these things are. My peers (blue tribe? grey tribe?) would never show off their Rolex, even if they found themselves owning a bloody stupid Rolex.
Is it fair? Maybe the person honestly thinks Fiji water tastes better; maybe they honestly believe in their greenwashing.
At worst, this is speciation. How could these two kinds of people ever come together?
"Some people think luxury is the opposite of poverty. It is not. It is the opposite of vulgarity."
- Coco Chanel
This isn't true - or rather, there is a set of people for whom this is perfectly backwards.
In the airport I see a litre of water for €1.50 and, next to it, 400ml of "Fiji Water" for €6.50. I feel *powerful* disdain; a resolve rises in me to sneer at anyone I see with Fiji water. So buying this status good lowers your status, at least to me and some minority.