Perceptual Processing

Perceptual Processing is new to me.  What little I know about it I just learned from this review of a book.  So what follows is a brief attempt on my part to summarize a summary of what is apparently a difficult scholarly book by a philosopher about a theory of the mind that has arise in the psychology community.

The idea is that the mind has many many layers.  The low level layers are sensors, muscles, etc.  The higher up we go the more matching, modeling, and cognition we find.   Each layer is then at the intersection between the rough model from above, and the noisy data from below.

Layers compare the noisy data and the rough model, which naturally don’t match perfectly.  That may annoy it – in which case it complains – or the mismatch is tolerable – in which case it let’s it ignores it.  How fastidious, how easily annoyed is it?   These are adjustable.  The term of of art for the complaining is “surprisal.”   Each layer can presumably kick problems up or down stairs.

It’s as if the entire brain is about managing confirmation bias.

Moving beyond what the review talks about…

A layer whose subordinate notifies is to something it doesn’t expect has three choice, I think.  Expect means: fails to confirm against the model it has.

  • Ignore it: Presumably the problem doesn’t cross the threshold that triggers the layer to become annoyed.  I’d assume there is both a duration and a level component to those thresholds – at minimum.
  • Kick the problem back to the lower level:
    • If the subordinate is a sensor it might tell it that the info is boring and to ignore it.  Which is to adjust the model and thresholds of the subordinate.  I.e. “Yeah, your model is irritating me, please consider this alternative.”
    • If the subordinate is a muscle it says: “Yeah, fix that.”  Or more likely “Yeah, keep working on fixing that.”  The review suggests this is actually how we move.  We set up a model of the end state and then let the muscles work out how to get there.
  • Kick the problem upstairs.  Reporting that the thresholds have been crossed and the model isn’t working.
    • The sensor might report: “Gosh boss I see a snake!”
    • The muscle might report: “Dude, I’m stuck.”

If some high level layer decides to move the pot off the stove it pushes a model downstream, and in that model the pot is no longer on the stove.  The layers below all push model adjustments to their subordinates until the muscles and sensors get to work on moving the physical world into alignment.

If at some point the report comes in that your fingers are burning that gets push upstream and the goal adjusted to resolve the problem.  A bit later the high level system get’s told it’s goals have not been met.  Annoyed some cursing then is planned.

The review has some fascinating things to say.  That there are global control system that can turn up or down the fastidiousness, etc.  Really interesting stuff.

I’m amused about how this has (weak) mimics in software systems.

For example in model-view systems these days people are rediscovering the idea of having a few layers between the model and the screen and using differencing engines to bring them into alignment (aka React).

Or, the way layers of process managers work to assure retry and recovery when their subordinates fail.  Or even how the cable company’s first level support prescribe unplugging the modem.

None of these are as sophisticated as what as Perceptual Processing.  Which is exciting.

Regulatory Stories

It would be fun to accumulate a book of stories about regulation.  My book would be about what a messy complex necessary business this is.

For example, the story of a friend who’s contractor disappeared halfway thru the remodelling job.   When he got another guy to take over the building inspector insisted they remove the dry wall.  The wiring had not been inspected.

Today’s example:  A number of cities started using dry ice to kill rats in their burrows.  It was very cheap and very effective.  Soon, the media reported that the EPA had stepped in to say, “Ah guys?  That’s not an approved pesticide.”  So they stopped.   The media accounts all had this just the facts quality about ’em, but I sensed the underlying narrative was “Yo reader, ain’t regulation lame!”

I noticed story since I use dry-ice when I catch a squirrel trying to eat my house.  It is the recommended technique.

A few days ago New York city started again.  The cities pushed to get the technique approved.  But the story I read had a telling detail.  Apparently what was approved was not dry ice, but rather a product called “Rat Ice” made by some Bell Labs.

Which raises the question in my mind.  Who complained to the regulators?  In New York their original trial run was a park where the poured the dry ice into 60 burrows; so maybe the park’s users complained that the entire park was smoking.

A cynical observer would quickly guess that the rat poison vendors complained.

The Bell Labs is not the famous research laboratory in New Jersey.  Nah, it’s a firm that sells classic rat poisons, baits, and traps all over the planet.  They even have a registered trademark tag line: “The World Leader in Rodent Control Technology®”.  They haven’t gotten around to marketing Rat Ice on their web site.

To me the proof of this is this bit from an article from USA Today that appeared back when the flurry of media reports about how the EPA was telling the cities to stop using dry ice:

Ruth Kerzee, executive director of the Midwest Pesticide Action Center, said her organization raised concerns with regional EPA officials and the city of Chicago about the new rat-killing method.

Kerzee, whose organization promotes minimizing the use of pesticides, said while dry ice is less toxic than some conventional pesticides it remains unclear what, if any, guidelines cities created to ensure the product is being safely handled by personnel.

“We think it could be a sea changer, a great thing to be able to use, but it does need to be vetted and go through the process, so that we don’t end up in a situation where we throw the baby out with the bathwater,” Kerzee said.

The National Pest Management Association, a trade group representing private pest control companies, also inquired with EPA and the Illinois Department of Public Health about the use of dry ice after Chicago launched its pilot and was told it could not be legally used as rodenticide, said Jim Fredericks, chief entomologist for the association. The group published a message to members in its newsletter last month that “any use of CO2/dry ice to control rodents would be a violation of federal law.”

Fredericks said the industry association is not calling for the EPA to permit dry ice as a rodenticide. “It’s not one of our priorities right now,” he said.

There is a joke to be made here about inventing the better mousetrap and “It would be a shame if some innovation where to upset that nice business you have there.”

Mastodon

Yet another attempt to create a social network.  This one’s called Mastodon.  It is analogous to Twitter, i.e. short status updates with following, liking, comments.    Web UI, and apps for assorted devices.   It’s usenet like in with a user accounts residing on nodes and then the nodes stitched together into an exchange network.  Open source with ties to the FSF/Gnu community.

We wish them the best of luck, this is hard rabbit to pull out the damn hat.

Here are some charts based on data taken from this page enumerating some of the nodes in the network.  These are log log charts, and each point is for a single node.  Their equivalent of Twiter’s tweet is being called a toot.  Though in these charts it’s called a status.

A not unusual distribution for an unregulated social networks.  It’s always delightful make up little stories about why there is a node who’s users have made an huge number of toots per user.

The Backfire Effect

You may have noticed that sometimes: you argue with somebody and you come away thinking: “My that backfired!”   Rather than loosening their attachment to their foolish belief they have become more committed.

In years since the effect was named studies have revealed that the effect is common and potent.   They have discovered that some public health advertising campaigns backfire. The target audiences become much less likely to change behavior.  Even bizarrely after the audience admitted that they accepted the facts.

With a public health mindset you can then start to wonder what dosage of facts and information is optimal to change a person’s mind.  Studies that attempted to start to get a handle on that (see links below).  But slight spoiler – it’s really hard! – but not too hot, not too cold.

So what’s going here?  Naturally we all labor to keep a consistent world view.  Whenever new information comes over the transom our minds devote some calories to folding it into that world view.   Let’s call that work skepticism.  It can be defensive, curious, even light hearted  skepticism – smart people take pride in this work.    If the information is at odds with our current world view we are motivated to take the exercise more seriously.  The name for that syndrome is “motivated skepticism.”

It’s not actually that surprising that engaging in the exercise would often strength the existing world view.

That all reminded me of what in back in the 70s the AI community used to call truth maintenance.   Failure to keep the software’s model of truth well maintained was treated as an existential threat to the system.  Because, it’s well known that in simple sets of equations a single mistake doesn’t just lead to bad results; it lets you prove that anything is true.

Here are three podcasts (123) about this.   Part of David McRaney’s the “Your not so smart” series.   David’s turf is around questions of what social science can tell us about discourse, debate, and changing people’s minds.  If you are not into podcasts you can skim the posts enumerated above for an overview and links to other materials.

Sleeper cells

I’ve recently been enjoying a podcast on microbiology.

They recently mentioned that some bacterial infections include a tiny fraction of hibernating cells.   The sleepers are unaffected if somebody tries to murder with an antibiotic.  Later when they awaken the antibiotic is gone and the infection returns.

Bad faddish ideas are like this too.

Why FIFO?

Amazon’s AWS has a message queue system, aka SQS, to which they have recently started adding a variant which assures that your messages are delivered in the same order that they got sent.  I.e. first in first out.

If you are surprised that they don’t do that by default you may enjoy thinking about what would be involved if the Post Office was to decide to offer this feature.

That said, I’ve been surprised that it took so long for this to show up and I remain surprised how slowly they are rolling it out.

So it is with some amusement I read in their doc an example of why you might want this feature.

“Display the correct product price by sending price modifications in the right order.”

Which I think helps explain why Seattle is one of the few AWS sites that supports it so far.

Democracy for Realists – space and time

Some more about this book, Democracy for Realists: …

If we accept that voters do not vote for their policy preferences (and you can read the book if you want to see the evidence) then what is driving their voting behavior.

Here are two models that Political Scientists have put forward – space and time.  Both model presume that voters, being humans, lack the time or talent to engage in a very subtle or complex analysis of what to do with their vote; so they simplify things.  They approximate.

The spacial model: all of politics is boiled down to some simple metric: left-wing v.s. right-wing say.  Or maybe a few two,  like both an economic and social variant of left/right.  The voter then “merely” asks the question how close are these candidates’ metrics to my personal metrics. He then votes for the one closest to him.

In the time based model the voter need only look at his personal experience over time.  He then aligns that with who ever is change during various time frames and votes for the candidates that deliver better outcomes.  It’s feedback loop, and presumably the statistics of large numbers of voters might make this work out nicely.

Again this is Science.  A theory is only interesting if we can proceed to try and disprove it.

The spacial theory is easy to disprove.  You just ask.   Compare the voter and the candidate he selects on the metric.   Questionnaires can dependably tease out where they are on the scales.  For example: Support for lower taxes verse more government services? What the data shows is only the lightest correlation.  In fact in some cases voters do the opposite of what they prefer.  So this theory isn’t helping us.

The problem with the time based theory is two fold.

The first problem: The usual ones found in feedback based systems.  These systems only work if (a) the signal the feedback is based on is accurate and (b) the feedback’s timing is adjusted correctly.  In Engineering school I spent a few years learning how to get that right for simple electronic systems like amplifiers.  In that context if you get it wrong you get nothing or horrific feedback noise.  Big social systems are even harder.  So first off voters get a signal (they lose their job, the weather is lousy, the crop fails, the town has an awesome fair,  the kid gets a lovely teacher) and they sum that up and vote for against the current candidate.   Then we have timing.  This model rewards the politicians for taking actions that have short term benefits; i.e. they show up in the voter’s impression before the next election.   Worse, long term benefits will accrue to the account of some other guy.

Like the spacial model voters have a very noisy model of the candidates.  In this case the their model of credit/blame is very poor.

So what are two models worth anything?  Turns out yes.

The spacial model is the gold standard for understanding legislatures.  While it’s useless for discovering how a voter will pick his candidate, it useful for predicting how Bob, your legislator, will choose to vote on any given bill.   This is good news:  Bob is fairly well informed about the position taken by the bill.  On the other the voters who elected Bob do not have a good model of Bob.

The time based model is actually quite predictive of how voters will behave.   But, oh my, they are largely miss informed about blame/credit and their sample is narrow minded.  They only look back a few months.  This is not good if you want responsive government.   It is useful if your placing bets on an election.  You can do a damn good job of predicting the outcome of elections by measuring just GDP growth over the last few months.

While these models are not as useless as the folklore model ( i.e. that voters give their votes to candidates who reflect their personal policy preferences).   But if your goal is to explain how Democratic governance is responsive to the voters preferences; they they aren’t going to help you.

More to follow…

Democracy for Realists – acting on falacies

Part 2 – So let’s step into this book a bit.

The reason to prefer a realistic view of politics is fear.  Fear that your unrealistic premises will lead to unfortunate outcomes.  So political scientists have spun up models for voter behavior. And then, tested them!  if you want to win elections it’s probably best to pay attention.

Personally my thinking about politics was entirely up-ended by the work on the voting patterns in Congress.   This book may be forcing a major resorting in my head.  I’m not sure how that will settle out.   It’s very discomforting to think that the model I took on board from that book might be wrong, that I’ve been extremely deluded.

Books that are attempting to force a painful dose of realism into their audience probably need to spend a lot of time addressing their audience’s bogus beliefs.  Scientists to this with studies, data, statistics.  It takes years to convince people that the world is not flat, the sun doesn’t spin around us, that punishment is effective, that bleeding out the bad blood doesn’t help.

So let’s start with the most most popular model of how democracy works.  It’s widely presumed that voters vote their preferences.  Say Sam is extremely concerned about Global Warming.  We’d assume he’d seek out the candidate who is most aligned with his concerns and then vote for him.   What the data say?  The data says:  NO!

If you take that to heart you really need to stop taking seriously sentence like:  “The voters, outraged about X, voted for Mr. P.”  Because it’s not true!  Talk of the “will of the people” is aspirational, but it too is not true.  The whole idea of a mandate splits thru your fingers like sand.

Good science is all about disconfirming models   Postulate a theory/model and then see if you can prove it’s wrong.   The audience may hate that, they may love the model, but science doesn’t care.

So this first model of politics in the democratic states is wrong.  The authors call this the folklore theory.

Once it became clear that the folklore theory doesn’t fit the data the political scientists went looking for other theories.  But that’s a story for another day.

<X> for Realists

I’m currently reading a book titled “Democracy for Realists: …”   It’s triggered a bemused fantasy about a series of “… for Realists: …” books.  In the tradition of those “… for Idiots: …”

I remember books stores.  They had lots of shelf space for self help books.  A popular genre. Let’s imagine some titles:   Schooling for Realists, Vacations for Realists, Project management for Realists, Home brewing for Realists, Gardening for Realists.

So’s why not?  I have my theories.  For example picking up a book of this title would seem to signal one’s appetite for disconfirmation.  Where’s the fun in that?  Or possible like the Monty Python argument skit it implies your shopping for a scolding or abuse.  At minimum it would seem to signal that the author is war weary, scarred, old, cranky?

One take on self help books is that they are selling a treatment for stress.   Realism doesn’t sound like a miracle cure, more like chemo.