Miscellaneous notes 2021

This document contains only my personal opinions and calls of judgement, and any comment is an opinion, in my judgement.

[file this blog page at: dig del.icio.us Technocrat]

These notes are opinions about miscellaneous (non-computing) topics, often brief informal reviews of products or shops or places ranging from canned food to pubs.

20210105 Mar: NN Taleb is doing operations research

When NN Taleb writes if I am uncertain about the skills of the pilot, I get off the plane he is not making a point about probability or statistics, he is making a point about operations research, as he is arguing about the expected payoff of an operation (a decision).

Note: that it is operations research is also trivially obvious from his description of his interest being risk management, as one of his insights is Risk management is concerned with tail properties and distribution of extremes, not averages or survival functions.

The point that he is making across his work is the (subjectivity's) one that we study probability and statistics for operational reasons, that is to essentially make useful bets. Therefore we should want to know about not just simple summaries of samples and populations, but also about the shapes of their distributions, especially if they are highly asymmetric, because the operation that needs to be informed by such knowledge is betting, that is making decisions.

Note: The betting approach is clear even just from the title of his paper How to Price an Election: A Martingale Approach.

So in the case about uncertainty about the skill of the pilot, that is we don't know which is more likely, that the pilot is skilled or unskilled, we compare the payoff of betting right or wrong that the pilot is skilled or unskilled, where the payoff of a successful flight is some advantage like a profit from a deal, and that of an unsuccessful one is losing limbs or life.

If we bet wrong that the pilot is skilled the likely payoff is hugely negative, while if we bet right the likely payoff may be moderately positive, and if don't know which way to bet, the safe bet dominates.

That it is an operations research is easily demonstrated by changing one of the payoffs: if one needs to flight to have a life-saving operation that cannot be delayed, then the payoff of a successful flight is huge, and between the likelihood of death if one does not fly, and the uncertainty of living if one flies, the payoff of the former overrides the uncertainty.

More generally statistical summaries that result in expected values are good for betting only in the long run, assuming that the long run exists and the outcomes are erotic.

20210310 Wed: The difference between ergodicity and statefulness

NN Taleb is thankfully spreading more awareness of better understanding of probability and statistics but sometimes he simplifies a bit too far, for example:

As we saw, a situation is deemed non ergodic here when observed past probabilities do not apply to future processes. There is a “stop” somewhere, an absorbing barrier that prevents people with skin in the game from emerging from it

The problem here is that the word ergodic is often used in several different meanings, but its proper meaning is not the one used here, because the quote above describes statefulness rather than ergodicity.

Ergodicity is sort of a primitive concept so it can be defined in many ways, but it is a concept most relevant in information theory to describe a type of source of signals under sampling. Fundamentally ergodicity happens when the sampling is from an unchanging population, that is the samples may be biased but not misleading. A radio station that transmits only classical music is ergodic, one that transmits 50% classical and 50% pop is ergodic, one that transmits classical, jazz, talk radio etc. unpredictably is not ergodic.

Ergodicity is very related to one of the meanings of entropy, the one in information theory; a source of signals (samples) is the more ergodic the lower its entropy is, that is the less its samples are surprising.

That has as such nothing to do with statefulness, and it is statefulness that creates absorbing barriers.

There is however a link between ergodicity in information theory and statefulness, and it is that in thermodynamics there are notions of ergodicity and statefulness that are tied to that of statefulness, and indeed that is a pretty common situation. But it is best to clearly separate the concepts of entropy and statefulness.

20210420 Fri: An update about safety can openers

Some years ago I wanted to buy some safety can openers and there are some updates after a few year:

• All three worked fairly well.
• The HeyLR lost its cutting edge rather sooner than the others, but it was also the cheapest.
• All three are no longer available in that shape, but have been replaced by equivalent models.

There is now a similar model the Culinare MagiCan Opener which is the one that I am trying next. I also figured out two features to avoid:

• If the head is made of plastic the opener will stop being good at cutting sooner, because it will deform sooner because of the pressure. The rotating blade and the gripping cog should be fixed onto a metal base.
• Some safety can opener cut the lid on the top instead of the side, and they give in effect only half safety: the can indeed is cut neatly without sharp edges, but the lid is cut also cleanly with very sharp edges, almost like a blade.

20210620 Sun: Why vendors prefer mobile phone apps to web sites

It is often a question why on mobile phones vendors prefer for users to install their apps to them using the web mobile site; in most cases the app is actually a somewhat customized browser core, usually Electron.

The most likely answer is simple: users who install apps tend to give them all the requested permissions, which are usually a lot wider than the permissions the equivalent web site would be given by a generic browser accessing it: generic browsers are in part designed for the interests of the users, while apps as a rule designed only for the interests of the vendor.

20210729 Thu: Which cookie settings to enable

Many sites, especially if accessed from EU countries, offer options as to which tracking cookies to enable. To avoid hassle many people just enable all of them, which allows a lot of online surveillance. But often choosing a subset makes sites less useful or more annoying later.

There are usually two distinct sets of settings, LEGITIMATE INTEREST and CONSENT, where the latter enables the more intrusive tracking, plus various categories in each.

The most important setting to enable is often under CONSENT and it is to that to Store information on a device. This sounds ominous, but it is essential to enable the cookie that stores the other tracking settings. Without enabling this the tracking settings will have to be re-specified for each page.

Apart from that it is often useful to enable at least the LEGITIMATE INTEREST level of two other tracking settings:

• Functional: many sites use these cookies to maintain state between pages such as their logic sequence or choices made my users in earlier pages. Sometimes they require repeated entry of choices if they are disabled.
• Content Performance: there is little harm in enabling this as the content is anyhow provided by the site and accesses are logged regardless.

Some site offer a tracking settings menu that initially offers only a choice about the Consent level settings and they are unthreateningly all REJECTed by default, while the LEGITIMATE INTEREST settings are in a sub-menu and are all turned on by default. Please remember to OBJECT to them and only enable those you really want.

20210821 Thu: Shafting the customer, O2-UK edition

Most large companies have huge pricing teams whose only purpose is to find new ways to shaft their customers by finding ways to have complicated fee structures that charge more than headline prices seem to be like.

This is part of the general corporate strategy to aim to offer the worst possible service at every price-point, in order to get customers to upgrade to the next price-point and pay more than they would otherwise. Which indicates that competition is largely absent as otherwise corporates would try to compete on simple and good value, rather than trying to fool customers more than competitors.

A particular example I recently found is the O2-UK monthly tariff, which every year increased by RPI plus 3.9%. As always the increase in prices is RPI as it is larger than CPI (which is instead used for increases that benefit people), but why the additional 3.9%, which means that prices increase by 50% every 10 years? Do the costs of telecom companies really increase by 3.9% above inflation? Of course not: it is just a clever way to shaft the customer every year a bit more, hoping that they don't notice.

The model chosen by mobile telephone companies is to assume that customers dislike to change phone numbers, and therefore to price the initial tariff low, and then pile on increases and fees once they are locked into their new phone number.

What are the solutions for customers? I think three counter moves:

• Use different accounts for making and receiving phone calls, because corporates lock customers in on the phone numbers used to receive calls, but customers pay on accounts used to make calls. It helps considerably to have mobile phones capable of using two SIMs, one for calling, on for receiving calls, and having a backup phone and SIM for receiving calls.
• Have at least two accounts with two different vendors for receiving phone calls and use both numbers for things like two step authentication and contact numbers for banks etc.; this is also essential in case a mobilephone or a SIM is lost or stolen.
• Change the vendors used to make phone calls every 1-2 years.
• Every few years also change the vendors used to receive phone calls, using phone number portability.