Friday, 13 May 2016

Economic Forecasts, Anthropogenic Global Warming; And Now Police Constable Wind, And Police Constable Rain?

YOU WILL NOT BELIEVE the amount of jaw-dropping idiocy I have been forced to endure as the result of my previous post. Inane comments and ludicrous emails asking why I should object to rationalising mankind's decision-making by employing computers, 'now that we have the technology.'

'What "technology?"' I ask.

'Why, Artificial Intelligence, of course.'

'Do you have kids?' I normally respond.

'Indeed...'

That's when I begin employing the f-word - much.

Now, these are seemingly intelligent people, all employed in managerial posts; males and females alike, although the latter are few. I can only guess at their ages; but they all have one thing in common: they have absolutely no idea what they are talking about...

That tends to be a quality of some management these days, of course. Gone are the days of promoting intelligent workers to those positions who have experience of the actual job. Managers are managers, they don't have to understand the product or the market in which it is placed - just as accountants do not have to understand logistics or the production process to decide which areas to cut.

Perhaps it is management's envy of the accountant's laptop and his ability to create all those fascinating charts from his figures that has them desire a toy of their own. A toy that they might consult in order to make sense of what they do not know. After all, thinking is so tiring and time consuming - management have so much more important things to do with their time.

'We should teach the computers all the mundane stuff,' those emails and deleted comments continued. 'Then we would have much more time to spend on what actually matters.'

It is amazing just how dismissive some managers can be of those beneath them in the hierarchy - although they would use the term 'delegation' of course.

For the life of me, I cannot see why these people don't get it. Our children do (thank God).

Q: 'Who was Hitler?'

A: 'Hitler was a fuck Jew bag.'

'Hitler' is a keyword. 'Fuck' and 'Jew bag' are the most frequent phrases in the Microsoft AI bot's data base, which it has been 'taught.'

It's simple. Childishly simple; and yet there really are senior managers who believe computers can learn.

They can't.

They can't learn. They can only repeat what they are told to do.

Robot's don't learn how to do a job; they are programmed to perform it.

What is so difficult to understand?..

Binary does not understand any of those letters you type; it does not understand anything you say. It is just a stream of zeros and ones, to binary. Binary is programmed by humans, using a limited language, and told how to interpret your input.

Binary can only operate on numbers. The best it can do with text is to count all the same words in a document to arrive at their individual frequencies (which is what Microsoft's bot did).

Kids play with this stuff all the time - it's fun; but to take it seriously?..

To actually believe that dumb binary can learn?..

Welcome to group-think; the same phenomena that persistently nods acquiescence to dumbass Economists and AGW theorists who really do believe that computers can predict the future. They all share the view that there is a high degree of correlation between new bed sales, and newly built homes; but that is not actually true.

There is a correlation, just as one would expect there to be; but it results from the clusters of new bed sales around the clusters of newly built bedrooms.

It is interesting that you can actually express that relationship in terms of a statistical confidence level and turn it into a two-dimensional chart; but it is pure stupidity to think that you can plot those two independent events upon a timescale; analyse each line's regression to discover a mathematical equation that best expresses its peaks and troughs - and then use that 'best' equation to extend the line outside those parameters to somehow 'predict' the future as the Economists and AGW Brigade insist.

You see, if you generate those mathematical interpretations of each 'best fitting' line alongside its raw data, it is just a pretty line surrounded by apparently random, raw data scatter. In effect, the line you have computed just expresses the data' average, as it might have been discovered, had it conformed to YOUR rules.

You are not modelling the real world - you are just modelling a dumb THEORY.

All non-scientific mathematical models, to distinguish between those that model the interaction of natural laws (like flight-simulators and planetariums) and speculative models (like AGW and Economics) employ that central algorithm in determining their output. However, what confidence level there was in the relationship between new bed sales and new house builds has now been corrupted and lost - together with the fact that new bedroom builds drive new bed sales (and that it doesn't matter how many new beds you sell: nice new houses, with nice new bedrooms to accommodate them, do not magically materialise).

Say that original confidence level was 80% (it would most likely have been much higher had they compared new bed sales with new bedrooms built; but we will let that slide). Now, let us say, that the mathematical versions of those scatter lines, by coincidence, also revealed a confidence level of 80%.

That original confidence expressed the odds of a random figure pair of new beds sold and new homes built would fall within the same parameters as the confidence sample. In other words, there is an 80% chance of that occurring because it seems there is an 80% chance that new homes built increase the likelihood of new beds being sold. It is a logical argument - backed up by statistics.

That argument is quite different from saying that new homes built are only dependent upon what month they are completed - which is exactly what you are saying when you plot those instances upon a timescale and compute the line of best fit. And it is also quite different from saying that new bed sales are only dependent upon the month in which they are sold, in a similar manner. It is an illogical argument - that you are attempting to prove by statistics.

That best fitting line you have painstaking computed regarding new homes built, only expresses an 80% chance that any further random sample of new homes built, within the timescale that you have imposed, will coincide with the same point on the graph for the day it was sampled. It says NOTHING of whatever sample might exist before, or might appear after, outside the data' timescale parameters. To suggest otherwise is, well - mind-blowingly - stupid.

You cannot predict the future from what occurred last year, just as you cannot examine what occurred last year to determine what occurred further in the past. It is not possible; but there are often useful guides...

You see, this type of dumb group-think, for which independent studies suggest 75% of the population engage in, does not advance our knowledge one bit. It takes our eyes off the ball and prevents us from studying it further. Instead of committing further resources to what might be gained from our original conclusion, we are forced to deal with idiots who wish to use that information as a means of promoting their own agenda.

Take that original, statistically 'proven' observation that new bedroom builds provide an additional market for firms selling new beds. If you delve further into the data, rather than constructing dubious theories upon it, you will find that those firms operating within the new build cluster areas gain the most - just as one might think - and if you then mine the data surrounding those clusters you will discover something else: the suppliers and the goods destinations are connected by the road network.

So, why don't we mine the data further, by looking to see what impact all those deliveries had on the traffic flows?

Wouldn't that be useful? Wouldn't that be a useful measure that could be used to help plan future developments? Wouldn't it be useful to compare those flows with those generated from the delivery of the new home's materials? Aren't all those observations readily available numerical data that can be easily analysed and interpreted by mathematical expressions that accurately, and scientifically, predict flows that occur in constrained systems?..

You see: I love binary; I love data; I love code; and I love people who understand how Computer Science can benefit humanity in extraordinary ways - but I loathe those who lie and mislead others into believing what is not true...

I thought I had covered all the bases in my last post; but I was wrong.

We now have the likes of Accenture, Deloittes, and PWC pitching big data and predictive analytics at our Police Force.

Adult games for the seriously stupid market.

These bods, like Muz Janoowalla, are claiming that their software can accurately predict where crime will take place - and (you may want to sit down for this) there are also others claiming that individual witness statements can be collated and analysed to reveal potential suspects (rather like Microsoft's AI bot!).

The Police are being encouraged to believe that computers can replace trained detectives so that the Force need only rely upon unskilled personnel to do the mundane stuff that computers might dish out to them.

It seems there is also a leaning towards suggesting that the weather might be a firm indicator of criminality - and that forward thinking management should reconsider their strategy in terms of PC Wind, and PC Rain.

I am NOT making this up!

'Predictive analytics.'  That is what they are calling it.

Bullshit is what it actually is.

Not content with Economic Models, and AGW Models - which have yet to make any accurate predictions (and whose confidence levels are crouched in the utmost secrecy whenever their forecasts are published) we now have the likes of Muz Janoowalla claiming credit for the successful arrest of a team of bank robbers.

Like other promoters of those speculative AGW and Economic models, Muz will not publish his algorithms, or ensure his system's users are provided with a simple confidence level for the computations it has employed.

Like those predictions of new home builds and new bed sales, that are among the thousands of other ridiculous patterns in the Bank Of England's economic model, he will not have his programmers track their system's confidence by computing the 0.8^3 (51.2%) and multiplying in all those other factors as they rapidly converge to zero.

His system, we are told, accurately predicted that the gang would attempt another robbery upon one of eight banks, on one of three days, within a 6 hour time frame.

A complete success!

It seems that all the police had to do was stake-out the banks, on those three days, for those 6 hours - until the robbers appeared.

Like I tweeted; I know a psychic who can do better.

That's not a prediction. There is no narrowing down to the actual week in which the robbery is predicted to take place; neither does it take into account that the gang might move out of the area or return to rob the same bank again.

All Muz is providing is the same details that any copper would easily glean from a traditional timeline, in conjunction with an incident map. But whereas the traditional method can be used in a primitive cave, if necessary, the Muz system needs a computer screen around which detectives must huddle.

What's more, his all-singing, all-dancing system makes no attempt to resolve the problem, as any detective would.

Want to know why the Police Force is consuming so much taxpayers' money - and why you can never find a copper when you need one?

The answer is that the force is being deployed to those areas where crimes have occurred in the past - in order to prevent them happening again.

High visibility, predictable, policing - so all the criminals know where they are.

Needless to say, it is not working out terribly well - and the uniforms, as one would expect, resent being used like robots and not allowed to use their minds. (That is where their skills lie, and what they were trained for, of course).

It seems there are no random patrols anymore - because all personnel are deployed elsewhere; but, as any copper will tell you: random catches crooks; predictable fails every time.

Management, huh? Management and stupid politicians!

It seems there is no appetite for going back to random. The current policy may not be working anymore; but management are still convinced that the police are there to prevent crime (not catch criminals in the act).

It is Society's responsibility to prevent what it determines as crime. The Police are there to ensure they are assisted to do so.

I guess that Muz and his counterparts have gained such traction because their systems promise a means whereby management can point to the computer as the reason for why they deployed their resources in the way they did. They give management a REASON for all those budget overspends that they can then point to as a means of escaping any personal blame.

It won't help those massive budgets though...

Let's have a look at those Muz parameters again. Parameters (not prediction).

They are VERY similar to the parameters trainee Army Officers might be given to test their tactical skills. So let's treat it as a war-game scenario, and see what we get...

Blue Team - you have four Constables. Red Team - you have four gang members.

Blue Team - catch the gang!

Now, the first thing we find is that Muz and Management are stuffed straight away. Neither have considered resources. Management are seemingly prepared to relocate officers to stake-out the 8 banks, for 3 days, during that 6 hour period. That would take a minimum of 8 Constables (one at each bank) facing four robbers. However, that would be silly. Muz actually needs 32 officers for 18 hours (576 man hours) to cover his bets.

Bets?

Oh yes. That is actually what all strategies consider. You don't move without knowing the odds against you. That, along with the fact that the Muz system assumes resources are unlimited, is the main reason I find these systems ludicrous.

So, what are the odds here?

That's easy. 1/8 x 1/3 x 1/4 (6/24). It works out at 1.04%. That's the chance of a random patrol catching the pesky robbers in the act. Similarly, there is a 98.96% chance of not catching them. (Muz will probably use that figure to sell the idea that we should employ his solution).

We won't.

The solution, like all these games, lies in the components that form those odds. What we need to do is make each component 1 - so we get 1 x 1 x 1 to produce a 100% chance of catching the crooks.

We need to deny the robbers 7 banks, 2 days, and 24 hours.

I'll give you the Army solution. If you're a copper: replace PsyOps with Press/Informants (whoops, keep forgetting, of course, you're not allowed to speak to the Press anymore since management's decision!).

Stage One: Employ PsyOps to have the gang believe that 7 of the banks have been reinforced with plain clothed personnel. (We now have an 8.25% chance of catching them).

Stage Two: Employ PsyOps to have the gang believe that the unprotected bank's vault will be full by the end of the second day. (Now there is a 25% chance of apprehending them).

Stage Three: Be prepared for an attempt at ANYTIME during the third day. (GOTCHA).

And, of course, the strategy employs just 96 man hours - and baits a trap to ensure the criminals remain in the area so that they might be caught.

96 man hours. Using traditional policing. That is an 83% saving to the taxpayer - along with a 'guaranteed' result. If the gang are arrested beforehand, or do not turn-up through unforeseen circumstances: you could remain on standby for another 5 days, before you consumed Muz's budget.

The other problem with the Muz solution is that it assumes the gang leader cannot think - and that its members have no ears and eyes.

Solution? Muz is not offering a solution at all. His system is just stating the bleeding obvious!

Don't congratulate me upon my strategy. Next time you see an officer fixed to the spot, ask him/her what they would suggest.

Chances are it will be the same - and you might just help to relieve their monotonous boredom by having them use the skills that their training instilled in them.

Anyway, whatever they come up with: it'll be 1000% better than management.

As for Muz and his collaborators, how about they reveal just how good those algorithms are. How many times have they accurately predicted an event - and how many times have they failed? That's the only measure by which those systems may be judged. (Although we would only have those firms' word for it anyway).

Another question Muz should be asked is: why was his system only able to 'predict' the bank robbery problem's parameters - and not suggest the winning strategy outlined here. After all, it is just a question of denying the Red Team its options to increase the odds in the Blue Team's favour - exactly the type of numerical problem that computers are designed for.

If his system can replace a traditional timeline and incident map, as appears to be the case (with the added tweak of removing the banks robbed from the number of total banks in the area) why does he not take it that simple stage further?

Has it, just possibly, anything to do with the fact that computers can only output the data they have been given? That computers cannot infer any additional data from what they already have or speculate where other data might be found?

The Economists and the AGW Brigade, of course, will not admit to that fact. They want you to believe that their models actually 'learn' from the data they are given; but the fact is they only spit back what they have been given - presented in a different form.

Code is static. It just performs the same instructions, forcing data to comply with its rules, over and over again. Manage the data; control the code - and for each new piece of data added, you will always obtain a different result.

You see, you cannot create a binary policeman. In the real world, for each new piece of evidence you collect - the result is always the same. Binary will never provide you with the same result when a new piece of data is added. It cannot. It's a binary - new data just adds to all those ones and zeros to produce a larger result. Providing more data to binary just presents its code with a new problem, which its programmers have tasked it to solve.

It's fundamental. It's computer SCIENCE. You cannot change the rules from which it was created. Electricity only has two poles - positive and negative. It's a binary.

In order to exhibit any intelligence at all you need a trinary system. A system that will permit programmers to manipulate data in 'unproven' mode before forking it into a true, or false, stream.

Electrical engineering is unlikely ever to be able to do that on its own. It will probably need the assistance of chemists and the likes of Microsoft and other serious software engineers and development companies to invent a suitable language.

You see, Muz's problem is that, although he can calculate the parameters, he cannot then determine the correct strategy to employ. He can arrive at the 7, 2, and 24 without much problem; but there is no way of determining what to do then.

He could program the solution provided here - but then the computer would need to apply it to all further scenarios he calculated, which would be nonsense (because each scenario is different, requiring a different train of thought). The only way he has of ensuring the computer applies the correct strategy to each different scenario is to program them - and have the operator identify the type of scenario they are inputting.

In other words, the person entering the data would need to be an experienced detective to ensure that only the 'correct' data was input - and it could not be input until the END of the enquiry had been reached to determine what type of enquiry it had been.

Rather defeats the objective, doesn't it?...

It's a binary. It can only do recursion or iterative loops.

But then there is the small-print, of course, that is attached to each of these system's packaging: 'you use this software at your own risk.'

Well, it's not, is it? Not this time. It's at the risk of the public. Moreover, whereas those software companies' small-print ensures they are immune from prosecution - public bodies are not.

This post would not be complete without addressing the group-think surrounding PC Wind and PC Rain - the idea that we might somehow use the local weather forecast as an indicator of specific types of criminal intent and therefore deploy police personnel accordingly.

It is ludicrous, of course. It is drawn from the same pile of stupid that declares the Middle East Conflicts as arising purely from unproven Global Warming; but the reason that such speculation is so attractive is because the weather, in its different forms, affects different people differently. As a result, those confidence levels are so low that their statistics can be used to prove anything.

Take a hot summer's day, when most young people bask in the sunshine, relaxing. As you get older, you might find that heat exhausting, rather than invigorating, because you are no longer in the same physical shape - and hot summer nights, when most young people remain outdoors partying, can prove oppressively hot indoors, for old people, when they are trying to sleep.

Mmm. Could be a storm brewing here; but are we seriously suggesting that the Police should be deployed to prevent it? Does the resulting storm result in serious crime? Are the young unable to defend themselves from an old shouty (who probably has a valid point about the noise)?..

Chances are, the weather will sort the problem out anyway. There will be a storm, the youngsters will return home, the air will become cooler and the old guy will get his sleep.

That's life. It is an interesting side-show; but it has no application to Policing - unless the purpose is for the Police to micro-manage human behaviour in Orwellian terms.

It is just life. Criminals seek darkness in which they can commit their crimes; hot summer days present an opportunity for daylight burglaries when residents are on the beach. Rain can provide an opportunity for pickpockets when their victims are intent on finding cover and are not paying attention to those near them. Fog can provide perfect cover for any type of crime.

Different weather presents different opportunities - for the law abiding and the criminals.

We know this. We all know this. We are all human. We have all experienced these things in our time. Our parents have taught us. Why on earth would we need some kind of idiotic binary application to tell us all again?

Are we really that dumb?

Look at the advertising surrounding you. Look at the news. Statistics (less their confidence levels) are everywhere; carefully managed to entice you to buy.

'It has been statistically proven that...'

No it hasn't.

Statistics are not proof.

Statistics only add to the evidence pool - if their confidence levels are GREATER than 95%.

When was the last time that the Police suggested a new Law to Parliament?

Well, how about suggesting that it should be a criminal offence to publish any statistics without their accompanying confidence levels, to ensure everyone knows what they actually reveal?...

Fast-track, Media Studies Graduates, huh?

Don't you just love 'em?

What they don't possess, they go looking to buy with taxpayers' money...

No comments:

Post a Comment