The inevitability of punishment, the effectiveness of sudden checks
This is a philosophical note about management and upbringing, as well as a very unexpected insight associated with modeling digital mold. Inspired by conversations about the problems of construction management, as well as a network of essentially remote branches.
Fighting digital mildew populations under the influence of a sizzling sun.
Preliminary notice. The article consists of three parts, and the first two seem to be completely irrelevant to each other, but there is a third one that unites them.
The second warning. There are different ways of motivation and there are adherents of the approach in which punishment (and generally negative reinforcement as a whole) does not apply. Not bad described in the book of Karen Pryor "Do not growl at the dog!" I highly recommend to anyone who has not read. However, this note is about a different approach, and more likely about the effectiveness of different options for negative reinforcement.
The third warning. In my views, I am an evolutionist and I am confident that natural selection occurs not only among living organisms but also among professionals, entrepreneurs, companies and enterprises, peoples and sometimes even ideas.
I have repeatedly heard about the principle of inevitability of punishment, and that it is not the amount of punishment that is important but its inevitability, and then such a measure of action will work more efficiently.
But life is difficult and to produce continuous continuous monitoring is expensive and impractical. In response to this, mankind invented the concept of sudden and unscheduled inspections. Whether it is periodic surprise verification work at school, or fines for non-compliance with traffic rules, which are distributed by car inspection cars hidden in the bushes, unscheduled audits of enterprises, test purchases, etc.
Let us briefly fix on the example of the traffic regulations: A camera that stands and finishes all with violation is an inevitable punishment (we do not include those who sue here). The crew of traffic police in the bushes which stands in a random place and selectively controls the speed - a sudden check.
In theory, sudden checks should work as follows: a person understands that at any moment his activity can be checked, therefore he will act in compliance with the norms. And actually, if it works normally, then indeed, unscheduled inspections will reveal the absence of violations and everything will be fine.
However, if we look around, we will suddenly find out that unscheduled checks, to put it mildly, are not as effective as we would like.
I played with the mold simulation, based on the article habr.com/post/418545 Also as described by the author, in my implementation, each cell has its own genome and executes it. Maybe someday I will publish my implementation. The essential difference is that different genes have significantly (about two decimal orders of magnitude) different execution times. Well, for example, to make a decision on the basis of temperature is easy. And making decisions based on an assessment of how close your neighbors are to you (for this you need to compare a lot of external signs) many times harder.
Initially, I made the same internal energy costs for any team.
Having added another complex gene and having started this case, I found a significant performance decline. The analysis revealed a population of freaks who used very long genes for their execution, and it must be said very effectively, so that everyone who mastered these genes was a rod with terrible force and crushed the rest, but wanted more variety and competition of species. And here I came up with the idea that the costs of internal energy for the execution of a gene can be made proportional to the time it is performed. And I stupidly entered the time measurement and multiplied the spent nanoseconds by a factor and subtracted from the internal energy of the cell. It seemed to me a good decision. fully automated consideration of the complexity of the execution of the gene and, in general, was well placed on evolutionary ideas, let those who really use complex things survive,
The launch showed even greater sinking of performance, the timing of the measurement was quite expensive on the platform used, the genome interpreter code was written quite productively and began to outweigh the measurement against the background of the usual interpretation. However, the malicious exploiters are very expensive but effective operations have ceased to firmly dominate. I checked, the gene is and sometimes works, but now not so mindlessly often. Diversity has returned, again it was possible to observe the war of multi-colored mold.
And then I thought, and let me take a selective measurement, with a probability of 10%, I will measure the cost of time, and the time-to-energy conversion factor will increase 10 times. On average, the impact on the system will be the same (I thought naive), and the time spent on metering will fall 10 times.
I start and again all the same geezer who very actively use a heavy but profitable team are expanding with a terry color. What? Why? Once again I turn on the measurement at each iteration and they are not so twig. I turn on selective metering and they are there again. Uniform costs and 10 times increased costs applied with a 10% probability give a radically different result.. On dull mold. And the thing seemed like this. The usual costs of this gene were large, but affordable for the cell. However, the costs increased 10 times turned out to be overwhelming and simply completely killed the cell on the spot. And now let's estimate. Malignant exploiters of this gene, for example, execute it once every 5 cycles of simulation (this is very common). And with a 10% chance they will be killed and killed at this. This essentially turns into a condition - do what you want, but every beat of 2% of your population will fall into the use of a very expensive operation and they will be killed. And in such conditions, these genes again showed an advantage and maliciously crushed all the others. All that should make such a population - to multiply more than 2% in time with the simulation, which they successfully did. And worst of all, an even larger increase in the fine for these genes does not change anything at all,
Those. it turned out that instead of the price for this gene there is an accidental punishment (murder) of those who use it. On average, integrally, the energy costs are exactly the same. But the result is radically different.
And here for me in the new key, the meaning of the statement about the inevitability of punishment and about was opened, then why the sudden checks do not work.
There are illegal receptions in business. For example, take the sellers who hang on the market (manipulating the scales will sell you less than they sound). Those who do not cheat have one indicator of profit. Those who begin to deceive begin to extract substantial additional profits, and even can put very low price tags, thereby attracting a buyer. If for each body kit the seller would be punished with a small fine that makes this kit unprofitable, then no one would do that. But let's consider the situation, if we periodically, at a random point in time, conduct test purchases and issue a very large fine that will take the seller out of the market. Suppose the seller in 30% of sales decides to deceive the buyer. And we also assume that we reach every seller with a control check on a random day once a month, which means about a 3% chance that this seller will be checked that day. 3% * 30% chance that he decides to deceive our controlling buyer gives 1% of what he will be discovered that day. Suppose all sellers are. This means that every day we will catch on the fraud 1% of our sellers and write them a huge fine and they will leave the market. And now interesting. 1% penalty means that 99% of the violation will not be noticed. Now we estimate for 365 working days, what is the chance that they will never catch you? 0.99 ^ 365 ~ 0.025 Ie 2.5% of shopkeepers who every day deceive 30% of buyers during the year will not be caught by such checks. And they will have the best profit. And they will be much more competitive than those who do not use such a hack. Similarly, it will happen, for example, with people who steal from work.
The worst thing in this situation is that systematic violators of the rules appear and who never come across. And for this, no conscious activity is needed. They will simply be because it is profitable to violate, those who got caught flew out and those who didn't get caught remained. Thus, by assigning such sudden checks, we ourselves create those who still break the rules and do not come across. And people still talk to each other, take an example. In the case of theft, a conditional Petya will appear, who regularly privorovat, also regularly passes spot checks and has never been caught in many years.
The result of the whole note: introducing random checks, we ourselves will create those who regularly break the rules and have never been caught. And they will have a bad influence on the rest, as a result we will have a lot of violators, constant catch of violators, we will dismiss them, but this will not reduce the proportion of violators.
For me, this was the key point of understanding why the scheme with random checks and instant dismissal did not result in good discipline.
And just as well with safety techniques. Violating it faster and easier to do their job, but because they are not always killing Uncle Vanya, “yes, I did it 100 times, not ssy.” That is why an employer interested in observing TB must constantly monitor its observance and the key point must be consistency, not the size of the penalty.
Creating conditions where breaking rules does not work much better than rare but very strong punishments. And now the mechanism has become completely clear why this is so.
Separately, gambling fell outside the scope of this article. Those. the excitement on the subject was not caught. And it also has a crucial role in the philosophy of breaking the rules or theft.
Fighting digital mildew populations under the influence of a sizzling sun.
Preliminary notice. The article consists of three parts, and the first two seem to be completely irrelevant to each other, but there is a third one that unites them.
The second warning. There are different ways of motivation and there are adherents of the approach in which punishment (and generally negative reinforcement as a whole) does not apply. Not bad described in the book of Karen Pryor "Do not growl at the dog!" I highly recommend to anyone who has not read. However, this note is about a different approach, and more likely about the effectiveness of different options for negative reinforcement.
The third warning. In my views, I am an evolutionist and I am confident that natural selection occurs not only among living organisms but also among professionals, entrepreneurs, companies and enterprises, peoples and sometimes even ideas.
Part one. Inevitable punishment and random inspection
I have repeatedly heard about the principle of inevitability of punishment, and that it is not the amount of punishment that is important but its inevitability, and then such a measure of action will work more efficiently.
But life is difficult and to produce continuous continuous monitoring is expensive and impractical. In response to this, mankind invented the concept of sudden and unscheduled inspections. Whether it is periodic surprise verification work at school, or fines for non-compliance with traffic rules, which are distributed by car inspection cars hidden in the bushes, unscheduled audits of enterprises, test purchases, etc.
Let us briefly fix on the example of the traffic regulations: A camera that stands and finishes all with violation is an inevitable punishment (we do not include those who sue here). The crew of traffic police in the bushes which stands in a random place and selectively controls the speed - a sudden check.
In theory, sudden checks should work as follows: a person understands that at any moment his activity can be checked, therefore he will act in compliance with the norms. And actually, if it works normally, then indeed, unscheduled inspections will reveal the absence of violations and everything will be fine.
However, if we look around, we will suddenly find out that unscheduled checks, to put it mildly, are not as effective as we would like.
Part two. Digital mold
I played with the mold simulation, based on the article habr.com/post/418545 Also as described by the author, in my implementation, each cell has its own genome and executes it. Maybe someday I will publish my implementation. The essential difference is that different genes have significantly (about two decimal orders of magnitude) different execution times. Well, for example, to make a decision on the basis of temperature is easy. And making decisions based on an assessment of how close your neighbors are to you (for this you need to compare a lot of external signs) many times harder.
Initially, I made the same internal energy costs for any team.
Having added another complex gene and having started this case, I found a significant performance decline. The analysis revealed a population of freaks who used very long genes for their execution, and it must be said very effectively, so that everyone who mastered these genes was a rod with terrible force and crushed the rest, but wanted more variety and competition of species. And here I came up with the idea that the costs of internal energy for the execution of a gene can be made proportional to the time it is performed. And I stupidly entered the time measurement and multiplied the spent nanoseconds by a factor and subtracted from the internal energy of the cell. It seemed to me a good decision. fully automated consideration of the complexity of the execution of the gene and, in general, was well placed on evolutionary ideas, let those who really use complex things survive,
The launch showed even greater sinking of performance, the timing of the measurement was quite expensive on the platform used, the genome interpreter code was written quite productively and began to outweigh the measurement against the background of the usual interpretation. However, the malicious exploiters are very expensive but effective operations have ceased to firmly dominate. I checked, the gene is and sometimes works, but now not so mindlessly often. Diversity has returned, again it was possible to observe the war of multi-colored mold.
And then I thought, and let me take a selective measurement, with a probability of 10%, I will measure the cost of time, and the time-to-energy conversion factor will increase 10 times. On average, the impact on the system will be the same (I thought naive), and the time spent on metering will fall 10 times.
I start and again all the same geezer who very actively use a heavy but profitable team are expanding with a terry color. What? Why? Once again I turn on the measurement at each iteration and they are not so twig. I turn on selective metering and they are there again. Uniform costs and 10 times increased costs applied with a 10% probability give a radically different result.. On dull mold. And the thing seemed like this. The usual costs of this gene were large, but affordable for the cell. However, the costs increased 10 times turned out to be overwhelming and simply completely killed the cell on the spot. And now let's estimate. Malignant exploiters of this gene, for example, execute it once every 5 cycles of simulation (this is very common). And with a 10% chance they will be killed and killed at this. This essentially turns into a condition - do what you want, but every beat of 2% of your population will fall into the use of a very expensive operation and they will be killed. And in such conditions, these genes again showed an advantage and maliciously crushed all the others. All that should make such a population - to multiply more than 2% in time with the simulation, which they successfully did. And worst of all, an even larger increase in the fine for these genes does not change anything at all,
Those. it turned out that instead of the price for this gene there is an accidental punishment (murder) of those who use it. On average, integrally, the energy costs are exactly the same. But the result is radically different.
And here for me in the new key, the meaning of the statement about the inevitability of punishment and about was opened, then why the sudden checks do not work.
Part Three Association
There are illegal receptions in business. For example, take the sellers who hang on the market (manipulating the scales will sell you less than they sound). Those who do not cheat have one indicator of profit. Those who begin to deceive begin to extract substantial additional profits, and even can put very low price tags, thereby attracting a buyer. If for each body kit the seller would be punished with a small fine that makes this kit unprofitable, then no one would do that. But let's consider the situation, if we periodically, at a random point in time, conduct test purchases and issue a very large fine that will take the seller out of the market. Suppose the seller in 30% of sales decides to deceive the buyer. And we also assume that we reach every seller with a control check on a random day once a month, which means about a 3% chance that this seller will be checked that day. 3% * 30% chance that he decides to deceive our controlling buyer gives 1% of what he will be discovered that day. Suppose all sellers are. This means that every day we will catch on the fraud 1% of our sellers and write them a huge fine and they will leave the market. And now interesting. 1% penalty means that 99% of the violation will not be noticed. Now we estimate for 365 working days, what is the chance that they will never catch you? 0.99 ^ 365 ~ 0.025 Ie 2.5% of shopkeepers who every day deceive 30% of buyers during the year will not be caught by such checks. And they will have the best profit. And they will be much more competitive than those who do not use such a hack. Similarly, it will happen, for example, with people who steal from work.
The worst thing in this situation is that systematic violators of the rules appear and who never come across. And for this, no conscious activity is needed. They will simply be because it is profitable to violate, those who got caught flew out and those who didn't get caught remained. Thus, by assigning such sudden checks, we ourselves create those who still break the rules and do not come across. And people still talk to each other, take an example. In the case of theft, a conditional Petya will appear, who regularly privorovat, also regularly passes spot checks and has never been caught in many years.
The result of the whole note: introducing random checks, we ourselves will create those who regularly break the rules and have never been caught. And they will have a bad influence on the rest, as a result we will have a lot of violators, constant catch of violators, we will dismiss them, but this will not reduce the proportion of violators.
For me, this was the key point of understanding why the scheme with random checks and instant dismissal did not result in good discipline.
And just as well with safety techniques. Violating it faster and easier to do their job, but because they are not always killing Uncle Vanya, “yes, I did it 100 times, not ssy.” That is why an employer interested in observing TB must constantly monitor its observance and the key point must be consistency, not the size of the penalty.
Creating conditions where breaking rules does not work much better than rare but very strong punishments. And now the mechanism has become completely clear why this is so.
Separately, gambling fell outside the scope of this article. Those. the excitement on the subject was not caught. And it also has a crucial role in the philosophy of breaking the rules or theft.