A great deal of pre-crime prediction is inherently racist. The Reality Theory is a case in point, which while pretending to be free of liberal presumptions remains full of conservative presumptions. Adrian Raines, the criminologist and psychiatrist, took Reality Theory and made it into a psychiatric science of psychopathology but amazingly his psychopaths all seem to belong to the lower (his words)levels of society not the higher levels (his group), and does not include military or political leaders.

Raines, and many other psychiatrists, hold all prisoners are psychopaths (even the innocent ones)under the notion of psychopathology, but as in the USA a majority of prisoners are black, this amounts to racism. Neither Reality theory or Raines have much truck with society and society's effects nor necessarily see their own good fortune.

Thank you so much for your response.

David, a great deal of psychology and psychiatry is predictive. The modern craze for psychopathology employs original sin, that is people are born bad. I have heard in some parts of the USA groups go around testing people for what I consider an imaginary problem and thereby predicting that one person or another will cause problems to others.

Nevertheless, a number of sociology ideas are equally subject to bias-in the past juvenile delinquency was considered widespread especially amongst the lower classes and controls were effected as was policing. Social and economic factors were largely to blame but the lower classes were viewed as having a predisposition for violent behaviour. At the same time riotous university students, then from the upper classes, were considered merely high spirited even if they were doing the same. The prominence of violence attributed to black youths changed policing too-in psychiatry it created new disorders concerning dancing and singing in the street and other west indian and african habits. It is interesting though that the police are often viewed as naturally racist but doctors are not.

What is the definition of "Pre-Crime Prediction"?

Who is going to be predicted?

Based on what kind of knowns, what kind of prediction is going to be made, using which method, if there is anything like a method?

Exactly what would be done with the prediction?

The law, as I understand, does not punish what is yet to happen (conspiring to do is something that has happened). Preventing a for-sure-to-happen from happening has nothing to do with punishment. If there is something to be punished, the prevention in the strict sense must have already failed. If a crime has been stopped, it is not that a crime has been prevented, but a consequence has been prevented. Whatever pre-crime prediction does, it is not (at least not just) about racism. Depending on how it works, it could cause crimes. Making a prediction indicates taking some measures based on the prediction. A justification must be found for any measure following a prediction, especially when the spirit of the law does not directly support such a measure.

If a prediction is based on factors that correlate with criminal consequences in the past, how reliable the prediction is is one thing: a prediction may well be biased because what has happened does not explain why that happened, and what is considered as has happened might actually have never happened; what is going to be done based on the prediction is another: inappropriate responses might make things worse. In particular, it seems reasonable to say that we should not make a yet-to-happen crime more likely to happen based on a prediction. On the other hand, we certainly shall not just kill everyone that has been predicted as "likely to commit a crime."

So the "value" of a prediction depends on its design and its application.

It is difficult to see how, even if unintended, pre-crime prediction will not be seen by many as racial.

I also find it difficult to see how, even if unintended, pre-crime prediction will not be seen by many as racial.

Ashley Lindsley Leisa S. McElreath I tend to think that a prediction can be racial if it takes inputs that represent an already radical reality and is followed by responses that reinforce the racial reality. But this "racism" is not the whole story of criminal justice that would be affected by a prediction.

One way to interpret this kind of "racism" is to say that we not only want to eliminate responses that reinforce the racial reality but also want to have a reality that does not input racial inequality. And a prediction should be designed and applied to improve what we have, if a prediction is to be made at all.

Xinyuan, predictions usually involve an ideological base. My son, who is mixed, has been stopped by police as his looks predict (for some) certain personality traits. People who grow up in lower income environments are safely predicted to turn to certain crimes, but those from much better environments might in fact be predicted to engage in other crimes. (See Adrian Raine above) Say, all policeman have a personality that predicts certain aspects of people from certain backgrounds caught up in the fact that the police engage in 'public' crimes. Public crimes are usually considered worse than private (white collar crimes), thereby certan 'safe' predictions can be made. Private crimes are not always prosecuted while public crimes usually are.

The term black crime widely used seems justified as black youths often do similar public crimes. My son, for example, says the term is both racist (colour based) and meaningless as the crimes are done by cultural groups who happen to be black.

Stanley Wilkin Thank you for your comment. I guess one of the central problems of prediction is to respond to which correlation. If a prediction is based on a summary of data, there could well be different skin colors associate with different averages of crime rate, as well as a strong negative correlation between income and crime rate. It is intuitive for me that the former correlation does not indicate a causal relation, while the latter should be about a causal relation. If there is no causal relation in the former, then there is no point in targeting certain skin color, regardless of the publicness or privateness nature of a likelihood predicted.

I am not a professional in criminology and would not become one, but I observe that some practices, such as, based on a prediction to increase the patrol rate for a certain area, in an emergency narrowing down the likelihood of an unknown subject based on a prediction, may be justified. I don't see, however, how might a non-random higher stop rate targeting a skin color be justified.

I have not read the works, but I couldn't agree on the claim that all criminal prisoners (let along innocents) are psychopaths. In fact, I think if that is literally the claim made by a criminology and psychiatry circle, I would even suspect that there can be a causal relation between crime rate and psychopathic professionals.

I hope this has entertained Mark Twain a little.

Xinyuan gu I am afraid that pre-crime prediction and resulting preventive measures will be based not on causal relations, but on the kinds of correlation you sketch. Pre-Crime prediction algorithms, like all big data algorithms, work on correlations, not causal relations. The resulting preventive measures will have value (since, to increase crime prevention, correlation will already suffice), but are also likely to have comparatively greater negative consequences (due to racist targeting of ethnic groups more likely to participate in crime than others). Hence, I do not agree with your saying „If there is no causal relation in the former, then there is no point in targeting certain skin color“ — unfortunately, there is a point, namely the prevention of crime, it‘s just a point which, due to the lack of causality you mention, can only be achieved by immoral means: by targeting ethnic groups as if they were causally responsible as a group for certain crimes. It is up to socially responsible administration to step up and emphasize that, due to a lack of causal relation, and due to preventing racism, such correlations must not be acted on —but that’s not part of the data analysis itself (and has nothing to do with the preventive value of such algorithms), it’s responsible social politics.

It is also worth noting that there are of course also crimes correlated with high incomes, which are thereby more prevalent in different ethnic groups than crimes in low-income contexts, and that such crimes are sadly rarely meant when we speak of crime-prevention, even though the social losses are immense. For example, over 55 Billion € were recently stolen from Germany due to a peculiar form of tax fraud perpetrated by stock brokers and banks.

Joachim Lipski I think the issue is very much a data analysis problem as well as a social politics problem. The data itself may be biased; the correlation calculation may be purely mathematical and non-biased; the way we interpret the correlations may well be biased.

First, we should question the preciseness of the data-based prediction and what the prediction actually means. Understanding what kind of correlation is found between which and which variables, understanding what exactly has been predicted, and understanding the preciseness of the prediction, these are essential in finding our conclusions on what should be acted on what/whom. I am not sure how to understand "to increase crime prevention, correlation will already suffice". If a crime has never happened and is said to have been successfully prevented from happening (that is, will never happen), how could we tell if any prevention has been successful? Is it not the case that when you were thinking of "increase crime prevention", you were actually thinking of "decrease the number of crimes that actually happen"? These are very different expressions: the former is the "Emperor's New Clothes", while the latter can be found by making comparisons across years. What does a prediction mean? This must be a fundamental question we have to ask, and it is to be asked first within data analysis.

Secondly, you have mentioned "due to racist targeting of ethnic groups more likely to participate in crime than others." If the only correlation we see is the one between ethnic groups and crime outcomes, we are certainly ill in our reading of data. Let's say there can be sex (or gender), age, ethnic group, education, income, employment state, savings, stress triggered or not, and many other possible variables from which we could find correlations. Why didn't we say "males are more likely to commit crimes" (which is a cross-culture fact) and target males? Why didn't we say "people aged 20-50 are more likely to commit crimes"? Why didn't we admit "less educated are more likely to commit crimes"? If less-educateds are more likely to commit crimes, why didn't we send police to high schools, junior high schools, to watch and prevent crimes from happening? When we interpret correlations, there are many assumptions that we don't even notice. If it is more likely that income has a causal relation with crime rate, why didn't we first tackle the income-crime rate relation? Why did we first pick out the skin color-crime rate correlation? Of course, the police couldn't pick out the income-crime rate relation because the police could do nothing about that, but data does not have to be read in the way that picks out skin color-crime rate while there are many other interpretations possible and possibly more meaningful. So the question might become, when we have the data, and we have the algorithms that can bring us correlations, why use the data and the algorithms to "predict" crimes in the first place? What did we expect?

Thirdly, talking about crimes that have very bad consequences, things can still be about how you read the data, and how you read the data may very much depend on your social-political view. If you screen a crime database with "victim number larger than n", I bet you will find that the larger the n is, the less obvious correlations you can find from such as income, education, etc (I am just betting on it). I make this bet only based on my observation on human beings: criminals who commit worse or horrible crimes are less "predictable" than those who commit common ones not only because the sample would be quite small but also because income, education, and other aspects could be less definitive in shaping or affecting the behavior of those most cold-blooded criminals. But if that is the case, do we turn to their gene data to find correlations? If one day you have a national or even a world-wide gene database, should you try to predict most cold-blooded criminals with such a database?

So I think there are in fact a lot to be reviewed when we talk about "pre-crime prediction", and how we read data does play an important role in determining if we are making the "prediction" valuable.

Xinyuan gu

I generally agree.

Regarding your questions:

“If a crime has never happened and is said to have been successfully prevented from happening (that is, will never happen), how could we tell if any prevention has been successful? Is it not the case that when you were thinking of "increase crime prevention", you were actually thinking of "decrease the number of crimes that actually happen"?“ - Indeed; under this consideration, the measure for successful crime prevention would be a lowering in actual crimes in comparison to expected crimes. I take it that any data sufficient to extrapolate pre-crime prevention measures from will also yield such expectations. Of course you cannot say whether you have actually prevented an event that never happened if your measure is just the occurrence of the actual event. (We are currently experiencing this in the kind of prevention paradox regarding COVID-19, which is just a fallacy: the relevant measure is constituted by justified expectations, not by what did or didn’t happen).

“Let's say there can be sex (or gender), age, ethnic group, education, income, employment state, savings, stress triggered or not, and many other possible variables from which we could find correlations. Why didn't we say "males are more likely to commit crimes" (which is a cross-culture fact) and target males? Why didn't we say "people aged 20-50 are more likely to commit crimes"? Why didn't we admit "less educated are more likely to commit crimes"? (...) why didn't we first tackle the income-crime rate relation? Why did we first pick out the skin color-crime rate correlation?“ - Thank you, this is exactly my point: Data analysis can yield all kinds of correlations, and acting on them might well decrease crime rates, and thereby have the kind of value the OP speaks of; but acting on them may still be wrong. As I said, I simply disagreed with your statement „If there is no causal relation in the former, then there is no point in targeting certain skin color“— there is a point, it‘s just a bad one.

Joachim Lipski "a lowering in actual crimes in comparison to expected crimes" is certainly one of the possible measurements. It would only make sense when the predicted / expected crimes are justified. I remember when I was a teenager, at a theme park there was a machine "predicted' my height in the future as an adult. Eventually, I become near 4 inch (10 cm) shorter than the prediction. Was it the prediction imprecise or I unconsciously did something to myself to fulfill the prophecy? A crime prediction algorithm should be much more reliable than the one used on the theme park machine. We would, however, want to make sure that we have put our best algorithm into use, and that our best algorithm is not non-sense or reinforcing inequality or easily used in generating self-fulfilling prophecy. This, I believe, is the "justified expectation" problem you mentioned.

Acting on correlations is already post-algorithm. I wonder if you believe that a "justified expectation" in a "pre-crime prediction" algorithm can ever be justified without taking causal relations among correlations into account. Predicting crimes, after all, is not or should not be just like predicting the color of the next ball we pull out from a box, is it?

Xinyuan, a good response. But looking at matters from the point of view of racialism or prejudice the psychiatric idea (I am a disbeliever in this so called science) is prejudice based usually on class. Psychiatrists form an enlite caste (the job assumed through generations). The evidence for psychopathology produced by Raines is based on the efficacy of brain scans, which again many modern neurologists deny, and his identification of brain aberration, which may, if it exists, be owned by millions.

One must tread carefully with quantitative analysis, nevertheless your sense of predictive seems object based, which little of it is. If you listen to the news you might hear black crime used or black on black crime. Conflation of residents in an area might be one problem with this. Nevertheless, we do not hear of white on white crime nor are black crimes ever broken into cultural parts, if true at all. If crime is high in an area predominantly black, it can be referenced within a colour identity but not if an area is predominantly white or any other ethnic groups.

Your belief in correlations must be widened outside of a crime environment. It must also be widened out of public crime fixation. Whether it is always possible to create thoroughly objective data on the behaviour of human populations, subject to influence and change, is doubtful.

Xinyuan gu Again, I generally agree. Some predictions yield interactive effects, so that the mere knowledge of such predictions changes the predicted outcome. (Often, this is quite desirable, of course — climate change would be one such area.) It seems to me that predictions drive so many of our social, political and corporate actions that I wouldn’t view acting on predictions itself as something controversial. But, as you say, not only must we make sure that predictions are based on our best available knowledge and methods, we must also estimate the risk that they yield false results or unintended negative side-effects.

Big data algorithms, as impressive as they can be, are „dumb“ when it comes to social consequences. Many agents only seek a reduction in certain qualitative measurements (in this case: incidents of crime; another important area is the number of insurance cases), so they may push for the use of big data and not care at all about the consequences the community or society must carry (insurance companies may, for example, restrict certain groups from being insured in order to drive down absolute numbers of insurance cases; they are not immediately affected by the larger social damage of excluding poor, high-risk or handicapped individuals; hence there must be political restrictions in place to keep companies from employing such algorithms or fetching the input data in the first place — I believe pre-crime detection is analogous, insofar as some agents only care about reducing absolute numbers, but not about the injustices which come with it).

In this sense, I do indeed believe, as you say, „that a "justified expectation" in a "pre-crime prediction" algorithm can ever be justified without taking causal relations among correlations into account“ — it’s just that the justification is relative to the respective agent‘s interest. For a company, any measure which positively affects their KPI is already justified, and that measure may require nothing more than correlations. If you are interested in allocating social resources justly, however, correlational knowledge will not be enough and require causal analyses (well, except if you simply allocate available resources per capita).

Stanley Wilkin

Joachim Lipski

Thank you for your comments. Regarding the pre-crime detection analogy, let us still check the definition of "prediction." As I understand, in the loosest sense, there can be many kinds of predictions made based on different kinds of data. For instance:

a prediction can be made upon one certain individual's behavior / a certain city's crime rate in the future based only on this particular individual's behavior / the data from this city in the past;

a prediction can also be made upon one certain individual's behavior / a certain city's crime rate in the future based on a data collection from which some pattern can be found and an assumption that the predicted individual/city follows the pattern of the data.

Predicting climate change, my guess, should be something like:

step a) temperature rise and sea-level rise observed first;

step b) find certain human activities responsible;

step c) if the current scale of certain human activities extends to the future, the climate and the entire ecosystem would be...

The step c) prediction is made after the (almost impossible to deny) causal explanation at step b). Then we alter the scale of certain human activities and predict how much the climate could change correspondingly, and we propose a solution.

So the climate change prediction might actually be different from the "interactive prediction" made upon crimes. The climate change prediction is more like predicting the next position of an object in this way: a) knowing the movement of the object, b) knowing the mechanism that determines the movement, c) when the property of the object and the mechanism remain the same, tell where the object would be in the next minute. d) if we want the object to change its movement in a certain way, we should add a certain force.

So again, I question how the so-called "pre-crime prediction" is made. If I am not mistaken, the "predictions" made in such as machine learning and the ones driven by big data are different from the "predictions" following traditional scientific reasonings. (Please correct me if I am very wrong about this) The spirit of machine learning / big data prediction is to see patterns from the past and to tell what would be the accompany when seeing similar patterns in the future and to improve predictions by comparing what is predicted and what actually happens. There is no causal relation involved in such predictions.

So if a scientist claims that certain neurological patterns have a causal relation to committing crimes, and so people carrying such neurological patterns are said to be more likely to commit crimes, this is different from a big-observation-upon-100,000-prisoners-based prediction predicting the chance for whoever carries a crime-correlated collection of patterns actually commits a crime.

If a prediction is made based on the latter method, and the law enforcement "interacts" with a predicted in a way that potentially alters the causal aspects in the environment for the predicted (for instance, before the intervention the predicted has never been exposed to drugs; due to the intervention the predicted is exposed to drugs), then it is impossible to justify the intervention. This problem is not necessarily racial. It might become racial only when the patterns found in the data source count racial. But the causal effects brought by an intervention would be critical.

These discussions have already gone beyond what I can be sure that I know. Thank you very much for reading.

Zinguan gu and Joachim Lipski,

There is an instability in human populations and the allocation of categories for those populations whether ethnic, national, parental ethnicity or simply cultural tendencies. You mention drugs, and some people focus on marihauna as a personality disruptive drug, but I suggest also psychotropic drugs. The latter would not necessarily be included but which I consider are having a p

rofound affect on crime-for example-and the expansion of mental illness. Read here A Dark Science: Women, Sexuality and Psychiatry by Jeffrey Masson. For neurology predictions read my text below. My The New Fascism covers the Neurology argument and its arguments based on original sin and the freedom of upper and privileged classes from psychopathic motives.

What is considered benevolent or malevolent processes in the prediction depends on the agency. Doctors would not include their own drugs in any assessment of crime but I would, and excluding it gives an unbalanced view and leads to the errors discussed here. My working with clients indicates that violence is a symptom of taking psychotropic drugs over time. Doctor's deny this believing, against all the evidence, that their drugs are efficacious and benevolent. I perceive the neurological arguments as deeply suspect as they replicate 19th century psychiatric thinking on the lower classes (sic) and consider only public, that is physical or violent, crimes and are class bound. The point is, predictions depend on what information is pertinant or not and thereby any study is immediately prejudiced.

Preprint CRIMINOLOGY, PSYCHOPATHOLOGY, DEHUMANISATION AND MORLOCKS "T...

The above may be the same problem in any set of predictions.

Dear Stanley Wilkin

In case I haven't made myself clear enough:

I believe that trying to find an explanation for causal relation is the foundation of science, but we should not try to find an explanation for the cause of crimes by experimenting on human beings. If we do not do experiments on human beings, we can never precisely conclude how much nature plays in the shaping of crimes, and therefore it is wrong to discriminate those who have never committed crimes or even those who have committed crimes based on their neurological patterns.

I didn't say that we should only use 'pertinent' information in data-based prediction. I was saying that we should be very careful when applying observation-based predictions in social justice. One way to be careful is to make sure that the law enforcement does not, based on any wrong understanding of the predictions, intendedly or unintendedly participate in causing crimes.

If one day we have a crystal ball that tells a 'deterministic' prediction, we should still be careful not to be causing or cultivating any disaster because of the extra information.

Revenging on a person's name is premature. Please respect yourself.

Xinguan,

My apologies but I am not quite sure how I have upset you.

I suspect we are talking at cross purposes, that is missing each others points.

Xinyuan gu It's surely a typo. The X and the Z are adjacent on the keyboard.

Oh, thanks Karl. I didn't realise my mistake.