British cops told to scrap 'discriminatory' algorithms in policing

Statistics

Has its influence on bias in algorithms, otherwise they would be pretty useless.

Predictive policing has the ability to be great if....

Sadly it is true that certain sectors of society are more likely to be sucked into crime, and especially gang crime. Typically young men from lower income households with lower educational attainment, and lower levels of aspiration.

The issue here is that instead of looking at 'young men who are poor, not doing well at school, and don't see a good future' (which is really difficult to profile) the algorithms take the easy way out and do a match with 'young black or muslim men'.

There is nothing inherently less aspirational, educational, or attaining about any particular genetic inheritance - but there may be about certain cultural groups, and how those groups have integrated (or failed to integrate) into UK society. Algorithms have a distressing tendency to jump on the back of easily identitfiable statistical factors, which may 'work', but which will exacerbate the underlying issues by objectifying and victimising groups which have already experienced significant negative bias.

Sadly unless society (or its proxy, Government) makes a concerted (for which you can read 'expensive') effort to actually address the underlying issues the problems will continue.

A rather interesting study into educational attainment in I think South America showed 10% better education, incomes and attainment now in communities clustered around the sites of educational religious missions 200 years ago. No such positive affect existed around doctrinal religious missions. These affects are long term, and persist long past the original causes, unless positive action is taken to correct problems.

Predictive policing is coming, and it will reduce crime, but it is the challenge for society to ensure it doesn't do so by making it even harder for those who are suffering under current systems to improve their position in society.

Re: Predictive mapping and individual risk assessments

If these two techniques actually work then by all means continue to use them but if they don't work then consider other options like building up intelligence networks in communities.

All that said, what would really help the police become more effective is if the Tories unsacked the 21,000 police officers and PCSOs that they slashed as part of their smaller state austerity agenda.

Well when ~95% of all knife crime in London is black on black gang violence - that's where the focus ends up going, whether its putting preventative measures in place (and trying to identify where said crime is likely to occur to make it most effective) or purely reactive (going to where emergency calls come from) the result is the same.

Is it still discrimination...

...when it's demonstrably true?

Just saying

PC Schweik blames the computer.

So instead of "everybody knowing" (for example) the South Side is the baddest part of town (as in the USA), PC Schweik blames the computer.

lots of police on patrol hereabouts.

mind how you go.

let's get a few things straight here

"Predictive Policing" is just new jargon for crime analysis. You analyse what has happened in the past, To figure out what types of crimes may occur in certain areas on certain days and times. You then organise the police in that area to be in those areas, days and times that the crimes are most likely to occur. There are no sophisticated algorithms. Very basic statistics. You find out 40% of the Break, Enter and Steal offences occur on a Saturday night between 8pm and 3am, in a certain area, you make sure the police on patrol are in that area during that time frame.

You couple this with "Community Policing". Remember the "ole days when you would see the same police officer(s) occasionally, or regularly, walking around your local area. They would know who lived in the area. If you were good they would greet you with a smile and maybe a quick chat on how you were doing. If you occasionally got into trouble. you would still get the greeting but it was followed with a warning. If you were a shitbag that didn't care about the law or the trouble you caused in the local area. They would stop you,,find out who you were with, where you were, where you are going and what you were doing. This was shared among the other officers in the local area and was built up over time. Apparently known as "Individual Risk Assessments" now days.

Combine these two and it is the most effective tactic for reducing crime in a given area. No amount of predictive modelling will ever trump it.

After 10 years in the police , on the street, as a detective and later as a crime analyst. I can personally attest, that the people who shout the loudest about the thuggery of police. Even the mere presence of them in their area or street are always the ones who cry the loudest. Demand the most immediate and brutal punishments. Blame the Police for the lack of foresight, skill or lack of action when a crime is committed against them. The hypocrisy is completely lost on them.

Here endeth my rant for the day

The 2011 London Riots

The 2011 London Riots would arguably have been a good opportunity to study human behaviour without taking participants' ethnicity into consideration (Wikipedia quotes those arrested as 29% white and 39% black in London). What was exhibited there was innate reaction to various environmental stimulii, regardless of ethnicity. The number of people in a specified area using certain body language inciting others to act similarly, setting off a chain reaction, even to those drawn in from the periphery.

Another research possibility is studying the effects of ambient temperature on crime statistics. So if the temperature is x in place abc, AI might be able to predict a corresponding percentage increase or decrease in crime in that locality.

To my mind these are areas of "macro criminal behaviour" where I feel AI could be usefully harnessed, rather than micro or person to person interaction where prejudice is more likely to be baked into observations.

Bias is self-perpetuating in lazy statistics

Don't be fooled by "what if this group really is more criminal" - apart from the main factor that others have mentioned, which is that socioeconomic factors are more predictive than race, you can easily envision how bias can set up a feedback loop.

Start with two groups, A and B, which actually have identical proportions of members who carry knives (or any other crime detectable by a search). Due to cultural bias, more people from group A are searched than from group B. Also due to bias, when charged, a greater proportion of members of group A are convicted than those from group B. Now take a look at the headline statistic of people convicted of knife offences - wow, people from group A are massively over-represented! We'd better make sure we focus our policing on them! Now not only do you have the support of those with (conscious or otherwise) bias against group A, you also have people with neither bias or the ability to properly check the validity of statistics.

The biggest danger is that it may be effective

I used to live in one of those postcode areas where "If it wasn't nailed down it was immediately nicked and if it was nailed down then it just took a few minutes to pry it up". Most people in that area were ordinary people, more or less, with only a relative minority being the actual criminals. After a bit you even got to know who was the likely puerp, the cops knew it as well but there's a lot of distance between knowing who the villain was and catching them with enough evidence to nail them.

The algorithms just systematize the informal data collection and area knowledge of the beat copper of old. They should be more objective than an individual and so a lot more effective. The algorithm may seem to be biased, especially when its first used, because it has only existing data to work on, data that is probably biased. It will rapidly adapt, though, and you'll quite likely to find that its stopped being interested in dark skinned people in wool hats and started zeroing in on ostensibly respectable people who almost seem to fit in but don't quite.

I should remark that as an ex-inner city dweller I'm not particularly 'pro' police. They're only human so they can spend far too much time going after the low hanging fruit (i.e. bothering me and my friends). I never found them particularly biased against creed or color but they do get a bit cynical about offenders since that's their stock in trade. This is what tends to cause bias -- they see the same faces, the same sorts of people day in, day out, so they tend to make generalizations about people. The computer may go a long way to putting them right.