https://padlet.com/moocdeliveryteam/cbrlnqvu1w69yj7p en-us 2020-04-21 11:46:26 UTC 2020-09-16 09:53:03 UTC hello@padlet.com moocdeliveryteam https://padlet.com/moocdeliveryteam/cbrlnqvu1w69yj7p/wish/519998274

Double click anywhere on the board to add a new note

Add your name and comment

Add any relevant links, images or attachments

Spend time reviewing your peers' posts and comment on and 'like' your favourites

]]> 2020-04-21 11:48:38 UTC https://padlet.com/moocdeliveryteam/cbrlnqvu1w69yj7p/wish/519998274 https://padlet.com/moocdeliveryteam/cbrlnqvu1w69yj7p/wish/580579896 Algorithmic bias is definitely present in predictive policing. Be it predictive mapping or individual assessments, the fact that proxies appear to play a major role in the data points that police use to build those algorithms is alarming, especially considering the long lasting detrimental impact that wrongfully identified targets will suffer.]]> 2020-05-18 17:32:46 UTC https://padlet.com/moocdeliveryteam/cbrlnqvu1w69yj7p/wish/580579896 https://padlet.com/moocdeliveryteam/cbrlnqvu1w69yj7p/wish/581548689 Key source of bias in this tool could be in the data collection and labelling where historical data is used for training models which quite likely included over-representation of black males residing in subsidized housing in previous incarceration records. Feature selection could be another source of bias in this algorithm: considering its opacity it may be difficult to justify why certain attributes were used as "good" representation of "likelihood to re-offend".]]> 2020-05-19 05:23:40 UTC https://padlet.com/moocdeliveryteam/cbrlnqvu1w69yj7p/wish/581548689 https://padlet.com/moocdeliveryteam/cbrlnqvu1w69yj7p/wish/583630859 Our discussions have been based on how to further streamline policing efficiency, based on the idea that more policing = less crime. But are we sure this is true?
I realize I am straying from the discussion but I think any way we increase policing, whether it is evident or opaque, we will only lead to more civil disobedience and more mistakes made on behalf of police people. 
There is an argument to be made for surveillance and how serious crime is less likely to occur in heavily surveilled zones, but any time/resources the government/police are putting into these systems would be better used to fight the underlying issues that lead to crime such as  poverty.
By only criticizing the techniques the police are using rather than the fundamental problems they are doing nothing to solve, we are letting them off the hook.]]> 2020-05-19 22:29:08 UTC https://padlet.com/moocdeliveryteam/cbrlnqvu1w69yj7p/wish/583630859 https://padlet.com/moocdeliveryteam/cbrlnqvu1w69yj7p/wish/586623727 appears to confirm bias. Perhaps some areas have a higher crime level, and there is a repeat offender, but does that validate the use of AI. Should the police ethically put increased scrutiny on these areas? It claims utility, but it does not necessarily promote fairness. Maybe train AI on the police to determine how fair they will be in using such a program. That way, it's not one way—just a thought.]]> 2020-05-21 09:34:59 UTC https://padlet.com/moocdeliveryteam/cbrlnqvu1w69yj7p/wish/586623727 https://padlet.com/moocdeliveryteam/cbrlnqvu1w69yj7p/wish/587981193 It seems kind fairly to increase overseen on areas with high historig trends of crime. However trends from other kind of crimes usually perpetrated by amercans should be included in the algorithms. Concepts, data and rules should be reviewed and wide analyzed before to run the programs.]]> 2020-05-21 22:27:37 UTC https://padlet.com/moocdeliveryteam/cbrlnqvu1w69yj7p/wish/587981193 https://padlet.com/moocdeliveryteam/cbrlnqvu1w69yj7p/wish/590505945 Certainly the examples given and discussed demonstrate that predictive policing is problematic and has the potential to further enforce social bias and prejudices. However, if we assume that these algorthims are constructed by experts and trained on real data then i think the bias can be attributed to a broader problem with society as a whole. If we could help balance out inequality in society as a whole then the bias that clearly exists in predictive algorithms could be negated ]]> 2020-05-23 16:07:15 UTC https://padlet.com/moocdeliveryteam/cbrlnqvu1w69yj7p/wish/590505945 https://padlet.com/moocdeliveryteam/cbrlnqvu1w69yj7p/wish/591487615 Algorithms are biased due to the type of data. To a certain extent by the data miner aswell. 
More policing could lead to less crime but this will put an undue stress on the community. Also the type of questions used to decide the bail. Has your parents ever been sent to jail. Is unfair we are stereotyping people. There is a threat there we could put more people under a bucket.]]> 2020-05-24 15:47:56 UTC https://padlet.com/moocdeliveryteam/cbrlnqvu1w69yj7p/wish/591487615 https://padlet.com/moocdeliveryteam/cbrlnqvu1w69yj7p/wish/593426143 Algorithms could be biased by the labeling of the data and also by the data collection. In both cases, whoever input that data could do it biased, intentionally or not.]]> 2020-05-25 18:20:32 UTC https://padlet.com/moocdeliveryteam/cbrlnqvu1w69yj7p/wish/593426143 https://padlet.com/moocdeliveryteam/cbrlnqvu1w69yj7p/wish/608971164 The PredPol system appears to be biased due to the datasets used to train the algorithm about where crimes are most likely to occur. It also appears to discriminate by Proxy i.e. 'it so happens' that high crime areas are predominantly Black/Hispanic ethnicity and poor demographic (as the policeman himself says, he regards the data as accurate, because it is accurate..) Understanding (or caring too much) about the bigger picture is not part of his job. ]]> 2020-06-03 14:02:56 UTC https://padlet.com/moocdeliveryteam/cbrlnqvu1w69yj7p/wish/608971164 https://padlet.com/moocdeliveryteam/cbrlnqvu1w69yj7p/wish/614501495
We can see multiple types of Algorithmic biases at play as far as Predictive Policing/Sentencing is concerned

Data/Sampling Bias- due to biased & Dirty historical data

The Feedback loops, which the Machine automatically generates, is another layer of bias, which is propped up due to the initial input of faulty & partial data

Bias by Proxy- which amounts to 'Racism by other means'. Eg- Questions regarding the 'Area of Residence'

Bias in the Design/Model itself- the Articles mention how it is the Police who ultimately decide, what 'type' of crimes to focus on. Eg- Petty roadside crimes vis-a-vis big ticket white collar crimes

Given the backdrop of the 'George Floyd's incident, we cannot rule out the inherent bias within the Legal systems & therefore the risk of bias via 'Masking' being Algorithms, seems more real than not]]> 2020-06-06 19:30:54 UTC https://padlet.com/moocdeliveryteam/cbrlnqvu1w69yj7p/wish/614501495 https://padlet.com/moocdeliveryteam/cbrlnqvu1w69yj7p/wish/617655041 Assuming that all qualitative data entry contains some bias (in design, implementation or analysis), surely the key is to identify that bias to better understand the uniqueness of the algorithm. Seeking to achieve no bias - or unbias - isn't practical. Better to be transparent and design around that.]]> 2020-06-08 20:32:21 UTC https://padlet.com/moocdeliveryteam/cbrlnqvu1w69yj7p/wish/617655041 https://padlet.com/moocdeliveryteam/cbrlnqvu1w69yj7p/wish/750984611 Thus far, we have discovered that it isn't possible to create a bias-free algorithm.  

This is particularly problematic when utilising algorithms in areas such as predictive policing.  The data available is flawed, as it reflects a system heavily weighted against minorities, and further perpetuates the imbalance.

Breonna Taylor and George Floyd are our most recent examples, but this is rampant.  There are other Breonnas and Georges targeted daily.

Time to throw it all out and start again?  If we haven't yet been scared off, here are two actionable approaches:
1- co-designing with complete representation to identify and remove bias would help, but it would be fair to say that we will never get to 100% unbiased, because we as humans are flawed.

2- legislation on AI and what corporates and governments can and can't use it for.]]> 2020-09-16 09:42:24 UTC https://padlet.com/moocdeliveryteam/cbrlnqvu1w69yj7p/wish/750984611