https://padlet.com/moocdeliveryteam/ezuj6netilfnh4ll en-us 2020-08-04 08:39:32 UTC 2020-12-04 19:59:12 UTC hello@padlet.com https://padlet.com/moocdeliveryteam/ezuj6netilfnh4ll/wish/897014854 - This represents institutional discrimination due to statistical bias in data.  Biased training data results in biased algorithms.
- Postcode bias is algorithmic bias, it will over-represent poorer areas.
- Previous bias in arrest history and harsher sentencing of BME groups will be exacerbated.
- The wrong feature selections are used when considering parole and sentencing, questions such as 'have your parents been arrested?' will likely bias the BME community by labelling them as high risk.
- Statisticians may claim that those represented in the historical data will show patterns of those likely to offend, but disadvantaged areas have historically been over-policed and harshly charged this will re-enforce the trend of over-policing and create a loop which they will say 'justifies' the algorithm.]]> 2020-11-06 08:50:48 UTC https://padlet.com/moocdeliveryteam/ezuj6netilfnh4ll/wish/897014854 https://padlet.com/moocdeliveryteam/ezuj6netilfnh4ll/wish/951949046 -  The Swiss Precobs automated predictive policing sytem used by the Swiss police and justice authorities between 2013 and 2019 could be described as encoding bias in the model itself. This is because the system was not transparent enougth to reliably determine how the tool had actually led to the reduction of the burglaries it claimed to have contributed to in the three cantons where it was used, namely, in Zurich, Aargau and Basel Land, given that similar reductions in the numbers of burglaries were reported during the same period in other cantons in Switzerland where this predictive policing tool was not used.]]> 2020-11-23 11:55:53 UTC https://padlet.com/moocdeliveryteam/ezuj6netilfnh4ll/wish/951949046 https://padlet.com/moocdeliveryteam/ezuj6netilfnh4ll/wish/953837504 2020-11-23 19:51:20 UTC https://padlet.com/moocdeliveryteam/ezuj6netilfnh4ll/wish/953837504 https://padlet.com/moocdeliveryteam/ezuj6netilfnh4ll/wish/958095354 I read about the predictive policing software used by the German police called "precobs". The article said that this software doesn't use any personal data like car license plates, for example, or on radio cell queries of cell phones. It uses a technique called "near repeat", which only uses anonymous crime data. Thus "Precobs" could not bring individuals into the focus of the investigators. However, police officers can check who they have recently checked in the endangered area for which the system has sounded the alarm. This model also appears to be potentially at risk in terms of statistical bias - for example, if there is a tendency towards racial profiling in the local police force.]]> 2020-11-24 22:55:42 UTC https://padlet.com/moocdeliveryteam/ezuj6netilfnh4ll/wish/958095354 https://padlet.com/moocdeliveryteam/ezuj6netilfnh4ll/wish/959659137 I read an article about the new trend called Giggle, an app created in 2020 in Australia. Basically, this app allows only girls to chat and meet and interact between each other. However, this app due to algorithms, is bias towards girls and therefore, excluding trans girls who by the way needed to verify its gender contacting the makers, the AI was incapable to recognise them as girls.

Which type of bias would this example be?

From my point of view, could be Proxy bias, since unintentionally this app discriminates when using variables that are proxies for gender.  ]]> 2020-11-25 12:38:49 UTC https://padlet.com/moocdeliveryteam/ezuj6netilfnh4ll/wish/959659137 https://padlet.com/moocdeliveryteam/ezuj6netilfnh4ll/wish/966170332 2020-11-27 18:35:16 UTC https://padlet.com/moocdeliveryteam/ezuj6netilfnh4ll/wish/966170332