These are chat archives for FreeCodeCamp/DataScience
discussion on how we can use statistical methods to measure and improve the efficacy of http://freeCodeCamp.com
The most interesting presentation went about Amsterdam "Smart City" project and the social effects of Big Data governance. Amsterdam has plans to become one of the Smart Cities in Europe. However there are many questions about the implications of becoming a Smart City:
One of the main concerns that caused more discussion was that about individual privacy: it is likely that the cost of privacy could increase, and that only those who have enough income will be able to protect their privacy: the rest will be either exposed or not included in the social welfare system: privacy could become a expensive good in a heavily big-data-dependent community.
Not mentioned during the presentation was the fact that big data and data science is overrated when it comes to its ability to really bring the actual picture of the population and individuals. As we have discussed here (particularly @erictleung), it is not totally appropriate to affirm that having the data and implement some good methods that minimise the error is enough: the pattern that you find in the data might not be representative of the social architecture, which could be more complex and absent from the data.
In fact, a similar issue about the risks of basing actions crudely on data only could be found in another study about the implementation of Big Brother surveillance devices in Oakland and the concerns about increasing social discrimination.
In short, data patterns should have a sensible social interpretation, not technocratic one, if you want the social welfare to become a reality. Recognising the exclusion, either by lack of access or due to own choice, should be also relevant.
The presentation by Linnet Taylor was a summary of a preliminary report that you can find at:
Wow I never knew Oakland tried to implement something as ambitious as a smart city, heck I'd be surprised if any place in the US tried it.
There is a story that I will link that is somewhat on topic, it's about prediction software being used to predict the occurrence of crime in areas of a city in real time.
Due to racial tensions in the US, I'd like to bring up a bit of a ethical dilemma.
Should prediction software use race as a feature as part of it's prediction either in the case of crime prediction or attempting to predict Recidivism, I'll explain two of my opinions on both sides of the argument. On the yes side by incorporating race as a feature it would possibly increase the accuracy of the model due to latent factors.
On the no side, it could create what I would call a "Race bias" whereas the software would more likely peg a minority at a much higher risk of reoffending just on the basis of the race feature maybe again due to latent factors.
razerh0 sends brownie points to @erictleung :sparkles: :thumbsup: :sparkles: