Weapons of Math Destruction: How Big Data Increases Inequality and Threatens Democracy by Cathy O'NeilMy rating: 3 of 5 stars
An actual, important and insightful book that everyone should read, but should be mandatory for every data practitioners, managers and decision/policy makers.
It provides several examples whereby mathematical/statistics models can go wary and produce outcomes significantly different from the ones original sought and create or reinforce bias, perpetuate trends, reinforce segregation and produce pure bad results and behaviors.
And all this happens while most of the times the models were created with the best intention, of course there are also the cases that some were created to generate and exploit systematical the debilities of specific groups or individuals.
My key takeaways of the book to address most of the identified problems are:
- all models are simplifications of complex realities (full stop);
- when creating a model one should be cognizant of the previous, thus adapt and embrace that it will not be able to address all the variables, so trade-offs have to be made, but those should be made explicit;
- the variables and trade-offs should be made by its relative importance and not because they are quantitively available and make the model work better from an statistical/econometric perspective;
- the models should not be black boxes and transparency should be granted to all the ones that are using it (the user have the right to know);
- the models should be continuously monitored and adapted to the new realities;
- the outcomes and the impact of the model should also be assessed on a systematically way, to ensure there are no unintended consequences of its use.
I'm still a strong believer that these models when properly applied are of great value to underpin the human development, but they can really become WMD if not understood, controlled and monitored by the experts and by the society as all (through regulators, policies in place and by the individuals).
However, there should always be an ethical commitment by the ones who are creating them, this community cannot be neutral to what they are creating.
I feel on my day-to-day work that the model creators do not like to share and are unwilling to share their work with the others (within the society or organization) and that has to change dramatically. The mathematicians, statisticians and data-scientist have to develop their communication skills to be able to explain the models created to any audience they face (not the other way around). If you cannot explain it in 5 minutes and easily is because you do not know it thoroughly!
Regarding the book i think it emphasizes too much on the examples of WMD (after 4/5 the reader gets the point the author is trying to make) and does not focus sufficiently on the potential solutions for the problems presented.
Do recommend its reading.
View all my reviews
It provides several examples whereby mathematical/statistics models can go wary and produce outcomes significantly different from the ones original sought and create or reinforce bias, perpetuate trends, reinforce segregation and produce pure bad results and behaviors.
And all this happens while most of the times the models were created with the best intention, of course there are also the cases that some were created to generate and exploit systematical the debilities of specific groups or individuals.
My key takeaways of the book to address most of the identified problems are:
- all models are simplifications of complex realities (full stop);
- when creating a model one should be cognizant of the previous, thus adapt and embrace that it will not be able to address all the variables, so trade-offs have to be made, but those should be made explicit;
- the variables and trade-offs should be made by its relative importance and not because they are quantitively available and make the model work better from an statistical/econometric perspective;
- the models should not be black boxes and transparency should be granted to all the ones that are using it (the user have the right to know);
- the models should be continuously monitored and adapted to the new realities;
- the outcomes and the impact of the model should also be assessed on a systematically way, to ensure there are no unintended consequences of its use.
I'm still a strong believer that these models when properly applied are of great value to underpin the human development, but they can really become WMD if not understood, controlled and monitored by the experts and by the society as all (through regulators, policies in place and by the individuals).
However, there should always be an ethical commitment by the ones who are creating them, this community cannot be neutral to what they are creating.
I feel on my day-to-day work that the model creators do not like to share and are unwilling to share their work with the others (within the society or organization) and that has to change dramatically. The mathematicians, statisticians and data-scientist have to develop their communication skills to be able to explain the models created to any audience they face (not the other way around). If you cannot explain it in 5 minutes and easily is because you do not know it thoroughly!
Regarding the book i think it emphasizes too much on the examples of WMD (after 4/5 the reader gets the point the author is trying to make) and does not focus sufficiently on the potential solutions for the problems presented.
Do recommend its reading.
View all my reviews
