@ -67,11 +67,11 @@ Vast amounts of data are incomprehensible for a human being, but once we create
## Types of Data
As we have already mentioned, data is everywhere. We just need to capture it in the right way! It is useful to distinguish between **structured** and **unstructured** data. The former is typically represented in some well-structured form, often as a table or number of tables, while the latter is just a collection of files. Sometimes we can also talk about **semistructured** data, that have some sort of a structure that may vary greatly.
As we have already mentioned, data is everywhere. We just need to capture it in the right way! It is useful to distinguish between **structured** and **unstructured** data. The former is typically represented in some well-structured form, often as a table or number of tables, while the latter is just a collection of files. Sometimes we can also talk about **semi-structured** data, that have some sort of a structure that may vary greatly.
| List of people with their phone numbers | Wikipedia pages with links | Text of Encyclopaedia Britannica |
| List of people with their phone numbers | Wikipedia pages with links | Text of Encyclopedia Britannica |
| Temperature in all rooms of a building at every minute for the last 20 years | Collection of scientific papers in JSON format with authors, data of publication, and abstract | File share with corporate documents |
| Data for age and gender of all people entering the building | Internet pages | Raw video feed from surveillance camera |
@ -12,7 +12,7 @@ Market trends tell us that by 2022, 1-in-3 large organizations will buy and sell
Trends also indicate that we will create and consume over [180 zettabytes](https://www.statista.com/statistics/871513/worldwide-data-created/) of data by 2025. As **Data Scientists**, this gives us unprecedented levels of access to personal data. This means we can build behavioral profiles of users and influence decision-making in ways that create an [illusion of free choice](https://www.datasciencecentral.com/profiles/blogs/the-illusion-of-choice) while potentially nudging users towards outcomes we prefer. It also raises broader questions on data privacy and user protections.
Data ethics are now _necessary guardrails_ for data science and engineering, helping us minimize potential harms and unintended consequences from our data-driven actions. The [Gartner Hype Cycle for AI](https://www.gartner.com/smarterwithgartner/2-megatrends-dominate-the-gartner-hype-cycle-for-artificial-intelligence-2020/) identifies relevant trends in digital ethics, responsible AI ,and AI governances as key drivers for larger megatrends around _democratization_ and _industrialization_ of AI.
Data ethics are now _necessary guardrails_ for data science and engineering, helping us minimize potential harms and unintended consequences from our data-driven actions. The [Gartner Hype Cycle for AI](https://www.gartner.com/smarterwithgartner/2-megatrends-dominate-the-gartner-hype-cycle-for-artificial-intelligence-2020/) identifies relevant trends in digital ethics, responsible AI ,and AI governance as key drivers for larger megatrends around _democratization_ and _industrialization_ of AI.
@ -179,7 +179,7 @@ Here are a few examples:
| Ethics Challenge | Case Study |
|--- |--- |
| **Informed Consent** | 1972 - [Tuskegee Syphillis Study](https://en.wikipedia.org/wiki/Tuskegee_Syphilis_Study) - African American men who participated in the study were promised free medical care _but deceived_ by researchers who failed to inform subjects of their diagnosis or about availability of treatment. Many subjects died & partners or children were affected; the study lasted 40 years. |
| **Informed Consent** | 1972 - [Tuskegee Syphilis Study](https://en.wikipedia.org/wiki/Tuskegee_Syphilis_Study) - African American men who participated in the study were promised free medical care _but deceived_ by researchers who failed to inform subjects of their diagnosis or about availability of treatment. Many subjects died & partners or children were affected; the study lasted 40 years. |
| **Data Privacy** | 2007 - The [Netflix data prize](https://www.wired.com/2007/12/why-anonymous-data-sometimes-isnt/) provided researchers with _10M anonymized movie rankings from 50K customers_ to help improve recommendation algorithms. However, researchers were able to correlate anonymized data with personally-identifiable data in _external datasets_ (e.g., IMDb comments) - effectively "de-anonymizing" some Netflix subscribers.|
| **Collection Bias** | 2013 - The City of Boston [developed Street Bump](https://www.boston.gov/transportation/street-bump), an app that let citizens report potholes, giving the city better roadway data to find and fix issues. However, [people in lower income groups had less access to cars and phones](https://hbr.org/2013/04/the-hidden-biases-in-big-data), making their roadway issues invisible in this app. Developers worked with academics to _equitable access and digital divides_ issues for fairness. |
| **Algorithmic Fairness** | 2018 - The MIT [Gender Shades Study](http://gendershades.org/overview.html) evaluated the accuracy of gender classification AI products, exposing gaps in accuracy for women and persons of color. A [2019 Apple Card](https://www.wired.com/story/the-apple-card-didnt-see-genderand-thats-the-problem/) seemed to offer less credit to women than men. Both illustrated issues in algorithmic bias leading to socio-economic harms.|
@ -250,7 +250,7 @@ Use the sample code in the notebook to test other hypothesis that:
Probability and statistics is such a broad topic that it deserves its own course. If you are interested to go deeper into theory, you may want to continue reading some of the following books:
1. [Carlos Fernanderz-Granda](https://cims.nyu.edu/~cfgranda/) from New York University has great lecture notes [Probability and Statistics for Data Science](https://cims.nyu.edu/~cfgranda/pages/stuff/probability_stats_for_DS.pdf) (available online)
1. [Carlos Fernandez-Granda](https://cims.nyu.edu/~cfgranda/) from New York University has great lecture notes [Probability and Statistics for Data Science](https://cims.nyu.edu/~cfgranda/pages/stuff/probability_stats_for_DS.pdf) (available online)
1. [Peter and Andrew Bruce. Practical Statistics for Data Scientists.](https://www.oreilly.com/library/view/practical-statistics-for/9781491952955/) [[sample code in R](https://github.com/andrewgbruce/statistics-for-data-scientists)].
1. [James D. Miller. Statistics for Data Science](https://www.packtpub.com/product/statistics-for-data-science/9781788290678) [[sample code in R](https://github.com/PacktPublishing/Statistics-for-Data-Science)]
@ -157,7 +157,7 @@ Some key factors are to consider when creating a compute resource and those choi
A CPU (Central Processing Unit) is the electronic circuitry that executes instructions comprising a computer program. A GPU (Graphics Processing Unit) is a specialized electronic circuit that can execute graphics-related code at a very high rate.
The main difference between CPU and GPU architecture is that a CPU is designed to handle a wide-range of tasks quickly (as measured by CPU clock speed), but are limited in the concurrency of tasks that can be running. GPUs are designed for parallel computing and therfore are much better at deep learning tasks.
The main difference between CPU and GPU architecture is that a CPU is designed to handle a wide-range of tasks quickly (as measured by CPU clock speed), but are limited in the concurrency of tasks that can be running. GPUs are designed for parallel computing and therefore are much better at deep learning tasks.
@ -51,7 +51,7 @@ Let's look at one example - the [MIT Gender Shades Study](http://gendershades.or
* **What:** The objective of the research project was to _evaluate bias present in automated facial analysis algorithms and datasets_ based on gender and skin type.
* **Why:** Facial analysis is used in areas like law enforcement, airport security, hiring systems and more - contexts where inaccurate classifications (e.g., due to bias) can cause potential economic and social harms to affected individuals or groups. Understanding (and eliminating or mitigating) biases is key to fairness in usage.
* **How:** Researchers recongized that existing benchmarks used predominantly lighter-skinned subjects, and curated a new data set (1000+ images) that was _more balanced_ by gender and skin type. The data set was used to evaluate the accuracy of three gender classification products (from Microsoft, IBM & Face++).
* **How:** Researchers recognized that existing benchmarks used predominantly lighter-skinned subjects, and curated a new data set (1000+ images) that was _more balanced_ by gender and skin type. The data set was used to evaluate the accuracy of three gender classification products (from Microsoft, IBM & Face++).
Results showed that though overall classification accuracy was good, there was a noticeable difference in error rates between various subgroups - with **misgendering** being higher for females or persons with darker skin types, indicative of bias.
Jasmine Greenaway
4 years ago
committed by
GitHub