An ethics checklist for data scientists
deon is a command line tool that allows you to easily add an ethics checklist to your data science projects. We support creating a new, standalone checklist file or appending a checklist to an existing analysis in many common formats.
To help get started, deon includes a default Data Science Ethics Checklist along with a list of real-world examples connected with each item. Users can draw on the default list or develop their own.
δÎον • (déon) [n.] (Ancient Greek) wikitionary
Duty; that which is binding, needful, right, proper.
The conversation about ethics in data science, machine learning, and AI is increasingly important. The goal of deon is to push that conversation forward and provide concrete, actionable reminders to the developers that have influence over how data science gets done.
Quickstart
You only need two lines of code to get started!
First, install deon:
$ pip install deon
Then, write out the default checklist to a markdown file called ETHICS.md:
$ deon -o ETHICS.md
Dig into the checklist questions to identify and navigate the ethical considerations in your data science project.
For more configuration details, see the sections on command line options, supported output file types, and custom checklists.
Background and perspective
We have a particular perspective with this package that we will use to make decisions about contributions, issues, PRs, and other maintenance and support activities.
First and foremost, our goal is not to be arbitrators of what ethical concerns merit inclusion. We have a process for changing the default checklist, but we believe that many domain-specific concerns are not included and teams will benefit from developing custom checklists. Not every checklist item will be relevant. We encourage teams to remove items, sections, or mark items as N/A as the concerns of their projects dictate.
Second, we built our initial list from a set of proposed items on multiple checklists that we referenced. This checklist was heavily inspired by an article written by Mike Loukides, Hilary Mason, and DJ Patil and published by O'Reilly: "Of Oaths and Checklists". We owe a great debt to the thinking that proceeded this, and we look forward to thoughtful engagement with the ongoing discussion about checklists for data science ethics.
Third, we believe in the power of examples to bring the principles of data ethics to bear on human experience. This repository includes a list of real-world examples connected with each item in the default checklist. We encourage you to contribute relevant use cases that you believe can benefit the community by their example. In addition, if you have a topic, idea, or comment that doesn't seem right for the documentation, please add it to the wiki page for this project!
Fourth, it's not up to data scientists alone to decide what the ethical course of action is. This has always been a responsibility of organizations that are part of civil society. This checklist is designed to provoke conversations around issues where data scientists have particular responsibility and perspective. This conversation should be part of a larger organizational commitment to doing what is right.
Fifth, we believe the primary benefit of a checklist is ensuring that we don't overlook important work. Sometimes it is difficult with pressing deadlines and a demand to multitask to make sure we do the hard work to think about the big picture. This package is meant to help ensure that those discussions happen, even in fast-moving environments. Ethics is hard, and we expect some of the conversations that arise from this checklist may also be hard.
Sixth, we are working at a level of abstraction that cannot concretely recommend a specific action (e.g., "remove variable X from your model"). Nearly all of the items on the checklist are meant to provoke discussion among good-faith actors who take their ethical responsibilities seriously. Because of this, most of the items are framed as prompts to discuss or consider. Teams will want to document these discussions and decisions for posterity.
Seventh, we can't define exhaustively every term that appears in the checklist. Some of these terms are open to interpretation or mean different things in different contexts. We recommend that when relevant, users create their own glossary for reference.
Eighth, we want to avoid any items that strictly fall into the realm of statistical best practices. Instead, we want to highlight the areas where we need to pay particular attention above and beyond best practices.
Ninth, we want all the checklist items to be as simple as possible (but no simpler), and to be actionable.
Using this tool
Prerequisites
- Python >3.6: Your project need not be Python 3, but you need Python 3 to execute this tool.
Installation
$ pip install deon
or
$ conda install deon -c conda-forge
Simple usage
We recommend adding a checklist as the first step in your data science project. After creating your project folder, you could run:
$ deon -o ETHICS.md
This will create a markdown file called ETHICS.md that you can add directly to your project.
For simple one-off analyses, you can append the checklist to a Jupyter notebook or RMarkdown file using the -o flag to indicate the output file. deon will automatically append if that file already exists.
$ jupyter notebook my-analysis.ipynb
...
$ deon -o my-analysis.ipynb # append cells to existing output file
This checklist can be used by individuals or teams to ensure that reviewing the ethical implications of their work is part of every project. The checklist is meant as a jumping-off point, and it should spark deeper and more thourough discussions rather than replace those discussions.
Proudly display your Deon badge
You can add a Deon badge to your project documentation, such as the README, to encourage wider adoption of these ethical practices in the data science community.
HTML badge
<a href="http://deon.drivendata.org/">
<img src="https://img.shields.io/badge/ethics%20checklist-deon-brightgreen.svg?style=popout-square" alt="Deon badge" />
</a>Markdown badge
[](http://deon.drivendata.org/)
Supported file types
Here are the currently supported file types. We will accept pull requests with new file types if there is a strong case for widespread use of that filetype.
.txt: ascii.html: html.ipynb: jupyter.md: markdown.rmd: rmarkdown.rst: rst
Command line options
Usage: deon [OPTIONS]
Easily create an ethics checklist for your data science project.
The checklist will be printed to standard output by default. Use the --output
option to write to a file instead.
Options:
-l, --checklist PATH Override default checklist file with a path to a custom
checklist.yml file.
-f, --format TEXT Output format. Default is "markdown". Can be one of
[ascii, html, jupyter, markdown, rmarkdown, rst].
Ignored and file extension used if --output is passed.
-o, --output PATH Output file path. Extension can be one of [.txt, .html,
.ipynb, .md, .rmd, .rst]. The checklist is appended if
the file exists.
-w, --overwrite Overwrite output file if it exists. Default is False,
which will append to existing file.
-m, --multicell For use with Jupyter format only. Write checklist with
multiple cells, one item per cell. Default is False,
which will write the checklist in a single cell.
--help Show this message and exit.
Default checklist
Data Science Ethics Checklist
A. Data Collection
- A.1 Informed consent: If there are human subjects, have they given informed consent, where subjects affirmatively opt-in and have a clear understanding of the data uses to which they consent?
- A.2 Collection bias: Have we considered sources of bias that could be introduced during data collection and survey design and taken steps to mitigate those?
- A.3 Limit PII exposure: Have we considered ways to minimize exposure of personally identifiable information (PII) for example through anonymization or not collecting information that isn't relevant for analysis?
- A.4 Downstream bias mitigation: Have we considered ways to enable testing downstream results for biased outcomes (e.g., collecting data on protected group status like race or gender)?
B. Data Storage
- B.1 Data security: Do we have a plan to protect and secure data (e.g., encryption at rest and in transit, access controls on internal users and third parties, access logs, and up-to-date software)?
- B.2 Right to be forgotten: Do we have a mechanism through which an individual can request their personal information be removed?
- B.3 Data retention plan: Is there a schedule or plan to delete the data after it is no longer needed?
C. Analysis
- C.1 Missing perspectives: Have we sought to address blindspots in the analysis through engagement with relevant stakeholders (e.g., checking assumptions and discussing implications with affected communities and subject matter experts)?
- C.2 Dataset bias: Have we examined the data for possible sources of bias and taken steps to mitigate or address these biases (e.g., stereotype perpetuation, confirmation bias, imbalanced classes, or omitted confounding variables)?
- C.3 Honest representation: Are our visualizations, summary statistics, and reports designed to honestly represent the underlying data?
- C.4 Privacy in analysis: Have we ensured that data with PII are not used or displayed unless necessary for the analysis?
- C.5 Auditability: Is the process of generating the analysis well documented and reproducible if we discover issues in the future?
D. Modeling
- D.1 Proxy discrimination: Have we ensured that the model does not rely on variables or proxies for variables that are unfairly discriminatory?
- D.2 Fairness across groups: Have we tested model results for fairness with respect to different affected groups (e.g., tested for disparate error rates)?
- D.3 Metric selection: Have we considered the effects of optimizing for our defined metrics and considered additional metrics?
- D.4 Explainability: Can we explain in understandable terms a decision the model made in cases where a justification is needed?
- D.5 Communicate bias: Have we communicated the shortcomings, limitations, and biases of the model to relevant stakeholders in ways that can be generally understood?
E. Deployment
- E.1 Monitoring and evaluation: How are we planning to monitor the model and its impacts after it is deployed (e.g., performance monitoring, regular audit of sample predictions, human review of high-stakes decisions, reviewing downstream impacts of errors or low-confidence decisions, testing for concept drift)?
- E.2 Redress: Have we discussed with our organization a plan for response if users are harmed by the results (e.g., how does the data science team evaluate these cases and update analysis and models to prevent future harm)?
- E.3 Roll back: Is there a way to turn off or roll back the model in production if necessary?
- E.4 Unintended use: Have we taken steps to identify and prevent unintended uses and abuse of the model and do we have a plan to monitor these once the model is deployed?
Data Science Ethics Checklist generated with deon.
Custom checklists
This is not meant to be the only ethical checklist, but instead we try to capture reasonable defaults that are general enough to be widely useful. For your own projects with particular concerns, we recommend your own checklist.yml file that is maintained by your team and passed to this tool with the -l flag.
Custom checklists must follow the same schema as checklist.yml. There must be a top-level title which is a string, and sections which is a list. Each section in the list sections must have a title, a section_id, and then a list of lines. Each line must have a line_id, a line_summary which is a 1-3 word shorthand, and a line string which is the content. The format is as follows:
title: TITLE
sections:
- title: SECTION TITLE
section_id: SECTION NUMBER
lines:
- line_id: LINE NUMBER
line_summary: LINE SUMMARY
line: LINE CONTENT
Changing the checklist
Please see the framing for an understanding of our perspective. Given this perspective, we will consider changes to the default checklist that fit with that perspective and follow this process.
Our goal is to have checklist items that are actionable as part of a review of data science work or as part of a plan. Please avoid suggesting items that are too vague (e.g., "do no harm") or too specific (e.g., "remove social security numbers from data").
Note: This process is an experiment and is subject to change based on how well it works.
A pull request to add an item should change:
deon/assets/checklist.yml: contains the default checklist itemsdeon/assets/examples_of_ethical_issues.yml: contains example of harms caused when the item was not considered
The description in the pull request must include:
- A justification for the change
- A consideration of related items that already exist, and why this change is different from what exists
- A published example (academic or press article) of where neglecting the principle has lead to concrete harm (articles that discuss potential or hypothetical harm will not be considered sufficient)
See detailed contributing instructions here.
Discussion and commentary
In addition to this documentation, the wiki pages for the GitHub repository are enabled. This is a good place for sharing of links and discussion of how the checklsits are used in practice.
If you have a topic, idea, or comment that doesn't seem right for the documentation, please add it to the wiki!
References, reading, and more
A robust discussion of data ethics is important for the profession. The goal of this tool is to make it easier to implement ethics review within technical projects. There are lots of great resources if you want to think about data ethics, and we encourage you to do so!
Checklist citations
We're excited to see so many articles popping up on data ethics! The short list below includes articles that directly informed the checklist content as well as a few case studies and thought-provoking pieces on the big picture.
- Of oaths and checklists
- How to build ethics into AI (Part I and Part II)
- An ethical checklist for data science
- How to recognize exclusion in AI
- Case studies in data ethics
- Technology is biased too. How do we fix it?
- The dark secret at the heart of AI
Where things have gone wrong
To make the ideas contained in the checklist more concrete, we've compiled examples of times when things have gone wrong. They're paired with the checklist questions to help illuminate where in the process ethics discussions may have helped provide a course correction.
We welcome contributions! Follow these instructions to add an example.
Related tools
There are other groups working on data ethics and thinking about how tools can help in this space. Here are a few we've seen so far:
- Aequitas (github)
- Ethical OS Toolkit
- Ethics & Algorithms Toolkit: A risk management framework for governments
- Ethics and Data Science (free ebook) and (write-up)
deon was created and is maintained by the team at DrivenData. Our mission is to bring the power of data science to social impact organizations.