very-good-science / data-ethics-club Goto Github PK

A CoC for the discussion space.
Examples:
Turing Way
Diversity Journal Club @ Arizona
Carpentries
Social Rules
JGI

• Discuss potential content (inc. how can people report incidents?)
• Prepare draft on repo (@ninadicara )
• Agree version for first meeting (@NatalieThurlby and @ninadicara)

I recommend Living in the Hidden Realm of AI from the Tech Worker's Coalition blog. It is a fairly quick read.

Some discussion ideas: How far down the pipeline should our ethical responsibility as Data Scientists reach? What are better ways of recognising the contributions of annotators who are crucial to the existence of clean data sets and the products based on these data sets? What are our future visions of labour in a world that is becoming more Data and AI centric?

Happy to be in the contributers page as Vanessa Hanschke

I've added the feb folder as '02-feb' so that the folders will display in date order (pedantic, I know!).
Jan would then be '01-jan' but if I change it now none of the links we've sent out will work so I'll leave it until after the jan meeting!

This is just to remind myself to do it.

It's not a very clear distinction - we could probably come up with better wording!

• For organisers:
  • How to add people using the all-contributors bot (with and without gh username).
  • How/when to email the mailing list (perhaps we can keep a password in our Sharepoint?)
  • How to run a meeting:
    • Poll next choice of paper.
    • (IDEA) We could have a poll at the end where we ask everyone "Would you recommend this content to a friend?" and "Did this content provoke a good discussion?", then our past papers would be more useful for other people.
• For participants:
  • How to suggest reading materials
  • How to become an organiser
  • How to lead a reading group

There is no awesome list for data ethics: https://github.com/sindresorhus/awesome, and the awesome papers list’s ethics section is… one paper. Also, I think it’s kind of orthogonal because we also have blog posts and things.

Anyway, maybe we can make our reading list an “awesome” list if it grows a bit unwieldy?

EDIT: also this was chris edsall's idea so I should add him as an ideas contributor in the all-contributors if we do this!

• Set up a mailing list: Mailchimp @ninadicara
• Advertise mailing list (along with advertising first session - advertising streams #5)

How easy it is to set this up - Nina to look into it. Would be helpful to know how many people actually use the site, and which pages get used.

Set up HackMD syncing with:

• Try one paper review syncing with gh-pages?
• For paper reviews: a website linking them, like this
• Write instructions letting people know how to use it best
• Figure out if it's possible to update all-contributors automatically based on contributions

Title: The Financial Modelers Manifesto
Type: 2 page doc, easy read.
Link: https://www.uio.no/studier/emner/sv/oekonomi/ECON4135/h09/undervisningsmateriale/FinancialModelersManifesto.pdf

I feel like this is a good jumping off point for conversations about...

• What would a data scientists' hippocratic oath look like? especially in behavioural research!
• Who should hold research data science to account?
  (There might be stuff on this already, feel free to ignore if so!)

Name: Measuring the predictability of life outcomes with a scientific mass collaboration (Salganik et al., 2020 in PNAS)
Type: Research Article
Link: https://www.pnas.org/content/117/15/8398

Questions 🤔
Will it ever be feasible to 'predict' life outcomes?
What are the boundaries of when it is ethical to use these algorithms (given some are in use right now)?

• Add writeup template stuff to hackmd template
• Add instructions for getting the writeup on the website to the template
• Get feedback from Huw
• Add instructions for updating meetings.md with the link (as well as writeup.md)

How should we categorise the reading list?

• Broad topics (e.g. "machine learning", "statistics", "EDI", "law")
• Narrower topics (e.g. "explainable AI", "algorithms", "accessibility", "privacy")
• Past/Planned/Not yet read?

How should we organise the reading list:

• Separate READINGLIST.md, signposted to in README?

Link: https://towardsdatascience.com/statistics-eugenics-and-me-29eaf43efac7

This is a blog about the authors 'personal reckoning' with the history of eugenics in statistics in the context of having a personal connection to the holocaust. It is relatively short and discusses Karl Pearson, Francis Galton, Ronald Fisher as the 'fathers of statistics'

We're planning to send out a short survey to help us plan for the next block of Data Ethics Club:

The draft is here

• Edits:
  • Third question down little typo on the word totally -> 'totally uninteresting'
  • Add "or suggest a more convenient time to meet" to examples for "anything else" question
  • Add "Include your GitHub username (or name and website) if you'd like us to credit you for this suggestion on GitHub".
• Send out survey to mailing list
• Tweet survey + ask JGI to retweet
• Add survey to JGI mailing list

Name: Missing Datasets
Type: A short-ish README accompanying the "missing datasets" project
Link: https://github.com/MimiOnuoha/missing-datasets

suggestion to watch

highly recommended
Code Bias Virtual Screening available from March 19 – 29, 2021 pass-code: CUCODBIA21 https://www.wmm.com/virtual-screening-room/coded-bias-watch-page-clarkson-university/

• Using something akin to the all-contributors guidelines

This issue is visible on the footer of all website pages, e.g. here

Gives the impression that materials are copyrighted in a way that we don't want people to use them, so instead it should explain the license (CC-BY), similar to how the Data Hazards website does.

Now that the reading list is in its own file we should probably have a more obvious "READING LIST IS OVER HERE" section :)

This is an article about the role of lithium mining in AI:
https://www.fastcompany.com/90618225/what-an-ancient-lake-in-nevada-reveals-about-the-future-of-tech?partner=rss&utm_campaign=rss+fastcompany&utm_content=rss&utm_medium=feed&utm_source=rss

Need to add everyone who has logged their name for a meeting to the all-contributors list.

DEC 2.0 - lets go! I've tried to do a brain dump of all the things we need to do and link to new or existing issues where relevant so we can keep track.

Aiming to restart 14th July

Organising tasks

• Organise an organisers meeting to organise the upcoming series (@ninadicara @NatalieThurlby @HDiscoDay)
• Set up a DEC HackMD account (#52) (@NatalieThurlby)
• Set up a new recurring Zoom meeting link that has all of us as hosts (@ninadicara)
• Finalise the write ups from the last couple of meetings on HackMD (#68) (@HDiscoDay to do one)
• Update the MEETINGS.MD file with new meeting times (@ninadicara @NatalieThurlby)
• Decide on material for the first meeting back (@HDiscoDay)
• Send out a survey for feedback? ( #74) (@NatalieThurlby)
• Set up calendar invites (@NatalieThurlby)

Repository maintenance tasks

• Check all our contributors are up to date (#62, #38)
• Add any outstanding suggestions to the reading list (#64 #61)
• Add a more obvious link to the reading list from the README (#60)

Other

• Website development (#41) (@NatalieThurlby)

The HackMD is now set up so that if we log in using the data-ethics-club login details, we can simply click the three dots next to "New" and choose "New from template" -> "Meeting discussion" template, instead of copying and pasting:

• Replace current instructions in facilitating.md
• Write instructions for how to set up your own club (e.g. make an account for HackMD)

• Choose date(s) for first sessions
• Decide how often we should meet
• Update README/or a dates markdown doc @ninadicara
• Advertise - fill in the JGI comms sheet @NatalieThurlby + send to mailing list when set up #6

Like the one on data hazards!

Currently the list of upcoming dates has the future ones in 2021.

I, for one, am happy to leave 2021 behind!

• Test deploy site on push (when does it trigger?)
• Create new build-site on PR
• Figure out if we can update automatically next-meeting link (in a sphinx extension?)

https://ieeexplore.ieee.org/abstract/document/5370707

Nudging Privacy: The Behavioural Economics of Personal Information by Alessandro Acquisti

Short paper about individual's preferences for privacy, how these can be conflicting and influenced by behavioural 'nudges' - describes how tech could implement those nudges for the benefit of society & individuals.

Discussion Qs:

• Do the article's statements about peoples' preferences for privacy align with your existing views?
• What are you views on the idea of nudging users towards certain sets of decisions? A related question might be who decides what those decisions should be.

I was scrolling tiktok and saw this goldmine: bit.ly/caseysclass

Will need looking through to identify some specific bits.

The clip I saw explained the concept of the pathetic dot which is the different things that can impact on people behaviour: law, architecture, social norms, and market. I thought that was cool.

• The name of the content: "Engineers’ Moral Responsibility: A Confucian Perspective"
• The type of content: paper (fairly long, 20 pages)
• A link where it can be accessed: https://link.springer.com/article/10.1007%2Fs11948-019-00093-4
• Any particular questions or points for discussion. TODO
• Are you happy to be credited for providing this recommendation in our contributors section? Yep

This issue is here to give us a specific place to talk to the all-contributors bot.
This is because we often need to add a contributor for a content suggestion, and the quickest way we know how to do this at the moment is to:

1. Ask the all contributors bot to add the person for content
2. Edit the README with the content link, and credit with the username + url that the all-contributors bot generates.

Note: If anyone knows of a better workflow for this, please comment here and tag @ninadicara or @NatalieThurlby to let us know!

The Mailchimp template is from the last set of meetings. It contains:

• the wrong zoom link
• it doesn't include Huw on the sign off
• it doesn't link to the website

@ninadicara - I hope to fix this soon, but I leave this here as a warning on the off chance I don't fix it before you must send an email.

Feedback from March 17th:

• In my room we got involved in a discussion on Q1 and didn't consider the other Qs until near the end, so it'd be helpful to consider other ways to capture the discussion that mean we think more holistically (maybe putting one space for notes after all the Qs, or encouraging us to read the Qs out at the start of the breakout, or something) (+1)
• Yes I agree with the suggestion above - it might be nice to have a general space - often the discussion deviates from the questions. Also, I found it a bit difficult to jump between all the questions and find the space for my group - if we keep the questions it might be good to have the main headings be groups and the sub headings as questions instead of the other way around.

The following links in the readme file return 404: Page not Found:

https://very-good-science.github.io/data-ethics-club/contents/meetings/next-meeting.html
https://very-good-science.github.io/data-ethics-club/contents/administration.html
https://very-good-science.github.io/data-ethics-club/contents/code-of-conduct.html

I think that the navigation on Data Ethics Club website has changed.

Templates (particularly the meeting template) could be made much more usable by using the style of markdown links where you just have to put in each link once at the bottom (see here for example.

• Edit meeting-template.md
• Check if other templates could be improved in this manner

• What are the headlines?
  • Good news?
  • Bad news?
• What actions can we take to improve?
  • Huw exists, so now we'll write up and share those a bit more timely-ly
  • No better time is jumping out for now
• What questions would we want to add to the survey next time?
  • Super-basic demographic e.g. "Do you consider yourself to be a data scientist?"

There are some outstanding write ups to do from the HackMD docs of the last series of Data Ethics Club.
They are:

• We created poverty - Apr 28th
  • Write @HDiscoDay
  • Add to website @NatalieThurlby
• Critical perspectives on computer vision - May 12th
  • Write @HDiscoDay
  • Add to website @NatalieThurlby
• Living in the hidden realms of AI - a workers perspective - May 26th
  • Write @HDiscoDay
  • Add to website @NatalieThurlby

I'd like to make sure that Data Ethics Club has a diversity of points of view and experience, and while I think we hit that really well in terms of disciplines within a university research and data science context, I'm wondering how we could reach out to new communities, particularly:

• interested members of the public
• small tech companies
• analysts in NHS/police/local + central government roles
• undergraduate/masters students

In addition, across all these categories and our existing community of data scientists and researchers, I also wonder what we can do to make Data Ethics Club more appealing and welcoming for a more diverse demographic of people. So, part of that might be advertising, so within the university, maybe we could reach out to the BAME network, or if we did a "decolonising data science" meeting, then the decolonise UoB group might be interested?

I'm leaving this issue here to gather some ideas for specific groups we could reach out to, or perhaps a special edition of Data Ethics Club that is in real life, invites the public, maybe at the JGI showcase (which will be in February at the MShed) for example. Then hopefully we can enact something from here at our next organiser's meeting.

TBC!

Options:

• zoom-style, like reproducibiliTea
• In an datawrapper/etherpad like data viz book club (e.g. https://notes.datawrapper.de/p/bookclub-datafeminism)

Decision: Zoom for now! (@NatalieThurlby will set up meetings), maybe etherpad later (after first term).

• Do it
• Put the login details in the Sharepoint doc
• Make sure there is a template in there for meetings

Name: Sharing learnings about our image cropping algorithm
Type: Blog
Link: https://blog.twitter.com/engineering/en_us/topics/insights/2021/sharing-learnings-about-our-image-cropping-algorithm.html

This is a blog about Twitter's investigation into complaints about its image cropping algorithm, which automatically chose which section of an image to display on Twitter if it was too big. They provide open code and describe how they evaluated the algorithm following complaints. They have decided to withdraw this feature, and let people choose their own preview section of the image.

This would fit well in the 'Ethics in Action' sub-category of the reading list.

Suggested questions:

• What do you think of Twitter's response to the complaints?
• Twitter have decided to withdraw the feature and lead this decision up to humans. Why do you think it was originally designed as an algorithmic solution?

This paper on ethics creep seems like something that should spark interesting discussion, and might be an opportunity for us to be a little more critical.

• Decide what to read/watch
• Add these to the 'upcoming meeting dates' markdown doc @ninadicara
• Add a document each in 'meetings'
  • 20/01 @NatalieThurlby (27/11/21)
  • 03/02 @ninadicara (20/01/21)

We want a gh-pages website for Data Ethics Club to make it more accessible.

We're going to try to use sphinx!

To-dos

• Create basic website
• Buy/add domain
• Update README and link to website so as not to duplicate information
• Update how-to-guide with updated workflows, including, e.g.:
  • How to update website
  • How to update next meeting link
  • New GH repo directory structure
  • Add all these to the website content under "Contribute"? And possible subcategory "Create your own Data Ethics Club"

Ideas

Must-haves:

• A nice landing page
• Easy to find information on our next meeting
• All the information that is in the README.
• Navigate our discussion write-ups

Nice to haves:

• Embedded mailchimp sign-up
• Sortable/searchable table for reading list: could be possible using a Jupyter Notebook page of the website using something like this itables package
• Zotero badges automatically assigned to all of the meeting write-ups, like SORSE.
• A nicely displayed calendar, where people can see upcoming events and download them to their calendars: this could potentially be done automatically either:
  • Using something like Open Research Calendar's approach, i.e. (using a GH Action to) run a script which extracts info from things in the meetings folder and adds them to a Google Calendar using the Google API - from there it should be easy to embed a google calendar.
  • Or populating an .ics calendar using JavaScript (SORSE use Full Calendar) from the markdown YAML, like the SORSE calendar or upcoming events (my fave option, which I think would work with sphinx)
  • Similarly, it would be cool to have a page which redirects to the information for the upcoming meeting. Ideally, we would have a GH pages action to automatically update to this when a new meeting is created, but perhaps to begin with, we can just manually add the link to the next meeting.
• (Future): Display the how-to-guide so that people can read that there.

(Waiting to check if Rob is okay with it)

• Create a template for "upcoming meetings" (which we can use when we send a reminder email)
  • Stuff from the issue template
  • Date that it's happening
  • Joining info: zoom link maybe to be sent out to mailing list for security?
  • Tempate bits and bobs for recording what's happening in the meeting and who came

Thanks for making a suggestion!

For each suggestion, please let us know:

• The name of the content
• The type of content (paper/blog/video, etc)
  • Is it a preprint?
  • Is it very long? If so, do you have a recommended section?
• A link where it can be accessed
• Any particular questions or points for discussion
• Are you happy to be credited for providing this recommendation in our contributors section?