Thread by @AndreaCoravos

(1/x) Similar to 'drug labels' and 'food nutrition labels', it is high time to develop labels for algorithms and connected products.

Thread inspired by this thoughtful piece published by @erbrod in @statnews this morning 👇

https://twitter.com/erbrod/status/1313103898031386625

(2/x) About a month ago, @lauradiang asked this provocative question.

Turns out... many teams are prototyping labels. Let's dive in and check out what folks are building.

https://twitter.com/lauradiang/status/1294398762794745856

(3/x) But first, some background.

Both medical devices and drugs can have an enormous impact on human lives—the trick is in balancing the benefits and risks.

🔎 What are mechanisms can you can use to evaluate a complex product?

(4/x) Consider drug labels and food nutrition labels. Labels are a powerful tool because they can make risk-benefit decisions more digestible when considering complex products.

⚠️ Today, tech products *don't* come with a warning label, but they could.

(5/x) BTW - I'm fascinated by the label topic because our team @ElektraLabs wants to know: "What connected sensors should YOU trust to collect your biometric data at home?"

Working with industry colleagues, here's our current hypothesis.

https://twitter.com/AndreaCoravos/status/1238546093362647041

(6/x) The big tech players are getting in on this game.

Ex: this summer, @Apple announced a "privacy label" that's modeled after the nutrition labels on food packaging, an idea that’s been promoted for years by privacy experts like @lorrietweet

consumerreports.org/privacy/apples…

(7/x) @lorrietweet along with Pardis Emami-Naeini and Yuvraj Agarwal released a "Privacy and Security Nutrition Label for Smart Devices" via the @ShorensteinCtr

privacy.shorensteincenter.org/nutrition-labe…

(8/x) @MarkSendak, @michaelgao88, and team drafted “Model Facts” labels as a communication tool for machine learning models to present model info to clinical end users

This was published this past March, which inspired @erbrod's original post

https://twitter.com/DrXiaoLiu/status/1244555298322419713

(9/x) The "Model Facts" work built off concepts @mmitchell_ai @timnitGebru who pioneered some of the first model cards for model reporting: arxiv.org/pdf/1810.03993… h/t @suchisaria

(10/x) And the work from @DartmouthInst on improving drug labels vox.com/2015/9/26/9398…

... and also work out of FDA on risk communication fda.gov/media/81597/do… h/t @MarkSendak

(11/x) The Dataset Nutrition Label Project (DNLP) was created by @hollasarahd, @ahmedhosny, Sarah Newman, Joshua Joseph, @kaschm in 2018 at @BKCHarvard datanutrition.org

(12/x) @kaschm has continued with the project and recently joined @ConsumerReports as a Digital Lab Fellow to build out a new set of tools and frameworks for the safe and effective use of datasets

https://twitter.com/ConsumerReports/status/1302000951692333060

(13/x) @ConsumerReports had previously come out with the Digital Standard - which has elements similar to a label. Its a set of principles for evaluating connected products and services for how well they live up to consumers' expectations thedigitalstandard.org/the-standard

(14/x) @DCMS commissioned an investigation into both the effectiveness of the labels and potential premium pricing for label-carrying IoT products.

Check out 20+ page report assets.publishing.service.gov.uk/government/upl…

(15/x) Similar to a Nutrition Label, the industry’s Data Transparency Label by @IABTechLab is intended to give every marketer, agency, data provider and publisher a clear view of the syndicated audience segments they use datalabel.org

(16/x) Drafted by @agropper, Patient Privacy Rights Information Governance Label has five separately self-asserted statements papers.ssrn.com/sol3/papers.cf…

(16/x) The team at Symantec Research Labs crafted proposal, after becoming more concerned about the growing number of home IoT devices symantec-enterprise-blogs.security.com/blogs/expert-p…

(17/x) @MeganStifel, @markpetersn and Dylan Gilbert designed 'Security Shield', a label to support sustainable cybersecurity publicknowledge.org/documents/secu…

(18/x) Trustable Technology Mark is an attestation-based trustmark for IoT devices, developed by @thingscon and supported by Mozilla Foundation.

(19/x) One last concept: consider the similarities between developing drugs, algorithms, and IoT — Artificial intelligence doesn’t come with a warning label, but it could.

qz.com/1540594/treati…

h/t @arieldora @irenetrampoline @AnkitGordhandas

(20/x) BIG thank you to my co-authors on the Nature npj Digital Medicine manuscript (nature.com/articles/s4174…) for sourcing these examples:

@beauwoods
@ChristineManta
@markshervey
@GoldsackJen
@bwoodmd
@MegDoerr

(21/x) I'm sure there's more label prototypes out there than the ones listed in this thread for algorithms and/or connected sensors/IoT.

What else have you seen?

Excited to see prototypes go into production! Make it easier to use technologies that are worthy of our trust⚡️

(22/x) And most of all: @HealthEDena who has some exciting updates coming up on labels over the next couple months.

She's a great follow if you want to be tracking these themes.