Joshua Kroll

Computer Scientist. Hacker. Recovering Physicist. Assist. Prof researching trustworthiness, governance, and values in computing @NPS_Monterey (tweets personal).

[Joshua Kroll @realjoshkroll]

My paper, "Unsafe at any AUC: Unlearned Lessons from Sociotechnical Disasters for Responsible AI", with @andrewthesmart (Google Research, @staeiou (UCSD), and Abigail Z. Jacobs (University of Michigan), is now in the Harvard Data Science Review! https://t.co/X7Yg450JBu

[Where the Tweets have no name @andrewthesmart]

Really happy this is out finally!!

[Joshua Kroll @realjoshkroll]

The determinants of safety are not social/organizational or technical alone, but sociotechnical. Disasters provide a good reference for this, as they mark the moments when complex systems receive exquisite scrutiny. Astounding parallels echo across accidents in many domains.

[Joshua Kroll @realjoshkroll]

My paper, "Unsafe at any AUC: Unlearned Lessons from Sociotechnical Disasters for Responsible AI", with @andrewthesmart (Google Research, @staeiou (UCSD), and Abigail Z. Jacobs (University of Michigan), is now in the Harvard Data Science Review! https://t.co/X7Yg450JBu

[Joshua Kroll @realjoshkroll]

By treating these as first-order engineering concerns, it is possible to make order out of the chaos of AI evaluation and safety techniques and tools. Without treating safety as a system property, for example by focusing on component reliability/performance, failure follows.

[Joshua Kroll @realjoshkroll]

Proud to be part of an august group of tech policy scholars! https://t.co/HJt2kEpPbB

[AS @agstrait]

Anyway, the paper is excellent, and includes much more than what's above. Massive kudos to @ShalalehRismani @renee_m_shelby @andrewthesmart Edgar Jatho @realjoshkroll @RoboEthics @negar_rz

[Negar Rostamzadeh @negar_rz]

Our #CHI2023 paper *From plane crashes to algorithmic harm: applicability of safety engineering frameworks for responsible ML* led by our incredible intern @ShalalehRismani will be presented on Tue, April 25 @ 17:17
https://t.co/PdlfxkhaV2
https://t.co/14seMqiYzZ
1/4

[The Regulatory Review @TheRegReview]

New series on #AI & procurement featuring a wide-ranging exchange of ideas by @AshleyCasovan, @az_jacobs, Carlos Ignacio Gutierrez, Cary Coglianese, @dchenok, @DMulliganUCB, David Rubenstein, @JTillipman, @realjoshkroll, Lavi Ben Dor, Tim Cooke & @var0101. https://t.co/9uvWXXSLHn

[Joshua Kroll @realjoshkroll]

Honored to be part of a distinguished group writing on “AI and Procurement” for @TheRegReview!

[The Regulatory Review @TheRegReview]

Responsible AI systems must comply with applicable law, mitigate risk to stakeholders, and eliminate structural biases, argues @realjoshkroll of @NPS_Monterey. https://t.co/LQUNmONoro

[ACM Public Policy @acmpolicy]

Just out! New https://t.co/7QDmdtLF0t on #FacialRecognition says its "fundamental limitations" create "profound #privacy and ethical challenges," system bias is "pervasive and profound," and FR is "insufficiently trustworthy" to replace human ID. Online at https://t.co/DPRyK0F2Jo

[Joshua Kroll @realjoshkroll]

Pleased to release a new paper, “Outlining Traceability: A Principle for Operationalizing Accountability in Computing Systems”, to appear in the 2021 @FAccTConference. https://t.co/bfCRLhDuvk.

[Joshua Kroll @realjoshkroll]

@Broniatowski @az_jacobs @drturnerlee @krvarshney Thanks, David! It was a fun day. The recording will come out someday, too.

[Joshua Kroll @realjoshkroll]

@math_rachel All insight about measurement problems in ML is due to the amazing @az_jacobs and @hannawallach, who have a forthcoming paper on this topic that we should have referenced more clearly. A critical and often overlooked deep subject worth significant attention and study!

[Dr. Gabriela Zanfir-Fortuna @gabrielazanfir]

@realjoshkroll @jreidenberg ~ 'Transparency alone will not solve all issues abt #algorithms and their accountability' #PrivacyPapers

[Benjamin Herold @BenjaminBHerold]

How do you make algorithms accountable? Transparency isnt enough, say @jreidenberg & @realjoshkroll #privacypapers

[Benjamin Herold @BenjaminBHerold]

Oppty for technologists & policymakers to build better automated decision-making systems @jreidenberg @realjoshkroll #privacypapers