Huge thanks to speakers and organisers ( @DrAliceTowler , @dunnhappy) for making @ufigAustralia 2021. 300 registrations, virtual head count of 180 -- so the biggest @unsw UFIG in our 8 year history. Some key points in this thread from my notes, videos of some talks to follow 1/n

Growth of @ufigAustralia reflects the growing importance of #FacialRecognition #AI #biometrics #forensics in modern society (2/n)

Wide use of #FacialRecognition by law enforcement in the US, but use is not disclosed to courts. False positive errors reported in media may be the tip of an iceberg @ClareAngelyn (3/n)

We do not know the accuracy of #FacialRecognition in operation, because it depends on complex human interactions e.g. police officers' choosing an image to search with, choosing matches from gallery to follow up. A sociotechnical problem without a solution @ClareAngelyn (4/n)

Some of the most informative -- and courageous -- talks at UFIG2021 were from #FacialRecognition and #forensics practitioners in police opening up and explaining what they do in their work -- in close detail -- to scientists (5/n)

Recent international survey (UK US Australia China): the general public accept police using #FacialRecognition technology for some tasks (e.g. searching criminal databases), but oppose others (tracking citizens). People want regulation and control of use @kayritchiepsych (6/n)

Wide use of #FacialRecognition by police in UK USA Australia means legislation playing catch up. Codes of #ethics published by e.g. @MicrosoftAI , @BiometricsInsti a useful start but do not safeguard human rights. But no substitute for regulation, legislation @lizjcampbell (7/n)

Thanks everyone for attending and all the great speakers! @carinahahnphd @C_AMSutherland @mfmatheson @danferguson84 @ntrtrn @Victor_Varelaa @pjbhancock @akbobak @MeikeRamon @BindemannMarkus @Thompsontweets @socpsychupdate @mattfysh @younan_mariam @MeheraSanRoque @amy_n_yates

Tag any Twitter attendees I have missed please @BethanyGrowns @DrDanCarragher @drjbeaudry @FavelleSimone @GRecognisers

3 myths about FR bias debunked by Jackie Cavazos: 1) Race #bias in #FacialRecognition is unique to #AI (humans are more biased); 2) Algorithms were fair before #DeepLearning (AI from 2002 was biased); 3) Race is categorical (the labels do not have genetic basis) (8/n) ...

Bias in #FacialRecognition and its measurement, is more complex than media reports suggest. e.g. race categories themselves do not have scientific basis (J Cavazos), but race categories do carry legal implications ( @John_J_Howard). An example of problems in science<->policy link

Explainable #AI researchers must work with psychologists to develop measures of explainability. Whether an explanation is understandable and meaningful, is a question that can only be answered via #psychology research. Jonathon Phillips, NIST, see: https://www.nist.gov/artificial-intelligence/ai-foundational-research-explainability (9/n)

Problems for Explainable AI decisions also apply to human decisions (e.g. forensic scientists) [Jonathon Phillips, NIST]. Kristy Martire shows people don't understand #forensics explanations. Learn to understand each other before we understand the robots? (10/n)

#Cognitive #psychologists and #AI researchers typically do not understand the applied practice of #FacialRecognition, and should listen carefully to practitioner communities more often @mikeburton47 (11/n)

Practitioners often mistake disagreement in scientific communities as a weakness in scientific evidence (yes, scientists disagree, but that doesn't undermine the evidence. Disagreement is how scientific progress is made) @mikeburton47 (12/n)

Practitioners and policy makers: Scientists and researchers need to generalise from their results (meaning they typically will not know accuracy for the precise task under the precise conditions that a given practitioner made a given decision @mikeburton47 (13/n)

Experts in facial #forensics have disappeared from Australian courts since Honeysett V The Queen (2014) [Janice Yung] ... throwing the baby out with the bath water?(14/n)