PRC Minority Ethnicity Recognition Research Targeted Uyghurs, Breached Ethical Standards

Published Sep 13, 2021 14:14 PM
PUBLIC - This article does not require an IPVM subscription. Feel free to share.

A research paper analyzing "Chinese ethnical groups" focused on "Korean, Tibetan, and Uyghur" ethnicities has been found by an Australian University to breach core ethical standards, with the University's Professor abruptly resigning, leaving to a PRC University.

IPVM Image

Australia's Curtin University determined that Wanquan Liu "did not gain informed consent" or "required ethical approvals" for the study which used AI-powered analysis of "ethnical facial features" to best predict whether a person is Uyghur, Tibetan, or Korean.

Liu "refused to respond" to parts of Curtin University's investigation, resigning and moving to one of the PRC's top 10 universities. The study's academic publisher Wiley is "reviewing" whether to retract it, while a different study based on its data has already been retracted.

The incident showcases the serious ethical risks of 'ethnicity analytics' research. This news comes as The New York Times reports that Chinese DNA studies that included Uyghurs' genetic material were retracted for failing to gain subjects' full consent.

Paper Background

In 2018, a paper titled "Facial feature discovery for ethnicity recognition" was published, co-authored by Curtin University's Wanquan Liu and three other academics from two PRC universities:

IPVM Image

The study photographed 300 Uyghur, Tibetan, and Korean university students - 100 for each ethnic group - at Dalian Minzu University, a PRC tech university for ethnic minorities:

IPVM Image

The students, aged 18-22, were photographed in ideal pose/lighting conditions to construct a "Multiethnic Group Database" of frontal images:

IPVM Image

The authors "extract[ed] ethnic salient features" from the photos, concluding that "T3", a face area that includes the eyes, mouth, and nose, had the highest ethnicity recognition "accuracy" at 78%.

IPVM Image

An example of the claimed accuracy identifying Uyghurs, Tibetans, and Koreans from the paper:

IPVM Image

However, the study found these T-regions were "not suitable for general face recognition", only "ethnicity recognition":

IPVM Image

The study did not give a detailed purpose other than it may be "helpful to the research in anthropology as it may indicate the facial features evolution".

Uyghur Recognition Widely Used By PRC Police

PRC police widely deploy Uyghur recognition technology to be automatically notified any time a Uyghur-looking person is filmed by security cameras. The PRC's top video surveillance manufacturers and facial recognition companies have been caught offering the technology: e.g. Huawei and Megvii's 'Uyghur Alarms', Dahua's 'Real Time Uyghur Warnings', Hikvision's Uyghur-detecting AI camera, etc.

While IPVM has found no evidence of Tibetan or Korean ethnicity analytics, the PRC also harshly represses Tibetans and regularly tracks and deports North Korean refugees back to North Korea.

Curtin University Investigation: "Did Not Gain Informed Consent" Or "Ethical Approvals"

The paper inadvertently went viral when computer science grad student Os Keyes tweeted it out in May 2019; at the time Curtin responded it was "reviewing its research approval procedure".

In an August 2021 letter obtained by IPVM, Curtin University found Liu "breached" the Australian Code for the Responsible Conduct of Research for failing to "gain informed consent" of the PRC university students nor "required ethical approvals":

IPVM Image

No Informed Consent

IPVM obtained two documents the paper's authors sent to Wiley: an undated, unsigned consent form and a "University Approval" letter from Dalian Minzu University.

However, the consent form is bare-bones and does not give any details about the research's purpose beyond the fact that it is "non-commercial", failing to meet "informed consent" standards for Human Subject Research (HSR) which mandate detailed descriptions of the research's purpose and "special protections" for "vulnerable populations". Specifically, the provided documents did not disclose to the subjects of the study that the study would be used to identify specific ethnicities.

Meanwhile, the approval letter is dated two months after the study was submitted to Wiley, raising questions about why Ethics Committee approval was not obtained before as is standard for HSR.

Liu Now Teaching In PRC

The letter also said Liu had resigned and is "now understood to be a Professor at Sun Yat Sen University" which is confirmed by the university's own website where he is listed as a Professor of Intelligent Control Theory and Engineering since May 2021:

IPVM Image

SYSU is regularly ranked among the top 10 universities within the PRC. IPVM contacted SYSU about Liu but it has not responded yet; if they do, we will update.

Curtin University Calls For Paper's Retraction

In the letter, Curtin University called for the paper "to be retracted" and any reference to Curtin "should be immediately removed":

IPVM Image

Wiley Originally Defended Study For 'Enriching' Field 'Beyond White, Caucasian'

In November 2019, the paper's publisher, academic publishing giant Wiley, originally defended the paper, stating that facial recognition research "is enriched by expanding beyond white, Caucasian subjects":

We respectfully reiterate that this article is about a specific technology, not any application of that technology. Facial recognition technology has been designed using mostly white, Caucasian subjects (such as by Facebook and Google). The technology and the research behind that technology is enriched by expanding beyond white, Caucasian subjects. [emphasis added]

However, the paper is not about "facial recognition technology" as Wiley stated but for improving ethnicity recognition technology, as its own authors explicitly noted:

IPVM Image

Wiley Now "Reviewing" Whether To Retract

However, Wiley told IPVM it is now "reviewing the matter again" given Curtin University's findings, noting it had already issued a Publisher's Note and an Expression of Concern:

WIREs Data Mining and Knowledge Discovery has previously launched an investigation in accordance with the Committee on Publication Ethics (COPE) guidelines, which resulted in a Publisher’s Note and an Expression of Concern. We take every concern seriously and are reviewing the matter again taking into account the new information provided by Curtin University.

Resignation Details

Curtin University's letter did not specify why Liu resigned, while an email obtained by IPVM from Curtin's Deputy Research Vice-Chancellor Chris Moran stated Liu "refused to respond" to "aspects of the allegations" and "became ill" in the middle of the investigation, then resigned: IPVM Image

in short, the researcher, who became ill, has now resigned and we are moving to convene an independent external panel to investigate the aspects of the allegations that he refused to respond to and for which we cannot draw conclusions.

Liu did not respond to a request for comment and Curtin University gave IPVM a generic statement avoiding our questions about why he resigned:

IPVM Image

Curtin University works to the Australian Code for the Responsible Conduct of Research.

We have well-developed policies and procedures for dealing with complaints and/or alleged breaches of the Code.

This complaint was handled consistent with those policies and procedures.

Retraction Of Other Study

A separate study based on the database of Uyghur/Tibetan/Korean faces been retracted by its publisher, the IEEE, which said the retraction was for two reasons - the face database was deleted and its authors "did not respond on whether" consent was obtained:

IPVM Image

"We Have Been Playing With Fire"

Yves Moreau, an engineering professor at KU Leuven University in Belgium and self-described "concerned scientist", has been pushing for the Wiley paper's retraction since 2019 and provided the letters/evidence above.

Moreau told IPVM that the facial recognition/biometrics research community has been "playing with fire in an arsenal" when it comes to ethics, calling for Wiley to retract the study and a block on certain types of PRC research targeting vulnerable groups:

IPVM Image

Do we really need to have models that [track] groups like Tibetans and Uyghurs that are really quite vulnerable in China?

Computer scientists should really start reflecting about what research is acceptable and not acceptable. We have been playing with fire in an arsenal with what we've been doing with surveillance technology

[the view that] 'tech is neutral, I have no ill intent, good people do good stuff with it, bad people do bad stuff with it', that's moral disengagement and that's wrong

I'm at a loss as to why Wiley does not want to see why this is not acceptable, there are so many issues as to why this should be retracted [...] we should block certain types of research from China

Comments are shown for subscribers only. Login or Join