By Cohen Snelling
Cohen Snelling is a teenager from Perth, Western Australia. He approached EFA for help sharing his concerns about WA Police’s AI surveillance vans. This post, and the accompanying social media content on EFA’s Facebook and LinkedIn platforms, is the outcome of that collaboration.
Hi, my name is Cohen Snelling and I am a teenager from Perth, Western Australia. Right now I am really concerned by the rollout of these WAPol mobile vans with AI-powered facial recognition tools.
WAPol are testing new AI-enabled surveillance vans across WA, specifically the Perth region. These vans use artificial intelligence to scan large crowds of people and run all images through biometric facial recognition software to identify supposedly known criminal offenders or persons of interest.
Who are the suspects? Everyone who walks past these WAPol cameras.
At face-value, this sounds like a way to keep WA citizens safe, but once you dig deeper, you find the real risks of this technology, the most major being the fact that there are currently no specific laws dealing with AI, biometrics and facial recognition software in any significant way. This means the WAPOL have been given little to no restrictions regarding this technology. In addition, WAPol bypassed the Western Australia Office of the Information Commissioner which surely they are expected to do. Instead, they rolled out the trial of this privacy invasive technology without ever consulting the WA Information Commissioner.
The real risk: false positives
Here is the actual, genuine risk of this technology: false positives, specifically for women and ethnic individuals. Looking at the 2018 MIT Media Lab study Gender Shades, we can see that there are higher false positive rates for female and/or ethnic individuals.
U.S. government-backed benchmark statistics (IJB-A)
- Percentage of the IJB-A made up of light-skinned males: 79.6%
- Misidentification rate for dark-skinned women: 34.7%
- Misidentification rate for light-skinned males: 0.8%
This happens because of a lack of ethnic, and gender representation in these benchmark databases, causing a learning gap in the AI’s facial recognition proficiency. This learning gap could cause detrimental effects, especially since Australia/WA is a multicultural and diverse society, this technology could potentially lead to people of colour being disproportionately and wrongly stopped by police, since they would have a higher chance at being falsely flagged.
Additionally, even if these learning gaps weren’t a problem, a tiny error rate could cause problems in large numbers. As EFA have pointed out in a media release on this issue, a crowd of 100,000 people, with a 0.1% false-positive rate would still cause approx. 100 false positive alerts. If the police are looking for a specific suspect, then this would cause 99 innocent members of the public being accosted by police and potentially interviewed by them. There is also the risk of a false arrest arising from this technology.
If you’re anything like me, this all sounds like an extremely dangerous overreaction by WAPol, supposedly in the interests of public safety. I do not see it that way. WAPol is treating everyone in a public space in WA as if they were already criminals.
As EFA has previously advocated: we are citizens, not suspects.
My call to action
If you are as concerned as I am about the loss of privacy in public places, and about WAPol trampling over our right to privacy in public spaces, then please sign my petition on change.org: Stop WAPOL AI Surveillance Vans In WA.
Thank you to EFA, and in particular its Chair, John Pane, for helping me to highlight these new privacy harms being suffered by Western Australians.
Image credit: Change.org