Amazon says it has added “fear” to the range of emotions that its Rekognition facial recognition software can now detect
Release follows amid immigration protests — AWS has added “fear” to the range of emotions identified by its Amazon Rekognition facial analysis service …
Computer Business ReviewEd Targett
Context & Ripple Effects
Rekognition has been on an expansion cadence since AWS added real-time face search and text detection in late-2017 feature updates and then shipped Rekognition Video for analyzing live streams. The 'fear' label extends that same pipeline from identifying who is in a frame to inferring an internal state.
The timing lands in the middle of a bruising stretch for the product: in February, reporting showed Rekognition's only known law enforcement client skips the confidence thresholdAmazon cites in its defense against bias research, and days later Amazon joined Microsoft in backing regulation while publishing 'responsible use' guidelines. Adding a new emotion category amid immigration protests tests whether those guidelines constrain anything.
First-order effects
Government and law enforcement customers can now flag 'fear' alongside existing emotion labels in face and video analysis, widening what a single Rekognition query returns about people in protest footage.
Amazon's own responsible-use framing is immediately on the hook: the guidelines it published in February now have to cover a capability that infers emotional state, not just identity.
Second-order effects
Civil-rights critics gain a sharper test case: if emotion labels are as error-prone as the bias research suggested for face matching, every misread 'fear' flag in protest or immigration contexts becomes evidence in the accuracy debate Amazon is already losing ground on.
Rivals selling emotion or facial analysis to the same public-sector buyers face pressure to match the feature set — or to differentiate on refusing it, turning capability breadth into a governance talking point.
Third-order effects
The pattern points toward affect recognition becoming a standard layer of surveillance infrastructure, with regulators forced to decide whether inferring emotion from faces gets the same scrutiny — thresholds, audits, accuracy standards — that face matching is starting to receive.
If the confidence-threshold gap persists across new capabilities, the practical constraint on these tools shifts from what vendors ship to whether procurement rules force customers to actually use the safeguards vendors advertise.
The trend: Facial analysis is expanding from identifying people to inferring their emotional states, with each added label arriving faster than the accuracy and governance standards meant to govern it.
Rekognition's claim that it can detect “all seven emotions” is a perfect opportunity for a reminder about how the machinic metaphysics of data science operates: https://twitter.com/... https://twitter.com/...
Rekognition new emotion: ‘Fear’, improved age range estimation accuracy; you also get narrower age ranges across most age groups. #rekognition #aws https://aws.amazon.com/...
How long until a shooting of a Black person is captured and the lawyers argue they asked for it because the AI didn't detect fear on their face? https://twitter.com/...
Setting aside the complete bullshit of emotion detection (hint: facial expressions are contextual, their meanings socially constructed), Amazon provides no data to support these claims. Just self-congratulatory PR, which sadly, even @motherboard seems to accept without question. …
Any article that can say this without also referring to the meta survey by @LFeldmanBarrett and others is failing to call BS on Amazon's claims https://twitter.com/...
This blog post from Amazon is...🤬🤬🤬 They claim to have “improved the accuracy of gender identification” and improved accuracy detection “for all 7 emotions” 😱 https://aws.amazon.com/...
I like how Amazon adding “fear” as an emotion analyzed by its facial recognition system reads as if it's like new addition to the McDonalds dollar menu. https://aws.amazon.com/... https://twitter.com/...
How many more of our ancestors, hiding and using fake names, would have died if the Nazis had a machine that could detect fear? Why would you build a machine like this, and then try to sell it to ICE @Amazon? #NeverAgainIsNow https://www.theverge.com/... https://www.cbronline.com…
i'd love to put the morons using MRIs to find transgender brain structures in the same room as the morons using face images for gender identification https://twitter.com/...
ALL 7 EMOTIONS YOU GUISE. ALL 7! With that type of robust objectivity we can trust @amazon will never feed false results to the fascists running this govt. No way will this tech be used to sort human beings into categories marked as deserving and not-deserving of human rights. ht…
Beyond the unscientific ‘emotion recognition’ aspect, this press release boasts about how Amazon's software can ‘identify gender’ with ‘improved accuracy.’ This is problematic. https://twitter.com/...
You can improve gender recognition all you want me and @morganklauss already wrote papers proving it's completely fucking impossible ya technochauvin-cis'ts https://twitter.com/...
Amazon claims it's improved accuracy for its facial recognition tool, which can now recognize fear. Academics spent two years reviewing over 1,000 papers and concluded that emotion detection still can't be trusted: https://www.technologyreview.com/ .... https://twitter.com/...
Are you fucking kidding me? AWS comes under fire for Rekognition sales to the federal government, who in turn is building concentration camps for children, and AWS's response is to improve “age range estimation” and “fear detection” in the service? Are you FUCKING KIDDING ME?! ht…