You sit alone in a spare room, laptop propped on a stack of books to get the camera level with your eyes, and you wait for the first question to appear. There is no interviewer. There is a progress bar, a countdown timer, and a small green light beside the webcam that tells you the machine is watching. A prompt fades onto the screen. You have thirty seconds to think and two minutes to answer, and there will be no second take. So you talk. You describe a time you handled conflict, a moment you showed leadership, the reason you want this job, and the whole while you are performing not for a person but for a piece of software that is measuring something you cannot see and were never told about. It is not weighing your words alone. It is parsing the cadence of your voice, the micro-movements at the corners of your mouth, the length of the pauses where you gather your thoughts, the steadiness or flicker of your gaze. From these signals it is assembling a verdict: a number for your conscientiousness, a number for your emotional stability, perhaps a quiet flag against your honesty. You will never see those numbers. You will receive, days later, a templated rejection that thanks you for your interest and wishes you well in your search.
This is the invisible audition, and millions of people now sit through some version of it every year, mostly without realising what is happening on the other side of the glass. The asynchronous video interview, in which a candidate records answers for a machine to score rather than a human to watch, has become a standard gate in high-volume hiring. The market that builds these systems is growing at a clip that industry analysts put in the region of fifteen per cent a year, with one estimate projecting the AI video interview sector to climb from under a billion dollars in 2024 towards nearly three billion within the decade, and the largest vendors process tens of millions of interviews. Most of that scoring is uncontroversial enough on its surface: transcribing speech, matching keywords, ranking competencies. But a strand of the industry reaches further, into the body and the face, claiming to read from a candidate's expressions and vocal tone the inner traits that a CV cannot show. And here the story stops being about efficiency and becomes about something stranger and more troubling: a hiring decision built on a science that does not exist, applied to the people least able to object, by employers who believe the very opacity of the process makes it fair.
What sets the invisible audition apart from the algorithmic hiring tools that have already drawn public scrutiny is precisely its reach. Most of the controversy of recent years has fixed on the resume-screening systems that parse a CV and rank applicants by keyword, qualification and inferred experience. Those tools are crude and frequently biased, but they at least confine themselves to information a candidate chose to submit. The affective layer operates somewhere else entirely. It does not read what you wrote; it reads what your body did while you spoke, the involuntary territory of expression and voice that no one consents to surrender and that most candidates do not even know is being recorded as data rather than mere footage. It is one thing for a machine to judge the words on your page. It is another for it to judge the flicker at the edge of your eye, and to do so silently, without ever telling you the trial is under way.
The Trait That Cannot Be Measured
Begin with the central claim, because everything else collapses without it. The premise of emotion-reading hiring software is that a person's interior state, their honesty, their enthusiasm, their emotional steadiness, their fit for a culture, can be inferred from the outward arrangement of their face and the acoustics of their speech. Smile in the right places and the algorithm reads warmth. Hold steady eye contact and it reads confidence. Let your voice wobble and it reads anxiety, or evasion, or instability. The whole edifice rests on the assumption that facial movements map reliably onto emotional and psychological states, the same way a thermometer maps onto temperature.
The trouble is that the assumption is wrong, and the people who know the science best have said so in the clearest possible terms. In 2019, five of the most prominent researchers in the field, led by the psychologist and neuroscientist Lisa Feldman Barrett and including Ralph Adolphs, Stacy Marsella, Aleix Martinez and Seth Pollak, published a sweeping review in the journal Psychological Science in the Public Interest under the title “Emotional Expressions Reconsidered.” They examined more than a thousand studies and reached a conclusion that ought to have detonated the entire premise of affective hiring. There are, they wrote, no objective measures, whether taken singly or as a pattern, that reliably, uniquely and replicably identify emotional categories from facial movements. A scowl does not mean anger. A smile does not mean happiness. People scowl when they are concentrating, smile when they are uncomfortable, sit stony-faced through grief and weep at weddings. The relationship between the face and the feeling is loose, contextual, cultural and individual, and it cannot be read off the surface like a barcode.
Much of the technology leans, knowingly or not, on what is called Basic Emotion Theory, the older idea that a small set of discrete emotions are expressed in fixed, universal ways across all human faces. The Barrett review is, in effect, a systematic dismantling of that idea using the field's own data. And if facial movements do not reliably reveal even the six so-called basic emotions, the proposition that they reveal abstractions as slippery as honesty or cultural fit belongs not to science but to its discredited ancestors. Critics of the field have repeatedly reached for the same comparison, and it is the right one. To infer character and competence from the configuration of a face is to revive physiognomy and phrenology, the nineteenth-century pseudosciences that claimed to read morality from skull shape and criminality from the set of a jaw. The technology is new. The fallacy is two hundred years old, and it was wrong then for the same reason it is wrong now. There is no validated, replicated relationship between the micro-expressions and vocal features these systems claim to measure and the actual performance of a person in an actual job. The vendors are selling a measurement of a thing that, as measured, does not exist.
How the Industry Learned to Hide
The most instructive episode in this whole saga belongs to HireVue, for a long time the most visible name in algorithmic video interviewing and, for that reason, the company that drew the first sustained fire. In its earlier incarnation, HireVue's platform analysed candidates' faces during recorded interviews, and the company acknowledged that somewhere between ten and thirty per cent of a candidate's assessment could derive from facial expression, with the remainder drawn from language. It marketed the ability to gauge cognitive ability, emotional intelligence, psychological traits and social aptitude from a recorded answer.
In November 2019, the Electronic Privacy Information Center, a Washington watchdog known as EPIC, filed a complaint with the United States Federal Trade Commission, alleging that HireVue's practices were unfair and deceptive. The complaint argued that the tools were unproven, invasive and prone to bias, and that representing them as objective when their scientific foundation was so shaky amounted to a deception of both candidates and the employers who bought the software. A little over a year later, in early 2021, HireVue announced that it would stop using facial analysis in its assessments. The company framed the retreat partly as a response to public concern, conceding that the visual component was not worth the unease it generated. By HireVue's own account the visual analysis had already been stripped out the previous March, in 2020, and the company maintained that advances in its language analysis meant the facial component “no longer significantly added value”, a revealing formulation: on the vendor's own telling the work had not been given up but simply absorbed by another modality.
It would be comforting to read that as the moment the industry came to its senses. It was not. HireVue did not abandon the underlying project of inferring traits from a candidate's body; it narrowed it. The platform continued to analyse speech, intonation and language, which carry many of the same risks of bias and unvalidated inference as the facial component it dropped, only with less of the visceral alarm that the word “facial recognition” provokes. And the broader market did not contract. It diversified and consolidated. A cohort of vendors now offer one-way video interviews layered with personality and culture-fit analysis. The platform myInterview, which blended recorded video answers with personality and cultural assessment, was acquired by the recruitment marketing firm Radancy in September 2025, one of several deals that folded these capabilities into ever larger hiring stacks. The lesson the industry absorbed from the HireVue affair was not that reading character off a body was unsound. It was that the reading should be done more quietly, in modalities that attract less scrutiny, sold to employers who will rarely think to ask what is happening beneath the interface.
The migration from face to voice deserves particular attention, because it is so often mistaken for a reform. Vocal tone, pitch, pace and prosody are no more validated as windows onto honesty or stability than facial micro-expressions are, and they import their own demographic landmines: the pitch of a voice tracks with gender, its rhythm with regional and national origin, its pauses and fillers with neurology and with the simple cognitive load of speaking a second language. To drop the camera's verdict and keep the microphone's is not to fix the problem. It is to move it somewhere harder to see.
The Audit Behind the Screen
For years the response to all of this from the companies involved was a version of: you cannot prove our systems are biased, because you cannot see inside them. The models are proprietary, the training data is confidential, and an outsider has no way to run the controlled experiment that would isolate the effect of a candidate's race or gender. That defence has now started to fail, because researchers have worked out how to probe the black box from the outside.
In a paper titled “Behind the Screens: Uncovering Bias in AI-Driven Video Interview Assessments Using Counterfactuals,” posted to the arXiv preprint server in May 2025 and revised that November, the researchers Dena F. Mujtaba and Nihar R. Mahapatra of Michigan State University built a method to do precisely that. Their approach uses generative adversarial networks, a class of AI that can synthesise realistic media, to construct counterfactual versions of the same candidate. Take a recorded interview and produce alternate renderings in which a protected characteristic, the apparent race or gender of the speaker, is altered while everything else, the words, the content, the substance of the answer, is held constant. Then feed each version into a personality-prediction model and watch whether the score moves. If the only thing that changed was the candidate's apparent demographic, and the score changes too, the bias is not hypothetical. It is sitting in the output.
The elegance of the counterfactual approach is that it sidesteps the usual stalemate. A vendor can always argue that any disparity in real-world outcomes reflects real differences between applicant pools rather than the model's prejudice. The counterfactual closes that escape: the two versions of the candidate are the same person giving the same answer, differing only in an attribute that ought to be irrelevant to the score. The framework is also deliberately multimodal, examining visual, audio and textual features together rather than in isolation, which matters because real systems combine all three and bias can hide in the interaction between them. Applied to a state-of-the-art model that predicts the so-called Big Five personality traits, the openness, conscientiousness, extraversion, agreeableness and neuroticism that much of the personality-assessment industry treats as gospel, the method revealed significant disparities across demographic groups. The same answer, the same competence, the same content, scored differently depending on who appeared to be giving it. The authors were careful about what they were and were not claiming: their contribution is principally a scalable auditing tool for exactly the black-box commercial settings where the training data and model internals are locked away. But the direction of the finding is unambiguous. When you can finally test these systems for disparate treatment, you find it.
When the Machine Misreads a Face
A second line of evidence, published in 2026, addresses the layer beneath the personality scores: the emotion-recognition machinery that many of these systems use to read affect in the first place. In an article on ethics and bias in emotional AI, published in Frontiers in Artificial Intelligence in March 2026, the researchers Smrithy G. S, Balaji Chandrasekaran and Omana J set out the case that emotion-recognition systems discriminate systematically along lines of race and gender, and misread expression across cultural contexts, and they trace the mechanism rather than merely asserting the result.
Part of the problem is foundational and statistical. The facial-analysis systems on which emotion inference is built inherit the accuracy gaps that the computer scientists Joy Buolamwini and Timnit Gebru exposed in their landmark 2018 study, Gender Shades. Testing commercial gender-classification systems, Buolamwini and Gebru found error rates as high as 34.7 per cent for darker-skinned women, against a maximum of 0.8 per cent for lighter-skinned men. A system that cannot reliably tell who a person is will not reliably tell how they feel, and the errors do not distribute evenly: the Frontiers authors note that such systems have tended to misclassify the expressions of Black faces in particular, more readily reading a neutral expression as hostile or aggressive. Carry that into a hiring context and the consequence is not abstract. A candidate of colour, sitting calmly through a recorded interview, can have their composure rendered by the machine as something darker, and be marked down for an emotion they never felt.
Part of the problem is cultural. The Frontiers authors emphasise that emotions are context-dependent, culturally mediated and frequently ambiguous, and that the training data behind these systems skews heavily towards Western populations and Western norms of expression. A model calibrated on Western faces and Western display rules will misread the expressions of people outside that mould, and reinforce stereotypes about how different genders are supposed to emote in the bargain. There is a socioeconomic dimension too, which the paper only brushes against but which follows from the same logic, subtler but no less real: the way a person presents, the polish of their diction, the backdrop visible behind them, the quality of their webcam and their broadband, all carry the fingerprints of class, and all feed into a system that was never designed to disentangle privilege from merit. The opacity of the process, the authors argue, is precisely what makes it dangerous in high-stakes settings such as recruitment, where the emotional readings are treated as objective indicators of a candidate's suitability when they are nothing of the kind. The result is a discrimination engine wearing the costume of a meritocracy.
The People Who Cannot Pass
Bias along the familiar lines of race and gender is grave enough. But emotion-reading hiring tools also inflict a quieter and in some ways more total exclusion, on candidates whose faces and voices simply do not produce the signals the model is looking for. These systems do not measure whether you can do the job. They measure whether you perform, in front of a webcam, in the narrow manner the model was trained to reward. And for whole categories of people, that performance is not available at any price.
Consider autistic candidates and those with attention deficit conditions. The behaviours these systems treat as red flags, reduced eye contact, a flatter or less animated vocal range, pauses in the middle of a sentence, an expressive style that does not match the neurotypical template, are in many cases direct features of neurodivergence rather than evidence of disengagement or dishonesty. An autistic applicant who looks slightly away from the lens to concentrate, or who answers in a measured monotone, can be flagged by the system as low in engagement or confidence, and screened out before a single human being assesses whether they can actually perform the role. Disability-rights researchers and advocates have warned repeatedly that video and audio screening which scores eye contact, vocal cadence and facial affect risks penalising autistic, blind and otherwise disabled candidates for traits that have nothing to do with competence. The cruelty is compounded by the fact that many such candidates are highly capable in the roles they are applying for; the system filters not for ability but for a narrow performance of normalcy that ability does not require.
The same trap closes on non-native speakers, whose accents, prosody and pacing diverge from the speech patterns the model learned, and on anyone whose self-presentation is shaped by culture, class or simple nerves in ways the training data did not anticipate. The defence offered by vendors and employers, that candidates can always request an accommodation, founders on a brutal practicality. To request one, a candidate must first disclose a disability to a prospective employer, at the most vulnerable moment of the hiring process, before they have any offer, any leverage or any relationship to protect them. Many will not, and so they say nothing, and the system quietly downgrades them for being who they are. Under the Americans with Disabilities Act, pre-employment assessment is supposed to be job-related and consistent with business necessity, a standard these tools struggle to meet when the very traits they measure have no demonstrated link to job performance. The disparate impact is not a bug to be patched. It is the predictable output of asking a machine to reward one narrow way of being human.
No Reasons, No Appeal
Strip away the bias for a moment and a separate harm remains, one that would persist even if the systems were somehow perfectly fair. It is the harm of being judged by a process that owes you no explanation and offers you no way to answer back.
A human interview is many things, some of them flawed, but it is at least a relationship. The interviewer can be challenged, charmed, corrected. If they misunderstand your answer, you can clarify. If they harbour a prejudice, you can sometimes overcome it in the room. And if you are rejected, there is at least a person who knows why, a chain of reasoning that can in principle be questioned, complained about, learned from. The algorithmic audition dissolves all of this. There is no interviewer to persuade, no reasoning to interrogate, no explanation of why your face and voice produced the score they did. You are not told the traits being measured. You are not shown the result. You are not given a route of appeal. You are handed a rejection with no causal story attached, and the absence of a story is the point: it is what lets the system process thousands of candidates an hour, and it is what makes the decision impossible to contest.
This opacity is not incidental. It is the operating logic. A system that had to explain each rejection in terms a candidate could challenge would forfeit the very speed and scale that make it commercially attractive. So the explanation is dispensed with, and a verdict on your honesty and your emotional stability, derived from a pseudoscience and tilted by bias, is delivered with all the unanswerable finality of a closed door. The candidate is left to guess. Was it the answer, the pause, the accent, the face, the lighting in the spare room? The machine knows, or claims to, and it is not saying. And because the rejection is silent about its reasons, the candidate cannot even learn from it, cannot adjust, cannot improve, because there is nothing to adjust towards except a moving target they were never permitted to see.
The Employer's Comforting Illusion
It would be easy to cast the employers who deploy these tools as the villains of the piece, but the more accurate and more disquieting truth is that many of them believe they are doing the opposite of harm. The pitch that sells emotion-reading hiring software to a human-resources department is a pitch about fairness. Human interviewers, the argument runs, are riddled with bias: they favour candidates who look like them, who share their background, who went to the right schools and tell the right jokes. Replace the fallible human with a consistent algorithm, the pitch continues, and you remove the prejudice. Every candidate is scored by the same model against the same criteria. What could be fairer than that?
This is the comforting illusion, and it is precisely backwards. The algorithm does not remove bias; it launders it, taking the prejudices embedded in its training data and the flawed assumptions of its design and reissuing them as objective scores with a veneer of mathematical neutrality. A biased human interviewer is at least a discrete, identifiable, potentially correctable source of unfairness. A biased model deployed across an entire hiring funnel applies the same distortion to every candidate, at scale, invisibly, and clothes it in the authority of data. The very feature the employer prizes, the consistency, is what converts a scattering of individual prejudices into a single systematic one. And because the output arrives as a number, it carries an aura of rigour that a gut feeling never could, which makes it harder to question and easier to defend. The employer believes they have bought objectivity. They have bought the appearance of objectivity wrapped around a discriminatory core, and the appearance is worse than nothing, because it forecloses the very scrutiny that might catch the discrimination.
There is a further irony. The traits these systems claim to optimise for, honesty, stability, cultural fit, are not even shown to predict good hiring. “Cultural fit” in particular is a notoriously double-edged criterion, as likely to entrench an organisation's existing homogeneity as to improve it: a model trained to reward candidates who resemble a company's current, perhaps already skewed, workforce will simply reproduce that skew while calling it merit. The employer set out to widen the funnel and ended up narrowing it, convinced the whole time that the narrowing was fairness. There is even a legal sting in the tail: an employer who adopts one of these tools believing it neutralises bias may in fact be importing a disparate impact they would be liable for, having outsourced the discrimination to a vendor but not the responsibility for it.
The Law Catches a Glimpse
Regulators have begun, unevenly and belatedly, to register what is happening, and the most decisive response has come from the European Union. Under the EU AI Act, which began phasing into force in 2025, the use of AI systems to infer the emotions of a person in the workplace, and in educational settings, is not merely regulated but prohibited outright, with narrow exceptions for medical and safety purposes. The prohibition, set out in Article 5 and effective from February 2025, rests on an explicit recognition of the power imbalance between an employer and a job candidate or worker, a relationship in which genuine consent to emotional surveillance is something close to a fiction. The penalties are not symbolic: breaches of the prohibited-practices provisions can attract fines of up to thirty-five million euros or seven per cent of a company's global annual turnover, whichever is higher. An AI tool that purports to score a candidate's enthusiasm or confidence or cultural fit from their face or voice is, in the workplaces of the European Union, now illegal.
Elsewhere the picture is patchier. In the United States, Illinois enacted its Artificial Intelligence Video Interview Act, which requires employers to notify applicants when AI is used to analyse a video interview, to explain in general terms how the technology works, and to obtain consent before proceeding, with further obligations on AI in employment taking effect at the start of 2026. New York City's Local Law 144 takes a different tack, requiring annual independent bias audits of automated employment decision tools, publication of the results, and advance notice to candidates. These are real interventions, but they are also limited: notice and consent do little for a candidate who has no realistic power to refuse, and a bias audit catches only some of the harms and none of the underlying invalidity. In the United Kingdom, the Information Commissioner's Office published a report and draft guidance on automated decision-making in recruitment on 31 March 2026, drawing on evidence gathered from more than thirty employers and calling on them to review their AI hiring practices. Its central finding sharpens everything said here about opacity: many employers do not realise they are using automated decision-making at all, and so never put the safeguards the law requires anywhere near it. The regulator's scepticism about affective tools specifically is older and blunter. In October 2022 the ICO's then deputy commissioner, Stephen Bonner, warned organisations off biometric “emotional analysis” altogether, saying the office was “yet to see any emotion AI technology develop in a way that satisfies data protection requirements”, and that such technologies “may not work yet, or indeed ever.” Britain's broader legal scaffolding shifted, too, with the Data (Use and Access) Act 2025 replacing the old Article 22 prohibition on solely automated decisions with a framework built around a right to challenge and mandatory safeguards.
The transatlantic divergence is stark. On one side of the Atlantic, emotion recognition in hiring is banned as an unacceptable risk to fundamental rights. On the other, it is a disclosure-and-audit problem, permitted so long as the right boxes are ticked. Neither approach has yet caught up with the speed of deployment, and a candidate's protection now depends heavily on the accident of which jurisdiction they happen to be applying from. A multinational running a single hiring pipeline across borders faces the awkward reality that the same tool can be outright illegal in Frankfurt and merely disclosable in Chicago, which tells you less about the technology than about how unsettled the world's collective judgement of it still is.
What Is Actually Being Decided
Return, at the last, to the room with the laptop and the green light, and ask what is really at stake when an algorithm decides how your face makes you feel about telling the truth. The most immediate answer is a job, and jobs are not small things; they are housing, healthcare, dignity, the difference between a life with options and a life without. To have that gated by a system that cannot do what it claims, that misreads faces by race and voices by accent and stillness by neurology, and that will never tell you why, is a concrete injustice visited on real people one rejection at a time.
But there is a deeper thing being decided, and it concerns the kind of judgement we are prepared to accept. The promise of the invisible audition is that character can be quantified, that honesty has a facial signature and emotional stability a vocal one, and that a sufficiently sophisticated model can read these off a recording and rank human beings accordingly. The science says it cannot. The audits say that where the model does produce a number, the number is warped by who you appear to be. And the structure of the process ensures that you will never be allowed to argue. What is being normalised is not a better way to hire but a worse way to be judged: opaque, unaccountable, dressed as objectivity, and aimed first at the people with the least power to resist it.
The candidates being judged deserve, at minimum, to know when a machine is reading their face, to understand what it claims to measure, and to contest a verdict that shapes their livelihood. The employers doing the judging deserve to understand that the objectivity they have been sold is a costume, and that the consistency they prize is consistency in error. And the rest of us deserve a serious public reckoning with a simple question that the technology has so far been allowed to skip. If a thing cannot be measured, no amount of computation will measure it, and the only honest verdict a face-reading hiring algorithm can return on a candidate's honesty is that it does not, and cannot, know. The green light beside the webcam suggests otherwise. It is the most confident liar in the room.
References
- Mujtaba, Dena F., and Nihar R. Mahapatra. “Behind the Screens: Uncovering Bias in AI-Driven Video Interview Assessments Using Counterfactuals.” arXiv preprint arXiv:2505.12114, submitted May 2025, revised November 2025. https://arxiv.org/abs/2505.12114
- Smrithy, G. S., Balaji Chandrasekaran, and Omana J. “Ethics and bias in emotional AI.” Frontiers in Artificial Intelligence, 5 March 2026. https://www.frontiersin.org/journals/artificial-intelligence/articles/10.3389/frai.2026.1768696/full
- Barrett, Lisa Feldman, Ralph Adolphs, Stacy Marsella, Aleix M. Martinez, and Seth D. Pollak. “Emotional Expressions Reconsidered: Challenges to Inferring Emotion From Human Facial Movements.” Psychological Science in the Public Interest, 2019. https://journals.sagepub.com/doi/10.1177/1529100619832930
- Buolamwini, Joy, and Timnit Gebru. “Gender Shades: Intersectional Accuracy Disparities in Commercial Gender Classification.” Proceedings of Machine Learning Research, vol. 81, 2018. https://proceedings.mlr.press/v81/buolamwini18a.html
- Electronic Privacy Information Center. “HireVue, Facing FTC Complaint From EPIC, Halts Use of Facial Recognition.” EPIC, January 2021. https://epic.org/hirevue-facing-ftc-complaint-from-epic-halts-use-of-facial-recognition/
- Electronic Privacy Information Center. “In re HireVue.” EPIC, 2019. https://epic.org/documents/in-re-hirevue/
- European Union. “Article 5: Prohibited AI Practices.” EU Artificial Intelligence Act. https://artificialintelligenceact.eu/article/5/
- Future of Privacy Forum. “Red Lines under EU AI Act: Unpacking the prohibition of emotion recognition in the workplace and education institutions.” FPF Blog. https://fpf.org/blog/red-lines-under-eu-ai-act-unpacking-the-prohibition-of-emotion-recognition-in-the-workplace-and-education-institutions/
- Hinshaw & Culbertson LLP. “Illinois Adopts New AI-in-Employment Regulations: What Employers Need to Know for 2026.” Employment Law Observer. https://www.hinshawlaw.com/en/insights/blogs/employment-law-observer/illinois-adopts-new-ai-in-employment-regulations-what-employers-need-to-know-for-2026.html
- Warden AI. “NYC Local Law 144 Compliance Guide 2026.” https://www.warden-ai.com/resources/hr-tech-compliance-nyc-local-law-144
- DLA Piper Privacy Matters. “UK: ICO Report on Automated Decision-Making in Recruitment.” April 2026. https://privacymatters.dlapiper.com/2026/04/uk-ico-report-on-automated-decision-making-in-recruitment/
- Information Commissioner's Office. “'Immature biometric technologies could be discriminating against people' says ICO in warning to organisations.” ICO, 26 October 2022. https://ico.org.uk/about-the-ico/media-centre/news-and-blogs/2022/10/immature-biometric-technologies-could-be-discriminating-against-people-says-ico-in-warning-to-organisations/
- Bloomberg Law. “AI Hiring Tools Elevate Bias Danger for Autistic Job Applicants.” Daily Labor Report. https://news.bloomberglaw.com/daily-labor-report/ai-hiring-tools-elevate-bias-danger-for-autistic-job-applicants
- The Conversation. “Tech companies claim AI can recognise human emotions. But the science doesn't stack up.” https://theconversation.com/tech-companies-claim-ai-can-recognise-human-emotions-but-the-science-doesnt-stack-up-243591
- MIT Technology Review. “Emotion AI researchers say overblown claims give their work a bad name.” 14 February 2020. https://www.technologyreview.com/2020/02/14/844765/ai-emotion-recognition-affective-computing-hirevue-regulation-ethics/
- Radancy. “Radancy Extends AI Platform, Powering Faster, Smarter, More Cost-Efficient Hiring at Scale.” 9 September 2025. https://www.radancy.com/en/radancy-extends-ai-platform-powering-faster-smarter-more-cost-efficient-hiring-at-scale-myinterview/
- openPR. “AI Video Interview Market Size Accelerated by 14.8% CAGR.” https://www.openpr.com/news/4508316/ai-video-interview-market-size-accelerated-by-14-8-cagr-by-key
Tim Green
UK-based Systems Theorist & Independent Technology Writer
Tim explores the intersections of artificial intelligence, decentralised cognition, and posthuman ethics. His work, published at smarterarticles.co.uk, challenges dominant narratives of technological progress while proposing interdisciplinary frameworks for collective intelligence and digital stewardship.
His writing has been featured on Ground News and shared by independent researchers across both academic and technological communities.
ORCID: 0009-0002-0156-9795
Email: tim@smarterarticles.co.uk
Listen to the free weekly SmarterArticles Podcast
