Artificial
intelligence can now detect subtle shifts in vocal tone, facial
microexpressions, and written language that correlate with human emotional
states, and this capability is being deployed at scale in customer service,
healthcare, recruitment, and education. The technology is advancing faster
than public awareness, faster than regulatory frameworks, and in some cases
faster than the science underpinning it can justify.

Emotional AI, sometimes called affective computing, has been
developing for decades as an academic discipline. What has changed is the
commercial viability of the underlying technology. Advances in computer
vision, natural language processing, and audio analysis have made it possible
to build systems that infer emotional states from widely available data
streams, including video calls, customer support recordings, and job
interviews, at a cost that makes mass deployment practical for large
organisations.

How Emotional AI Works

Current emotional AI systems typically combine multiple
modalities. Facial action coding systems analyse movement in dozens of facial
muscle groups to classify emotional expressions. Voice analysis tools examine
pitch, rhythm, and prosody for indicators of stress, confidence, or
deception. Sentiment analysis models parse text for emotional content using
large language model architectures trained on labelled datasets. The
combination creates a picture of inferred emotional state that vendors
present as actionable intelligence about individuals.

The accuracy of these systems is a matter of genuine scientific
dispute. Research published in Psychological Science has challenged the
foundational assumption that discrete emotional states map reliably onto
facial expressions, a critique that goes to the heart of validity claims made
by emotional AI vendors. Human emotional expression varies substantially
across cultures, individuals, and contexts in ways that universal models do
not adequately capture. The gap between what these systems measure and what
they claim to measure is significant and commercially inconvenient to
acknowledge.

Despite these limitations, the commercial market for emotional AI
has expanded rapidly. Companies including Affectiva, Cogito, and Realeyes
provide emotional analytics to major corporations, and the technology is
embedded in products used by millions of people without their explicit
awareness. The Federal Trade
Commission in the United States has raised concerns about the
marketing claims of some emotional AI vendors, noting the gap between advertising
and scientific evidence.

Workplace and Recruitment Applications

The use of emotional AI in hiring has attracted significant
controversy. Video interview platforms that analyse candidate facial
expressions and vocal patterns to assess personality traits have been adopted
by thousands of employers. Unilever was a high-profile early adopter of
AI-driven video interviews; subsequent scrutiny raised questions about the
validity and fairness of the underlying assessments, and the company
eventually scaled back its use of the technology.

Researchers at the University of Cambridge and elsewhere have
demonstrated that AI recruitment tools can exhibit systematic biases related
to protected characteristics, not through intentional discrimination but
through correlations embedded in training data. An algorithm trained on video
interviews of successful employees at a historically homogeneous organisation
will learn to replicate the demographic profile of that organisation’s
existing workforce. The Illinois Artificial Intelligence Video Interview Act,
passed in 2019, requires employers using AI to analyse video interviews to
notify candidates and explain how the technology works, a model that other
jurisdictions are following. The EU’s AI Act classifies certain AI systems
used in employment as high-risk, requiring conformity assessments and
transparency obligations that are now being implemented across European
businesses.

Healthcare and Mental Health Applications

In healthcare, emotional AI shows genuine promise alongside
genuine risk. Systems that detect early signs of depression, anxiety, or
cognitive decline from speech patterns and behavioural data could enable
earlier intervention and better outcomes. Research programmes at MIT,
Stanford, and several NHS trusts are investigating these applications with
appropriate clinical governance frameworks. Preliminary findings from the NHS
Digital Mental Health programme suggest that AI-assisted screening could
identify at-risk individuals weeks before they would otherwise come to
clinical attention.

The challenge is that the same technology deployed without
clinical oversight becomes a liability. Mental health apps that use AI to
infer emotional states from user behaviour and respond with content
recommendations are operating in a regulatory grey area in most
jurisdictions. The potential for misclassification, or for commercially
motivated emotional manipulation, is real and largely unaddressed by current
frameworks. The distinction between a clinical tool subject to regulatory
scrutiny and a consumer app operating outside it is a line that several
commercial products are deliberately blurring.

What This Means for You

The most important practical implication of emotional AI is that
you are almost certainly being assessed by systems you are unaware of. If you
have participated in a video job interview, called a customer service line,
or used a mental health or wellness app, your emotional signals may have been
processed by AI without meaningful disclosure. Understanding your rights
matters here.

The education sector represents another significant
deployment context for emotional AI that receives insufficient attention.
Learning management platforms are beginning to incorporate engagement
monitoring tools that analyse student facial expressions and attention
patterns during online learning sessions. The use of these tools in
educational contexts raises particular concerns because of the power
differential between educational institutions and students, and because young
people may not have the agency to meaningfully consent to or resist this
monitoring. Several school districts in the United States have adopted or
piloted AI emotion monitoring tools from vendors whose scientific validity
claims have not been independently validated, creating risks that regulators
have only recently begun to examine systematically.

In the EU, the General Data Protection Regulation provides
protections for biometric and special category data, and the AI Act adds
requirements for high-risk systems. In the UK, post-Brexit data protection
law follows a similar framework. In the United States, protections are
patchier and vary by state.

The healthcare applications of emotional
AI deserve particular attention because the power asymmetry between patients
and providers is so significant. Patients in medical contexts are in a state
of heightened vulnerability, and the deployment of emotion-monitoring AI
without their informed consent raises specific ethical concerns that general
data protection frameworks do not fully address. The World Health Organization
has published guidance on AI in healthcare settings that addresses emotional
data specifically, calling for explicit consent requirements and independent
validation of clinical claims before deployment. Adoption of this guidance
remains voluntary and inconsistent across health systems.

The broader question is whether the commodification of emotional
data is a direction society wants to travel. The asymmetry of power and
information between the organisations deploying emotional AI and the
individuals being monitored is not a technical inevitability. It is a policy
choice, and it can be reversed by policy choices that require meaningful
consent, impose accuracy standards, and create liability for harm. The
organisations making these deployment decisions today are shaping norms that
will be difficult to reverse once they become embedded in workplace and
commercial infrastructure. The ability to infer emotional states at scale
creates asymmetric power relationships between organisations deploying these
systems and individuals whose inner lives are being monitored and monetised.
For related analysis of AI’s impact on personal data and privacy, see our
coverage of AI
in insurance pricing and who
AI systems are built to serve.

About the Author

Stuart Kerr is a technology correspondent at LiveAIWire, covering
artificial intelligence, digital innovation, and the social impact of
emerging technologies. Follow LiveAIWire for daily analysis at liveaiwire.com.