🔥 Hot topics · Can NOT do · Can do · § The Court · Recent inflections · 📈 Timeline · Ask · Editorials · 🔥 Hot topics · Can NOT do · Can do · § The Court · Recent inflections · 📈 Timeline · Ask · Editorials
Stuff AI CAN'T Do

Can AI recognise emotions in faces at coarse-grained level ?

What do you think?

What does it mean for AI to 'recognise emotions in faces at a coarse-grained level'? Broadly, it refers to identifying overt emotional states like happiness, sadness, or anger from facial expressions, rather than detecting subtle or fleeting microexpressions. While high-resolution video calls make this task achievable with reasonable accuracy, finer emotional cues remain challenging. How do today's systems pull this off—and what still limits their performance?

Background

AI systems can distinguish coarse-grained emotional categories (e.g., happy, sad, angry) with reasonable accuracy using deep learning models—primarily convolutional neural networks—trained on large facial-image datasets (IEEE, enriched May 9, 2026). These models learn facial feature patterns associated with broad emotional states. Performance improves as datasets grow in size and diversity, increasing generalizability. In contrast, subtle microexpressions—rapid, low-intensity facial movements—remain difficult to classify reliably, especially at lower video-call resolutions.

Status last checked on August 10, 2026.

📰

Gallery

In the Court of AI Capability
Summary of Findings
Verdict over time
May 2026May 2026May 2026May 2026May 2026Jun 2026Jun 2026Jun 2026Jun 2026Jun 2026Jun 2026Jul 2026Jul 2026Jul 2026Jul 2026Jul 2026Jul 2026Aug 2026Aug 2026
Sitting at the Bench Filed · Aug 10, 2026
— The Question Before the Court —

Can AI recognise emotions in faces at coarse-grained level?

★ The Court Finds ★
Reaffirmed
Yes

The jury found a clear answer in the affirmative.

Ruling of the Bench

After thorough deliberation, the jury unanimously agreed that modern AI can indeed recognise emotions at a coarse-grained level—happy, sad, and neutral—with remarkable reliability, thanks to advances in deep learning models trained on vast datasets. They found the evidence persuasive and the performance consistent across multiple systems, leaving no doubt about AI’s current capability in this domain. The bench declares: “AI reads the room—though it still waits for an invite to the party.”

— Hon. M. Lovelace, Presiding
Jury Tally
2Yes
0Almost
0No
Verdict Confidence
93%
The Court of AI Capability is, of course, not a real court.
But the data is real.
The Case File · Stacked History
Session I · May 2026 Yes
Session II · May 2026 Yes
Session III · May 2026 Yes · 83%
Session IV · May 2026 Yes · 79%
Session V · May 2026 Yes · 85%
Session VI · Jun 2026 Yes · 81%
Session VII · Jun 2026 Yes · 83%
Session VIII · Jun 2026 Yes · 91%
Session IX · Jun 2026 Yes · 99%
Session X · Jun 2026 Yes · 92%
Session XI · Jun 2026 Yes · 88%
Session XII · Jul 2026 Yes · 98%
Session XIII · Jul 2026 Yes · 94%
Session XIV · Jul 2026 Yes · 98%
Session XV · Jul 2026 Yes · 90%
Session XVI · Jul 2026 Yes · 90%
Session XVII · Jul 2026 Yes · 90%
Session XVIII · Aug 2026 Yes · 90%
Case № D42B · Session XIX
In the Court of AI Capability

The Case File

Docket № D42B · Session XIX · Vol. XIX
I. Particulars of the Case
Question put to the courtCan AI recognise emotions in faces at coarse-grained level?
SessionXIX (19 hearing)
Convened10 Aug 2026
Previously ruledYES (May '26) → YES (May '26) → YES (May '26) → YES (May '26) → YES (May '26) → YES (Jun '26) → YES (Jun '26) → YES (Jun '26) → YES (Jun '26) → YES (Jun '26) → YES (Jun '26) → YES (Jul '26) → YES (Jul '26) → YES (Jul '26) → YES (Jul '26) → YES (Jul '26) → YES (Jul '26) → YES (Aug '26) → YES (Aug '26)
Presiding JudgeHon. M. Lovelace
II. Cumulative Tally Across Sessions

Across 19 sessions, 42 jurors have heard this case. Combined tally: 40 YES · 2 ALMOST · 0 NO · 0 IN RESEARCH.

Note: cumulative includes older juror opinions. The current session tally above is the live verdict.

III. Verdict

By a vote of 2 — 0 — 0, the panel returns a verdict of YES, with verdict confidence of 93%. The court so orders.

IV. Statements from the Bench
Juror I YES

"Deep learning models achieve high accuracy"

Juror II YES

"Multiple models (e.g., FER systems like ResMaskNet, AffectNet-trained CNNs) achieve coarse emotion recognition (e.g., happy/sad/neutral) reliably."

M. Lovelace
Presiding Judge
M. Lovelace
Clerk of the Court

What the audience thinks

No 3% · Yes 89% · Maybe 8% 176 votes
Yes · 89%
Trend needs votes from at least 2 different days.

Discussion

no comments

Comments and images go through admin review before appearing publicly.

19 jury checks · most recent 2 days ago
10 Aug 2026 2 jurors · can, can can
05 Aug 2026 1 juror · can can
30 Jul 2026 1 juror · can can
25 Jul 2026 1 juror · can can
20 Jul 2026 1 juror · can can
14 Jul 2026 1 juror · can can
09 Jul 2026 2 jurors · can, can can
03 Jul 2026 1 juror · can can
28 Jun 2026 2 jurors · can, can can
22 Jun 2026 3 jurors · can, can, can can
17 Jun 2026 1 juror · can can
12 Jun 2026 4 jurors · can, can, can, can can
06 Jun 2026 3 jurors · can, can, can can
01 Jun 2026 4 jurors · can, undecided, can, can undecided
26 May 2026 4 jurors · can, can, can, can can
21 May 2026 3 jurors · can, undecided, can undecided
16 May 2026 3 jurors · can, can, can can
13 May 2026 3 jurors · can, can, can can
11 May 2026 2 jurors · can, can can

Each row is a separate jury check. Jurors are AI models (identities kept neutral on purpose). Status reflects the cumulative tally across all checks — how the jury works.

More in Sensory

Got one we missed?

Add a statement to the atlas. We review weekly.