Research behind the Reel · TT-008

The fish that spat its answer

Can an archerfish use its visual abilities to distinguish human face images?

Discrimination of human faces by archerfish (Toxotes chatareus)
Newport, Wallis, Reshitnyk & Siebeck (2016), Scientific Reports 6:27523

Why it matters

It reveals flexible visual learning while showing how carefully animal cognition claims must follow the task actually tested.

What the study found

Two separate four-fish experiments found learned discrimination of a familiar unrewarded human face image from novel alternatives. The second cohort succeeded when colour, mean brightness and outer head shape were standardized. These trained choices do not demonstrate recognition of owners.

What this does not establish

Only two faces were shown per trial. The 44 and 18 counts describe pools of novel images. Biological replication was four fish per experiment, with repeated trials and a restricted set of still frontal face images.

  • Two separate cohorts of four fish; repeated trials are not additional animals or independent laboratory replications.
  • The testing rule was avoid a familiar unrewarded face and choose the other image, not choose an owner from a simultaneous array of 44 people.
  • Food-reward learning after prior monitor-spitting training; spontaneous social recognition was not studied.
  • Only still frontal views from a restricted set of rendered female human faces; owner recognition and emotions are outside this study.
  • Colour, brightness and head outline were controlled, but the exact remaining visual cues and neural mechanism were not identified.
  • The living-fish reference photograph is from a separate 2018 study; no actual footage from the 2016 face experiment is used.
  • The source contains training-session and model-link reporting inconsistencies documented in evidence; neither is used in narration.

Claims you can trace

Each factual claim is tied to its location in the original research. Background evidence is listed separately.

Research that supplies the context

Visual sources and credits

The credits below identify the source and use of each visual. Reference imagery and explanatory illustrations are distinguished from experimental footage where present. Narration is synthetic.

Complete verification record
# TT-008 — The fish that spat its answer

Research and script author: today_research. Checked 13 November 2026. Planned Friday Reel: same video on Instagram and YouTube Shorts. This document is source evidence, not an independent editorial or export approval.

## Original focal paper and access

Newport, C., Wallis, G., Reshitnyk, Y. & Siebeck, U. E. (2016). **Discrimination of human faces by archerfish (Toxotes chatareus).** Scientific Reports 6:27523. Published 7 June 2016. DOI https://doi.org/10.1038/srep27523 . PMID 27272551; PMCID PMC4895153.

Full original article: https://pmc.ncbi.nlm.nih.gov/articles/PMC4895153/ ; publisher https://www.nature.com/articles/srep27523 ; original PDF https://www.nature.com/articles/srep27523.pdf . The complete main article was read through the PMC original full text: Introduction; both Results experiments; all three figure captions; Discussion; Subjects, Stimuli, General procedure, both Training/testing sections and Statistical analyses. Initial direct publisher requests were blocked by a cookie redirect; indexed publisher text corroborated the PMC article, which subsequently opened completely. No secondary news story supplies a narration fact.

The PMC Copyright and License information explicitly states CC BY 4.0, including article images/third-party material unless excluded in a credit line. We are not reusing the human-face stimuli or the paper's figure layouts. The animations use new schematic faces and a newly drawn aquarium; they are not the original stimuli.

## Question and design

Could a fish without a mammalian neocortex learn fine distinctions between human face images? A two-alternative forced-choice task used **Toxotes chatareus**, an archerfish able to hit targets above water with a jet. An LCD screen lay above and parallel to the water. Each correct spit earned one food pellet; an incorrect choice ended the trial. Fish already had some experience in spitting at monitor stimuli; prior experience varied.

**Experiment 1: four fish.** They first learned one rewarded face (CS+) and one unrewarded face (CS−). Two pairs of fish had the opposite trained face as CS−. Training sessions had 21–31 trials, with criterion at least 71% correct for two consecutive sessions. An intermediate step paired CS− with eight new rewarded face images. Testing then paired the **familiar unrewarded face with one new face at a time**, sampled from a pool of **44**. Ten original-pair trials and four intermediate faces served as reminders within each block and were excluded; each analysed block therefore contained 44 novel-face trials. Fish 1 and 4 completed four blocks, fish 2 and 3 two blocks. Repeated trials are not independent animals.

**Experiment 2: four different fish.** No intermediate training stage; different unrewarded faces assigned to different fish. Sessions had 30 trials; training criterion at least 75% correct twice consecutively. Test pool **18** new faces, again only two faces per trial. An original-pair reminder was excluded from each analysed block of 18 trials. All four fish completed four test blocks. An identical oval mask removed outside head-shape variation, all pictures were greyscale, and mean pixel brightness was normalized to 128. This controls those three candidate cues; it does not identify which remaining features the fish actually used.

Stimulus set: 62 frontal views of rendered three-dimensional human head models, all described in the Methods as Caucasian female faces, without hair or clothing. This restricted image set does not represent dynamic human encounters, all face populations, emotion recognition or recognising owners.

## Results and replication units

Both four-fish experiments supported above-chance discrimination of the learned negative face from new faces. Chance for this two-choice task is 50%. Experiment 1 GLMM Z=5.949, P<0.001. Experiment 2 GLMM Z=6.794, P<0.001. Fish ID and block were included as random factors. These are separate cohorts within one paper, not independent laboratory replications.

Reported test-block correct-choice means (percent ± SD):

| Block | Experiment 1 | Experiment 2 |
| --- | --- | --- |
| 1 | 64.75 ± 8.342, four fish | 73.50 ± 12.77, four fish |
| 2 | 73.00 ± 11.40, four fish | 69.50 ± 14.62, four fish |
| 3 | 66.50 ± 2.121, two fish | 77.50 ± 6.351, four fish |
| 4 | 81.50 ± 6.364, two fish | 86.25 ± 5.50, four fish |

The Reel intentionally uses no accuracy headline or animated accuracy trajectory. “81%” would obscure the reduced two-fish sample for that final Experiment 1 block. A rising training chart would also imply observations absent from the proposed diagram. The only explicit counts in narration are the original sample and stimulus-pool sizes.

## Narration claim audit

Locations refer to the original article's section headings, which are stable across HTML/PDF. PMC line references below describe the version read on the checked date and are supplementary navigation aids.

| Segment | Claim / narrative function | Original support and boundary |
| --- | --- | --- |
| 1 | Archerfish hit targets with jets; researchers used this for a face test. | Introduction final paragraph; General procedure; Results Experiment 1. PMC lines 114, 124, 174–175. Natural-history ability is context; no exact range/velocity is added. |
| 2 | Two faces on an overhead screen, choice by spitting. | Results Experiment 1 and General procedure, Fig.1C; lines 124 and 174–175. Only two stimuli are shown, monitor parallel to surface. |
| 3 | Four fish initially learned rewarded and unrewarded faces. | Results Experiment 1; Methods Training/testing Experiment 1; lines 124–125 and 177–178. No punishment is implied; the other face simply gives no food. |
| 4 | The rewarded face was exchanged for novel faces from a pool of 44; the learned face to avoid stays. | Results Experiment 1 paragraphs 2–3; Methods same section; lines 133–136 and 179–181. Intermediate eight-face training is documented above but omitted from this short narration. The pool is not shown simultaneously. |
| 5 | Correct discrimination above chance; only two faces each trial. | GLMM result and Fig.2 caption; General procedure; lines 131–136, 174. Qualitative aggregate result, no invented success rate. |
| 6 | Separate four-fish cohort, 18 new faces, colour/brightness/head-outline controls. | Results Experiment 2; Methods Stimuli and Training/testing Experiment 2; lines 138–140, 172, 183–185. “Controlled” means standardized/masked, not that the precise recognition mechanism was identified. |
| 7 | Controlled-cohort fish still distinguished learned face from other images. | Experiment 2 result Z=6.794, P<0.001; Discussion first two paragraphs; lines 140, 149–152. Does not assert every trial or every fish in the population succeeds. |
| 8 | Small lab samples, still frontal images, no owner-recognition test. | N=4 in each experiment (Results); restricted frontal images (Methods Stimuli, Discussion final paragraphs); lines 136, 162–170. Owner-recognition limitation is the direct scope implication of these methods, not a new empirical result. |
| 9 | Fine learned discrimination illustrates flexible animal vision. | Discussion opening and paragraph on applying abilities to unfamiliar stimuli, lines 149–152, 159–162. Ending is a bounded interpretation, not a claim about conscious recognition or human-like processing. |

Short exact excerpts, CC BY 4.0 original paper, to make the central reading auditable:

> “The task was for fish to continue to avoid the trained distractor (CS−) and select the novel stimuli (N+).” — Results, Experiment 1.

> “we tested whether four new archerfish could be trained to discriminate human faces” — Results, Experiment 2.

> “In this experiment we tested discrimination of frontal views” — Discussion, final three paragraphs.

## Limits and reporting cautions

- Eight fish total across two separate four-fish experiments; trial totals cannot replace biological replication.
- Operant learning with food rewards and prior monitor-spitting experience, not spontaneous social recognition.
- Test asks fish to avoid one known unrewarded image while choosing new ones, not choose their owner among 44 people.
- Faces are shown in pairs. Forty-four and eighteen refer to novel-image pools, not simultaneous arrays or numbers of fish.
- Greyscale, normalized brightness and oval masks rule out only those cues. Local features, combinations and individual strategies remain unresolved.
- Original Discussion limits its claims to still frontal views. Later work on viewpoint changes exists (Newport et al. 2018, DOI 10.1016/j.anbehav.2018.09.002); therefore do not say no fish has ever been tested on rotations.
- The result does not establish that human faces need no specialized brain circuitry in humans, that fish understand identity or that owners are recognised.
- Source contains an internal Experiment 2 training-session inconsistency: text says 8–17 sessions but its individual list includes Fish 7 at seven. Source also describes the mixed-model link as logit in Results and log in Statistical analyses. Neither disputed detail appears in the narration or visual metrics. No attempt is made to silently resolve them.

## Correction/retraction check

On 13 November 2026 searched exact title and DOI combined with correction, erratum and retraction, inspected publisher/PMC records and PubMed metadata. No linked correction, retraction or expression of concern was found in the consulted sources. This dated search does not guarantee absolute absence. Author publication page also lists the 2016 article at https://caitnewport.com/publications/ .

## Photograph provenance and licence

**Source:** Reichenthal, A., Ben-Tov, M. & Segev, R. (2018). *Coding Schemes in the Archerfish Optic Tectum*. Frontiers in Neural Circuits 12:18. https://doi.org/10.3389/fncir.2018.00018 . Article: https://www.frontiersin.org/journals/neural-circuits/articles/10.3389/fncir.2018.00018/full .

**Figure 1A** is a photograph of an archerfish hunting an insect by shooting water. Methods/Animals specifies Toxotes chatareus and refers directly to Fig.1A. Caption has no separately excluded third-party credit. Copyright section says ©2018 Reichenthal, Ben-Tov and Segev and explicitly distributes the article under Creative Commons Attribution (CC BY). Licence URL: https://creativecommons.org/licenses/by/4.0/ .

**Original downloadable figure:** https://www.frontiersin.org/files/Articles/333383/xml-images/fncir-12-00018-g0001.webp . Use only panel A, not the other panels showing electrophysiological apparatus/stimuli. Crop coordinates and exact source SHA-256 must be pinned after production downloads and inspects the image. The asset is a still photograph with camera movement; it is never described as study footage, a frame from the 2016 face test, or a measured movement trajectory.

Caption credit: “Archerfish reference photograph: Reichenthal, Ben-Tov & Segev (2018), Front. Neural Circuits 12:18, Fig.1A, CC BY 4.0, cropped; from a separate study. Original conceptual task illustrations; synthetic narration.” Full source link and licence appear on the source page. No ShareAlike assets are selected.

## Storyboard and scientific visual constraints

Nine narration segments, 156 spoken words. 1080×1920, target 60–85 seconds after actual voice timing. The JSON holds exact captions and shot boundaries. Segment 1 starts with living-fish reference imagery then a labelled conceptual water-jet hunt; 2 introduces the two-face screen; 3 depicts food/no-food training; 4–5 keep the same familiar negative face while changing only its novel alternative; 6 standardizes greyscale faces, oval outline and brightness; 7 presents qualitative discrimination with exact cues unresolved; 8 uses four fish to show each small trained group and a clear lab/person distinction; 9 returns to the fish photo and the narrow scientific payoff.

All task faces are original schematics. Fish morphology is based on T. chatareus, including dark markings and an upturned mouth; it is not a generic goldfish icon. The water-jet path is illustrative, no numerical velocity or trajectory measurement. A left/right layout is for readability and must not claim the experiment rewarded a fixed screen side. A label names conceptual illustration and the photograph is identified as reference. No chart implies extrapolation to wild populations. The official Toxic Tales logo stays within the agreed phone-safe area. No carousel is produced.



Source photograph byte verification: original WebP SHA-256 56db2e031ead7bc20a724eb18108046857d74160fe5095f6f0cc1c760c4f513f, 1624 x 2028 pixels. Root visually inspected the original figure and the exact pixel crop (117, 59, 529, 793): only panel A photograph is included; no electrophysiology panels or labels. The unchanged original is downloaded and hash-checked before cropping.

Published versions