A functional test cannot be failed on biological grounds

{"Goldenberg":[0],"and":[1,16,48,78,125,129,155,176,214,242],"Gross":[2],"(Nature":[3],"Human":[4],"Behaviour,":[5],"2026)":[6],"argue":[7],"that":[8,29,40,50,108,170,187,209,216],"large":[9],"language":[10,96,141,189,199,217],"models":[11,97,190,200,218],"do":[12,219,225],"not":[13,54,69,220],"have":[14],"emotions,":[15],"set":[17,175],"a":[18,86,140,171,207],"functional":[19,42,172,203],"test":[20,43,67],"for":[21,156],"the":[22,25,31,41,45,66,99,111,143,157,177,185],"question:":[23],"does":[24,53],"system":[26,112],"contain":[27],"processes":[28],"play":[30,34],"roles":[32],"emotions":[33,71,84,192],"in":[35,121,123,126,212],"humans?":[36],"This":[37,230],"Correspondence":[38],"agrees":[39],"is":[44,60,113,148,163,169,174,179,193,226,231],"right":[46],"one,":[47],"argues":[49],"their":[51],"letter":[52],"apply":[55],"it.":[56,222],"Their":[57,62],"stated":[58],"standard":[59],"substrate-neutral.":[61],"own":[63],"text":[64],"says":[65],"\\"does":[68],"require":[70],"to":[72,82,85,139],"take":[73],"any":[74],"particular":[75],"physical":[76],"form,\\"":[77],"they":[79,94,118,135,224],"are":[80,131,136],"willing":[81],"attribute":[83],"frightened":[87],"crowd.":[88],"Yet":[89],"at":[90,165],"each":[91],"step":[92],"where":[93],"rule":[95],"out,":[98],"evidence":[100,146,162],"concerns":[101],"brains,":[102],"neural":[103],"organization,":[104],"or":[105],"transformer":[106],"architecture,":[107],"is,":[109],"what":[110],"made":[114,181],"of.":[115],"The":[116,167],"tests":[117],"propose,":[119],"changes":[120],"attention,":[122],"speed,":[124],"response":[127],"type":[128],"likelihood,":[130],"themselves":[132],"substrate-neutral,":[133],"but":[134],"never":[137],"applied":[138],"model:":[142],"only":[144],"language-model":[145,161],"discussed":[147],"answered":[149],"with":[150],"findings":[151],"from":[152],"brain":[153],"imaging,":[154],"second":[158],"function":[159,208],"no":[160,237,240,243],"cited":[164],"all.":[166],"conclusion":[168],"bar":[173],"judgment":[178],"then":[180],"on":[182,202],"substrate,":[183],"so":[184],"claim":[186],"present-day":[188],"lack":[191],"assumed":[194],"rather":[195],"than":[196],"demonstrated.":[197],"Ruling":[198],"out":[201],"grounds":[204],"requires":[205],"identifying":[206],"emotion":[210],"serves":[211],"humans":[213],"showing":[215],"perform":[221],"Whether":[223],"an":[227,232],"empirical":[228],"question.":[229],"argumentative":[233],"piece.":[234],"It":[235],"reports":[236],"new":[238],"data,":[239],"experiments,":[241],"figures.":[244]}

Authors

Institutions

Publication Details

Journal
Zenodo (CERN European Organization for Nuclear Research)
Published
2026-09-17
DOI
https://doi.org/10.5281/zenodo.22814094
Primary Topic
Face Recognition and Perception
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
preprint

A functional test cannot be failed on biological grounds

Leo Ferres
Zenodo (CERN European Organization for Nuclear Research)
Face Recognition and Perception
preprint

A functional test cannot be failed on biological grounds

Leo Ferres
preprint en

Abstract

Goldenberg and Gross (Nature Human Behaviour, 2026) argue that large language models do not have emotions, and set a functional test for the question: does the system contain processes that play the roles emotions play in humans? This Correspondence agrees that the functional test is the right one, and argues that their letter does not apply it. Their stated standard is substrate-neutral. Their own text says the test "does not require emotions to take any particular physical form," and they are willing to attribute emotions to a frightened crowd. Yet at each step where they rule language models out, the evidence concerns brains, neural organization, or transformer architecture, that is, what the system is made of. The tests they propose, changes in attention, in speed, and in response type and likelihood, are themselves substrate-neutral, but they are never applied to a language model: the only language-model evidence discussed is answered with findings from brain imaging, and for the second function no language-model evidence is cited at all. The conclusion is that a functional bar is set and the judgment is then made on substrate, so the claim that present-day language models lack emotions is assumed rather than demonstrated. Ruling language models out on functional grounds requires identifying a function that emotion serves in humans and showing that language models do not perform it. Whether they do is an empirical question. This is an argumentative piece. It reports no new data, no experiments, and no figures.

Zenodo (CERN European Organization for Nuclear Research)
Universidad del Desarrollo del Estado de Puebla (MX)
Peace, Justice and strong institutions
Face Recognition and Perception
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.