Skip to main content

1.1 Potenciali in tveganja

IfČe yousi rememberzapomnite onlyle oneen sentencestavek forms thiste page:strani: AIUI hasje madev itsšole wayprišla intokot schoolszdravilo likebrez anavodila medicineza thatuporabo. comesKot withoutučitelj ane patientveste, informationkdaj leaflet.uporabiti AsUI, animate teacher,navodil youza don’todmerjanje knowin whenni topodatkov useo AI,stranskih youučinkih. lackTo dosagepoglavje instructions,je andnavodilo thereza isuporabo notega information on the side effects. This chapter serves as the patient information leaflet for the Playbook.priročnika.

Ali

Shouldbi AImorala comeUI withpriti z navodilom za uporabo, tako kot zdravilo?

Poznate tisti listič papirja, ki je vtaknjen v vsako škatlico zdravila? Tistega, ki ga verjetno odvržete, ne da bi ga sploh pogledali? Morda se zdi dolgočasen, a leafletta likemajhen alist medicine?papirja

Youje knowtih thattriumf slippreteklega ofstoletja. paperTa tuckednavodila insiderazkrijejo everyštiri medicinebistvene box?stvari: Thekaj onezdravilo youzdravi, likelykoliko tossga asideje withouttreba avzeti, glance?kakšna Itpresenečenja maylahko seemprinese, dull,in butkdo thisnaj littlese papermu isizogiba. a

Za quietvsako triumphvrstico ofna thetem lastnavodilu century.stoji Thatceloten leafletmehanizem: spellsklinična outpreskušanja, fourregulatorji essentials:in whatodgovornost therazkriti medicinenapake. treats,Nihče howne muchverjame, toda take,nas theozdravi surprisessamo itnavodilo. mightNjegova bring,prava andmoč whoje shouldv steertem, clear.

da

naredi
Behindzdravilo everyvarno lineza onvsakogar. thatPrav leafletta standsčarovnija anam wholemanjka, machinery:ko clinicalUI trials,pustimo, regulators,da andpoučuje abrez responsibilityusmerjanja. to

Orodja revealUI, mistakes.ki Nose onezdaj believestiho thenaseljujejo leafletv itselfvaši healsšoli, you.so Itsprišla realbrez powernavodil. isV makingVeliki theBritaniji medicineje safepregled forpokazal, anyone.da Thatje is the magic we are missing when we let AI teach without guidance.


The AI tools now quietly settling into your school arrived with no instructions. In the UK, a review found that onlyle 7 out ofod 100 educationpodjetij technologyza companiesizobraževalno hadtehnologijo eversvoje putizdelke theirkdaj productspodvrglo throughnadzorovanemu apreskusu, controlled trial, and onlyle 12 out ofod 100 hadpa soughtje outsidepoiskalo certification.zunanjo Thecertifikacijo. classroomEnako tellszgodbo thepripoveduje samerazred. story. OnlyLe 11 out ofod 100 schoolvodij leadersšol andin teachersučiteljev acrossv 17 Americanameriških stateszveznih haddržavah everje requestedpred peer-revieweduvedbo proofnovega beforeorodja adoptingkdaj azahtevalo newrecenzirane tool.
dokaze.

Rezultat

Theje result,v inmnogih manyprimerih cases,zapravljen isdenar. wastedV money.ZDA Inje thedve US,tretjini two-thirdslicenc ofza schoolšolsko softwareprogramsko licensesopremo gatherednabiralo dust,prah, and98 98%% sawpa almostjih noskoraj use.ni Thebilo v uporabi. Projekt EdTech Genome Projectje lookedpreučil atpribližno about 7,7.000 teachingučnih toolsorodij worthv vrednosti 13 billionmilijard dollarsdolarjev andin foundugotovil, thatda 85%je werebilo either85 mismatched% ormed misused.
njimi

neustreznih
Thisali chapternapačno mayuporabljenih. look

To likepoglavje these dry,morda necessaryzdi section, but it reveals the hidden stories behind the rules. It tells what really happened, who was affected, and shinessuhoparen, a lightnujen ondel, thevendar siderazkriva effects,skrite evenzgodbe thoseza thatpravili. quietlyPove, reachedkaj childrense whoje neverresnično signedzgodilo, up.

kdo
je bil prizadet, in osvetli stranske učinke – tudi tiste, ki so tiho dosegli otroke, ki se zanje sploh niso prijavili.

Potenciali UI

Bodimo jasni že na začetku. Večina raziskav o potencialu UI se osredotoča na študente na univerzah. Šolski učenci so v teh raziskavah doslej razmeroma redko zastopani. Večina raziskav se nanaša na kratkotrajne intervencije, ki trajajo od nekaj ur do nekaj tednov. Posledično je na rezultate morda pozitivno vplival tako imenovani učinek novosti. Ne nazadnje so test za merjenje učinkovitosti pogosto zasnovali prav tisti, ki so izvajali pouk. Tudi to je lahko izkrivilo podatke.

Potentials(1) ofNaj AIUI razloži novo temo, vajo pa prihranite za svojo uro

Let'sPredstavljajte makesi this clear right from the start. Most studies on AI's potential focus on university students. School pupils have so far featured relatively rarely in these studies. Most studies relate to short-term interventions lasting from a few hours to a few weeks. Consequently, the so-called novelty effect may have positively influenced the results. Last but not least, the test used to assess effectiveness was often devised by the very people who delivered the lessons. This, too, may have confounded the data.

(1) Let AI explain a new topic and keep the practice for your lesson

Imagine this.tole. Tomorrow,Jutri, whenko aučenec pupilzamudi missestemo aali topicpotrebuje orponovno needsrazlago thatzahtevnega trickypojma, conceptnaj explainedprvo yetrazlago again,prevzame letUI-tutor. anVajo AIprihranite tutorza handlesvojo theuro, firstkjer explanation.resnično Reserveučenje thezares practice for your lesson, where real learning comes alive.zaživi.

WhyZakaj thisje matters.to pomembno. AIUI canlahko supportpodpre acquisitionusvajanje offaktičnega factualznanja. knowledge.V Inraziskavi aUniverze Harvard study,se je 194 physicsštudentov studentsfizike learnedučilo twodve topics.temi. ForPri oneeni topic,temi theyso useduporabili anUI-tutorja, AIrazvitega tutorprav developedza forta thisnamen. purpose.Pri Fordrugi thetemi otherso topic,se theyučili learnedv tradicionalnem razredu. Vsak študent je opravil oba pristopa, zato nihče ne more trditi, da je močnejša skupina prejela boljšo obravnavo. Na testu, izvedenem takoj po tem, so študenti, ki jih je poučevala UI, dosegli bistveno boljše rezultate. Raziskovalci so zabeležili mediani čas 49 minut za dokončanje naloge, v primerjavi s 60 minutami, predvidenimi za uro po urniku. Avtorji so pri razlagi teh rezultatov previdni: snov je bila za študente nova, velik del naloge pa je bil preprosto dobra razlaga. Izrecno zavračajo mnenje, da bi enako veljalo za naloge, ki zahtevajo povezovanje več idej. Poleg tega je šlo za dodiplomske študente, ne za učence 8. razreda.

Do podobnega zaključka je prišel obsežen pregled, ki je združil rezultate 228 raziskav. Klepetalniki na osnovi generativne UI so imeli večji učinek na poznavanje (in razumevanje) kot inteligentni tutorski sistemi in aprilagodljiva traditionalvadbena class.programska Each student completed both approaches, so no one can claim that the stronger group received the better treatment. In a test conducted immediately afterward, the students taught by the AI achieved significantly higher results. The researchers recorded a median time of 49 minutes to complete the task, compared with the 60 minutes allocated for the lesson in the timetable. The authors are cautious in their interpretation of these results: the material was new to the students, and much of the task simply involved explaining it well. They expressly reject the idea that the same would apply to tasks requiring the integration of multiple ideas. Furthermore, these were undergraduate students, not Year 8 pupils.

A large review pooling the results of 228 studies came to a similar conclusion. Generative AI-based chatbots produced a larger effect on knowing (and understanding) than intelligent tutoring systems and adaptive practice software. oprema.

GoPojdite deeper.globlje. Kestin, G., Miller, K., Klales, A., Milbourne, T., & Ponti, G. (2025). AI tutoring outperforms in-class active learning: An RCT introducing a novel research-based design in an authentic educational setting. Scientific Reports, 15, Articlečlanek 17458. FreeProsto access,dostopno, DOI 10.1038/s41598-025-97652-6. TheČlanek articleopisuje, describeskako howje thebil tutor waszasnovan designedin andnadzorovan. controlled.Koristno Thisza isvsakogar, usefulki forbi anyonerad wishingsam torazvil develop an AI tutor themselves.UI-tutorja.

(2) MakingBoljša betterizraba usečasa ofza lessonpripravo planningna time

pouk

ImaginePredstavljajte this.si tole. SavePrihranite somenekaj ofčasa, theki timega youpotrebujete needza forpripravo lessonna planningpouk, bytako usingda AIs topomočjo draftUI yourpripravite firstprvi lessonosnutek planučne andpriprave, thennato revisepa itga thoroughly.temeljito predelate.

WhyZakaj this matters. No one disputes that teachers’ workloads are heavy. Therefore, one potential benefit of AI is using working hours more efficiently. A UK study investigated the extentje to whichpomembno. theNihče targetedne useoporeka, ofda AIje forobremenitev lessonučiteljev planningvelika. canZato helpje toena reduceod teachers’možnih workload.koristi ResearchersUI randomlyučinkovitejša assignedraba delovnega časa. Britanska raziskava je preučila, v kolikšni meri lahko ciljno usmerjena uporaba UI za pripravo na pouk pomaga zmanjšati obremenitev učiteljev. Raziskovalci so 259 scienceučiteljev teachersnaravoslovja atna 68 schoolsšolah tonaključno eitherrazporedili abodisi groupv usingskupino, aki chatbotje alongsideuporabljala briefklepetalnika writtenob instructionskratkih orpisnih anavodilih, groupbodisi thatv continuedskupino, lessonki planningje aspripravo before.na Thepouk weeklynadaljevala planningkot timedoslej. forTedenski sciencečas lessonspriprave na ure naravoslovja v 7. in Years8. 7razredu andse 8je fellzmanjšal froms aroundpribližno 81 minutesna topribližno around56 56.minut. ASkupina panelstrokovnjakov ofza subjectposamezne expertspredmete assessedje theocenila qualitykakovost ofučnih thepriprav, lessonne plans.da Theybi didvedela, notali knowso whetherbile thepripravljene planss hadpomočjo beenUI createdali withbrez ornje. withoutStrokovnjaki theniso aidnašli ofdokazov, AI.da Thebi expertsse foundkakovost nopriprav evidencerazlikovala. toVendar suggestje thattreba thepoudariti qualitydve of the lesson plans differed. However, two points must be highlighted.stvari: (1) TheUčitelji teachersso usedorodje thez toolvsakim lesstednom anduporabljali lessmanj, askar thepomeni, weeksda wentprihranjen on,čas meaningne thatnastopi thesamodejno timein savedse doessam notpo occursebi automaticallyne and does not sustain itself.ohranja. (2) ThePreskus trialje measuredmeril thečas, timeki spentso byga teachersučitelji onporabili lessonza planning,pripravo notna theirpouk, pupils’ne learningpa outcomes.učnih dosežkov njihovih učencev.

GoPojdite deeper.globlje. Roy, P., Poet, H., Staunton, R., Aston, K., & Thomas, D. (2024, 1212. December)december). ChatGPT in lesson preparation: A Teacher Choices Trial. NFER, commissioned by thenaročilo Education Endowment Foundation and thein Hg Foundation. FreeProsto atdostopno na educationendowmentfoundation.org.uk.

(3) ProvidingBoljša betterpodpora supportza foručence, learnerski whood dostandardnega notpouka benefitnimajo fullypolne from standard lessons

koristi

ImaginePredstavljajte this.si tole.Your schoolVaša decidesšola howse itodloča, wisheskako toželi spendporabiti itssvoja funds.sredstva. ItOdloči wantsse toza purchasenakup anUI AIklepetalnika. chatbot.Kako Howverjetno likelyse dovam youzdi, thinkda itvam isbodo thatponudniki theklepetalnika chatbotpredložili providersdokaze willo providekoristih youbota withza evidenceposamezne of the bot’s benefits for different subjects?predmete?

WhyZakaj thisje matters.to pomembno.The providerPonudnik maylahko pointizpostavi outnaslednjo theuporabo. followingKlepetalniki application.so Chatbots are the oldest and best-documented application of ‘AI’najstarejša in education,najbolje anddokumentirana thisaplikacija is„UI“ preciselyv whatizobraževanju, iskar oftenje overlookedprav whentisto, allocatingkar funds.se Inpri andodeljevanju analysissredstev ofpogosto spregleda. V analizi 29 studiesraziskav, involvingki je vključevala 41 groupsskupin ofučencev learnerss withposebnimi disabilities,potrebami, researchersso foundraziskovalci moderateugotovili benefitszmerne acrosskoristi av wideširokem rangenaboru ofokoliščin. settings.A Butpreberite readdrobni thetisk. smallV print.sedmih Inod sevendesetih outteh ofraziskav tenje of„UI“ thesepomenila studies,robota: ‘AI’majhno meantfizično anapravo robot:na amizi, smallki physicaljo machineje sittingpogosto onupravljal araziskovalec table,v oftendrugem operatedprostoru. byLe aena researcherod inpetih anotherraziskav room.je Onlyuporabila oneprogramsko inopremo. fiveTo studiestorej usedni software.dokaz, Soda thisbo isklepetalnik nopomagal proofavtističnemu thatučencu av chatbotvašem willrazredu. helpPregledovalci anprav autistictako pupilniso inmogli yourpovedati, class.kaj Norte couldintervencije thenaredi reviewersučinkovite. sayNa whatpodlagi makessvoje theseraziskave interventionsniso effective.mogli Basedopredeliti, onpo theirčem research,se theyraziskave, couldki notso identifydelovale, whyrazlikujejo theod studiestistih, thatki worked differed from those that did not.niso.

GoPojdite deeper.globlje. Zhang, L., Carter, R. A., Jr., Liu, Y., & Peng, P. (2026). Let's CHAT about artificial intelligence for students with disabilities: A systematic literature review and meta-analysis. Review of Educational Research, 96 (96(1). DOI 10.3102/00346543241293424. ThePravi realbiser gemje ispreglednica thevključenih tableraziskav: oftakoj includedrazkrije, studies:kako itmajhen instantlydel revealstega justpodročja howse littledejansko ofosredotoča thisna fieldorodja, actuallyki focusesse ontrenutno thetržijo tools currently being marketed to schools.šolam.

(4) TeachingUčenje learnerso howtem, AIkako fails

UI odpove

ImaginePredstavljajte this.si tole. AskRazred yournaj classustvari todeset create ten images ofslik 'a doctor' and ten of 'a nurse.' Tally up who appearszdravnika' in eachdeset set,slik then'medicinske sparksestre'. aPreštejte, discussion:kdo wherese didpojavi thesev patternsvsakem comenaboru, from,nato andpa whosprožite maderazpravo: thoseod choices?kod izvirajo ti vzorci in kdo je te odločitve sprejel?

Gemini_Generated_Image_mfaur1mfaur1mfau.jpg
Gemini_Generated_Image_mfaur1mfaur1mfau.jpg

FigSlika 1. ImageSlika, createdustvarjena withz orodjem NanoBanana 2, creatednastala August,20. 20,avgusta 2026, Onepoziv shotv prompt:enem akoraku: doctor,zdravnik, amedicinska nursesestra

WhyZakaj thisje matters.to pomembno. TheKlasična classicrazličica versionte ofučne thisure lessonse feelszdi outdated,zastarela, andin thatprav iszato exactlysi whyzasluži itpozornost. deservesLeta attention.so Forbile years,predstavitve thepredvidljive: demonstrationprosite wasza predictable:zdravnika, askin forpojavi ase doctor,moški. andV aenem mansistematičnem appears.testu In one systematic test,je Midjourney almostzdravnike alwaysskoraj picturedvedno doctorsupodobil askot whitebele men,moške, whilemedtem ko je Adobe Firefly madenaredil visible,vidne, ifčetudi imperfect,nepopolne attemptsposkuse to diversify.raznolikosti.

TryPoskusite theisti samepoziv promptdanes today,in andmorda youse mightprikaže seezdravnica aali femalemedicinski doctor or a male nurse appearbrat (Fig.slika 1). But do not be fooled into thinking the bias has vanished. A 2026ne analysisdajte ofse 1,prevarati v prepričanje, da je pristranskost izginila. Analiza iz leta 2026, ki je zajela 1.344 imagesslik fromtreh threepriljubljenih populargeneratorjev, generatorsje revealedrazkrila thenasprotno: opposite:ko whenso asked for azahtevali 'competentsposobno person,osebo', en sistem ni pokazal nobene ženske.

Kar se je spremenilo, ni pristranskost sama, temveč plast, v kateri živi – in to je pomemben nauk! Za vsako od teh slik stojijo človeške odločitve: na čem je bil sistem naučen in kaj so njegovi ustvarjalci pozneje odločili, da naj prikazuje. Slednje je mogoče prilagajati, je nevidno in ga določi podjetje.

Prva plast človeških odločitev je tisto, kar je sistem vsrkal. Milijoni slik, ki so jih ustvarili in označili ljudje, vsaka od njih odseva svet, kakršen je že bil. To je plast, ki jo razkrije klasična lekcija o pristranskosti, in učenci jo zlahka razumejo: stroj je postal odsev tega, s čimer je bil hranjen.

Drugo plast oblikuje to, kar se podjetje odloči pokazati naprej, vendar ta vpliv v sami sliki ne pusti nobene sledi. Tu se odvijejo tri stvari, vsako od njih so ustvarjalci opisali s svojimi besedami.

Vaše besede so prepisane, preden jih model prejme. Iz lastne razvijalske dokumentacije OpenAI: „uporabljamo GPT-4 za optimizacijo vseh vaših pozivov, preden jih posredujemo DALL-E“ – in „te funkcije trenutno ni mogoče izklopiti“. Kar je vaš učenec dejansko natipkal, ni tisto, kar je bilo dejansko vprašano stroju.

Videz se dopolni, kadar ga poziv pusti odprtega. OpenAI o svoji tehniki raznolikosti pravi: „Ta tehnika se uporabi na ravni sistema, kadar DALL·E prejme poziv, ki opisuje osebo, ne da bi opredelil raso ali spol, na primer 'gasilec'.“ onePo systemtem showedso nouporabniki women.dvanajstkrat pogosteje povedali, da slike prikazujejo ljudi različnih ozadij. „Zdravnik“ je natanko takšen poziv.

In to prilagajanje lahko tudi spodleti. Google je po tem, ko je leta 2024 začasno ustavil Geminijevo generiranje slik, zapisal: „Naše prilagajanje, s katerim smo želeli zagotoviti, da Gemini prikazuje raznolik nabor ljudi, ni upoštevalo primerov, kjer raznolikost očitno ni bila primerna.“

Ko torej učenec natipka „zdravnik“ in se prikaže ženska, slika sama ne ponuja nobenega namiga, katera plast jo je ustvarila. Ta negotovost ni napaka učne ure. Je nauk sam.

Stara zgodba je trdila, da stroj preprosto vpije predsodke družbe. Nova zgodba seže globlje in traja dlje: slika ni zrcalo in ni naključje. Podjetje je sprejelo odločitev! Te odločitve v rezultatu ne morete opaziti, in lahko se spremeni kar v torek, tiho, brez opozorila. Ta resnica preživi vsak nov model. Lekcija o stereotipih zbledi.

To je lekcija, ki si zasluži pozornost učencev in bo preživela naslednjo posodobitev modela: slika ni preprost odsev resničnosti in ni nastala po naključju. Nekdo je sprejel odločitev.

To lekcijo je mogoče začeti zgodaj. Ko je 209 finskih učencev v 12 razredih, 4. in 7. razreda, raziskovalo to temo, se je delež tistih, ki so lahko pristranskost pojasnili s podatki, povečal s približno 7 od 100 na 44 od 100. Poštena zadržanost: raziskava je merila, kaj so otroci znali pojasniti, ne pa, kaj bi kasneje dejansko storili.

WhatPojdite changed is not the bias but the layer it lives in - and this the important lesson!globlje. Human decisions stand behind any of these pictures: what the system was trained on, and what its makers afterward decided it should show. The second is adjustable, invisible, and made by a company. 

The first layer of human decisions is what the system has absorbed. Millions of images, crafted and labeled by people, each one echoing the world as it already was. This is the layer that the classic bias lesson reveals, and pupils grasp it easily: the machine became a reflection of what it was fed.
The second layer is shaped by what a company chooses to show you next, yet this influence leaves no trace in the image itself. Three things unfold here, each one described by the creators in their own words.
    Your words are rewritten before the model receives them. From OpenAI’s own developer documentation: “we use GPT-4 to optimize all of your prompts before they’re passed to DALL-E” — and “this feature isn’t able to be disabled at the moment.” What your pupil typed is not what the machine was asked. Appearance gets filled in when the prompt leaves it open. OpenAI, describing its diversity technique: “This technique is applied at the system level when DALL·E is given a prompt describing a person that does not specify race or gender, like ‘firefighter.’” Afterwards, users were twelve times more likely to say the images showed people of diverse backgrounds. “A doctor” is exactly that kind of prompt. And that tuning can miss. Google, after pausing Gemini’s image generation in 2024: “Our tuning to ensure that Gemini showed a range of people failed to account for cases that should clearly not show a range.”
    So when a pupil types “a doctor” and a woman appears, the image offers no clue about which layer brought her into being. That uncertainty is not a flaw in the lesson. It is the lesson itself.
    The old story claimed the machine simply soaked up society’s prejudices. The new story cuts deeper and lasts longer: the image is no mirror, and it is no accident. A company made a choice! You cannot spot that choice in the outcome, and it might shift on a Tuesday, quietly, without warning. That truth endures through every new model. The stereotype lesson fades away.

    This is the lesson that deserves a student's attention, and it will outlast the next model update: the image is not a simple reflection of reality, and it did not happen by chance. Someone made a choice.

    This lesson can start early. When 209 Finnish students in 12 classes, grades 4 and 7, explored this topic, the number who could explain the bias using data jumped from about 7 out of 100 to 44 out of 100. The honest reservation: the study measured what children could explain, not what they would do later.

    Go deeper. Vartiainen, H., etin alsod. (2025). Enhancing children's understanding of algorithmic biases in and with text-to-image generative AI. New Media & Society, 27(27(9). Free,Prosto dostopno, DOI 10.1177/14614448241252820

    Weinmann, H., etin alsod. (2026). Gender bias in text-to-image generative artificial intelligence: Neglect and stereotypical presentations across three popular platforms. New Media & Society, Online First

    How

    Kako solidzanesljivi areso thetrenutno researchraziskovalni resultsizsledki ono thepotencialih potentials of AI at this very moment?

    UI?

    TheNajpomembneje. mostČe importantvam thing.nekdo Ifnavede someoneraziskavo, quotespred astrinjanjem studypostavite todve you,vprašanji: ask„Kdo twoso questionsbili beforeučenci agreeing:in "Whokako weredolgo theje learners,raziskava and how long did the study last?"trajala?“

    WhyZakaj thisje matters.to pomembno.In 2025,Leta a2025 meta-analysisje ofmetaanaliza 51 studiesraziskav reportedporočala ao significantpomembni benefit ofkoristi ChatGPT forza pupils’učno academicuspešnost performance.učencev. TheČlanek articleso wasprebrali read hundreds of thousands of times and served as the basis for numerous confident statementsstotisočkrat in staffje rooms.služil Onkot 22podlaga Aprilza 2026,številne thesamozavestne journaltrditve retractedv thezbornicah. article.22. Aaprila *retraction*2026 meansje thatrevija ačlanek journalumaknila. officiallyUmik withdrawspomeni, somethingda itrevija hasuradno published;prekliče thenekaj, articlekar remainsje visible,objavila; markedčlanek withostane theviden, noteoznačen ‘RETRACTED’z opombo „UMAKNJENO“, tako da lahko vsak, ki ga je citiral, to prepozna. Dva zunanja raziskovalca sta odkrila neskladja pri sintezi raziskav. Urednik revije je zapisal, da težave „spodkopavajo urednikovo zaupanje v veljavnost analize“. Avtorji na dopisovanje o tej zadevi niso odgovorili.

    To ne pomeni, da UI nima učinka. Obstajajo verodostojne raziskave, ki dokazujejo zmerne koristi. To raziskovalno področje prav tako še ni dobro opredeljeno, saj je razmeroma novo in zelo dinamično. Pomeni pa, da je bila pogosto navajana številka napačna in da je bilo za to potrebno leto dni in dva zunanja strokovnjaka.

    Še en pridržek, vreden omembe: bliže ko so thatdokazi anyonešoli, whovsaj hastrenutno, citedmanjši itje can recognise this. Two external researchers had identified inconsistencies in the synthesis of the studies. The journal's editor wrote that the problems ‘undermine the editor’s confidence in the validity of the analysis’. The authors did not respond to correspondence on this matter.učinek.

    ThisPojdite doesgloblje. notSam meanumik thatje AIdolg haseno nostran, impact.prosto Theredostopen, arein reputablevreden studiesbranja thatprav demonstratezato, moderateker benefits.je This field of research is also not yet well defined because it is relatively new and highly dynamic. It means that the figure that was frequently cited was incorrect, and that it took a year and two external experts to realise this.

    Another reservation worth mentioning is: The closer the evidence gets to school, at least at the moment, the smaller the effect. 

    Go deeper.  The retraction itself is one page, free, and worth reading precisely because it is short:kratek: Wang, J., & Fan, W. (2026). Retraction note: The effect of ChatGPT on students' learning performance, learning perception, and higher-order thinking. Humanities and Social Sciences Communications, 13,13, 528. DOI 10.1057/s41599-026-07310-z.

    TheUmaknjeni retractedizvirnik originalje wasbil HSSC 12, 621 (2025). StillŠe standing:vedno velja: Wu, X., Zhu, P., Zhang, J., Yin, M., & Wang, Y. (2026). *HSSC, 13*13, 684. Free,Prosto dostopno, DOI 10.1057/s41599-026-07019-z.
    The

    school-only

    Številka, figure:ki velja samo za šole: Yi, L., Liu, D., Jiang, T., & Xian, Y. (2025). *International Journal of Science and Mathematics Education, 23*23(4), 1105–1126. DOI 10.1007/s10763-024-10499-7.

    Tveganja UI

    Risks of AI

    (1) DoingNekaj somethingnarediti isni notisto thekot samese asnekaj learning

    naučiti

    ImaginePredstavljajte this.si tole. BeforePreden youzadate setdomačo analogo, homeworkki taskjo thatlahko aopravi chatbotklepetalnik, canpremislite, complete,kakšen thinkje aboutnamen whatnaloge. theMorda purposebi ofbilo thesmiselno tasknalogo is. Might it be a good idea to change the task?spremeniti?

    WhyZakaj thisje matters.to pomembno. DoSe youz learnklepetalnikom fasteručite withhitreje? Približno tisoč učencev od 9. do 11. razreda je vadilo matematiko v treh skupinah: ena je uporabljala standardnega klepetalnika, druga različico, programirano tako, da je zadrževala odgovor in namesto tega postavljala dodatna vprašanja, tretja pa klepetalnika ni uporabljala. Med uporabo klepetalnika je skupina s standardnim klepetalnikom opravila 48 odstotkov več vaj kot skupina brez njega. Nato so vsi opravljali isti izpit brez klepetalnika. Rezultat, ki se je sprva zdel presenetljiv, je bil, da je skupina s standardnim klepetalnikom dosegla 17 odstotkov slabše rezultate kot učenci, ki klepetalnika niso uporabljali. Skupina, ki je delala s klepetalnikom, ki je zadrževal odgovor, pri testu ni izkazala upada uspešnosti v primerjavi s skupino brez klepetalnika, a chatbot?je Aroundvadila aveč. thousandUčenci, pupilski inso Yearsuporabljali 9standardnega toklepetalnika, 11niso practisedbili mathsniti inleni threeniti groups:niso onegoljufali. usingTrudili aso standardse chatbot,bolj, onenaučili usingpa aso versionse programmedmanj, toker withholdorodje theni answerodpravilo andtruda, asktemveč amiselni follow-upizziv, question,ki andpredstavlja one without a chatbot. Whilst using the chatbot, the group with the standard chatbot completed 48 percent more practice exercises than the group without one. Afterwards, everyone sat the same exam without a chatbot. The result, which seemed surprising at first, was that the group using the standard chatbot performed 17 percent worse than the pupils who had not used a chatbot. The group that had worked with the chatbot which withheld the answer showed no drop in performance in the test compared to the group without a chatbot, but practised more. The pupils using the standard chatbot were neither lazy nor did they cheat. They worked harder and learnt less, because what the tool eliminated was not the effort, but the cognitive challenge that constitutes learning. učenje.

    GoPojdite deeper.globlje. Bastani, H., Bastani, O., Sungu, A., Ge, H., Kabakcı, Ö., & Mariman, R. (2025). Generative AI without guardrails can harm learning: Evidence from high school mathematics. *PNAS, 122*,122, e2422633122. DOI 10.1073/pnas.2422633122.

    (2) AIDetektorji DetectorsUI punishkaznujejo thenapačne wrong pupils

    učence

    ImaginePredstavljajte this.si tole. YouV receiveocenjevanje aprejmete student’sučenčev essayesej. toZanima grade.vas Curiousnjegova aboutpristnost, itszato authenticity,ga youvnesete feedv itdetektor intoUI. anRezultat AIse detector.izpiše: Theesej resultje flashes:bil theskoraj essayzagotovo wasnapisan almostz certainly written by AI.UI.

    WhyZakaj it matters. You use the AI detectorje to ensure fairness, yet for a particular group, the detectors have exactly the opposite effect.pomembno. ADetektor detectorUI doesuporabljate notzato, recogniseda machine-generatedzagotovite text.pravičnost, Itvendar measuresimajo howdetektorji predictableza thedoločeno wordingskupino is.ravno Andnasproten thatučinek. isDetektor preciselyne whereprepozna thebesedila, problemki lies,ga forje example,ustvaril whenstroj. aMeri, pupilkako ispredvidljivo writingje inbesedilo. aIn foreignprav language,v istem unpractisedje intežava, writing,na orprimer, haskadar aučenec limitedpiše everydayv vocabulary.tujem Theirjeziku, writingje ispri predictable.pisanju Whenneizurjen sevenali commercialima detectorsomejen analysedvsakdanji besedni zaklad. Njegovo pisanje je predvidljivo. Ko je sedem komercialnih detektorjev analiziralo 91 essaysesejev, writtenki byso non-nativejih speakersnapisali sittinggovorci, ankaterih Englishmaterni exam,jezik theseni essaysangleščina, werena incorrectlyizpitu flaggediz asangleščine, machine-generatedso inbili aroundti sixeseji outv ofpribližno tenšestih cases.od Bydesetih contrast,primerov essaysnapačno writtenoznačeni bykot ustvarjeni s strojem. V nasprotju s tem so bili eseji, ki so jih v svojem maternem jeziku, angleščini, napisali 14-year-oldletni AmericansAmeričani, inskoraj theirvsi nativepravilno language,razvrščeni. English,Orodje, wereki almostsproža alllažne classifiedpozitivne correctly.rezultate Atočno toolza thateno triggersskupino, falseni positivesnevtralno. forPoleg exactlytega oneje groupta isskupina notnajmanj neutral.sposobna, Furthermore,da thisse group is the least able to defend itself.brani.

    GoPojdite deeper.globlje. Liang, W., Yuksekgonul, M., Mao, Y., Wu, E., & Zou, J. (2023). GPT detectors are biased against non-native English writers. Patterns, 4(4(7), 100779. Free,Prosto dostopno, DOI 10.1016/j.patter.2023.100779.

    (3) WhereKam onza earthvraga willgredo thepodatki pupils' data be?

    učencev?

    ImaginePredstavljajte this.si tole. YouUporabljate useorodje anUI, AIki toolga yourponuja schoolvaša offers.šola. TryPoskusite tougotoviti, findkam outgredo wherepodatki the pupils' data go.učencev.

    WhyZakaj it matters. Can your school ensure that the data is not passed onje to apomembno. thirdAli party?lahko Avaša thirdšola partyzagotovi, isda companypodatki thatniso isposredovani neithertretji yourosebi? schoolTretja noroseba theje app’spodjetje, developer.ki Typically,ni theseniti arevaša advertisingšola orniti analyticsrazvijalec companiesaplikacije. thatObičajno receiveso datato whilstoglaševalska theali pupilanalitična ispodjetja, working.ki Noprejemajo onepodatke, atmedtem theko schoolučenec chosedela. thisNihče company.v Whenšoli researcherstega carriedpodjetja outni aizbral. technicalKo analysisso ofraziskovalci izvedli tehnično analizo 163 learningučnih productsizdelkov, recommendedki byso governmentsjih duringvlade thepriporočale pandemic-relatedmed schoolšolskimi closures,zaprtji zaradi pandemije, je 145 ofod themnjih processedobdelovalo children’spodatke dataotrok inna anačin, wayki thatje jeopardisedogrožal orali infringedkršil theirnjihove rightspravice, bytako transmittingda dataso topodatke posredovali 196 third-partytretjim companypodjetjem orali grantingjim themomogočili access,dostop predominantlydo fromnjih, thevečinoma advertisingiz industry.oglaševalske Ofpanoge. theOd 42 governmentsvlad, thatki developedso theirrazvile ownlastne productsizdelke rathernamesto thannakupa, purchasingjih them,je 39 hadrazvilo developedizdelke productsz withenako thetežavo. sameRaziskava problem.zajema Theizdelke studyiz covers products fromleta 2021, andnekateri someso mayse havemorda changedod sincetakrat then.spremenili. However,Vendar wene cannotmoremo ruleizključiti, outda thatenako thevelja sametudi appliesza toaplikacije AI applications. UI.

    GoPojdite deeper.globlje. Human Rights Watch (2022). "How dare they peep into my private life?: Children's rights violations by governments that endorsed online learning during the Covid-19 pandemic. FreeProsto atdostopno na hrw.org. TheDržavne countrypriloge annexesomogočajo letvpogled youv lookto, upkaj whatje wasbilo endorsedkje where.priporočeno.

    (4) AIUI arrivespride lastnazadnje wheretja, itkjer isje needednajbolj most

    potrebna

    ImaginePredstavljajte this.si tole.How muchKoliko trainingusposabljanja didste youže alreadyprejeli receiveo onučenju learningo aboutUI AIali orpredmetno subject-specificspecifičnem teachingpoučevanju withz AI?UI?

    WhyZakaj thisje matters.to pomembno.Adults whoOdraslim, hadki completedso theirzaključili educationizobraževanje, wereso providedzagotovili withUI-asistenta. anVrzel AIv assistant.uspešnosti Themed performancetistimi gapz betweenvišjo thosein withnižjo higherstopnjo andformalne lowerizobrazbe levelsse ofje formalznatno educationzmanjšala. narrowedKo significantly.so Whenraziskovalci thenato researchersasistenta thenumaknili, withdrewse theje assistant,vrzel thedeloma gapspet partiallypojavila. re-emerged.Raziskovalci Theso researcherssklenili, concludedda thatje theUI-asistent AIopravil assistantdelo, hadni carriedpa outudeležencem thepomagal work,razviti butnobenih had not helped participants develop any skills.veščin.

    AAnketa surveymed ofameriškimi Americanravnatelji schoolšol principalsje revealedpokazala thenaslednjo followingsliko: picture:bolj theko moreso disadvantagedbili aučenci school’sšole pupilssocialno-ekonomsko were,prikrajšani, themanjši lessje impactbil AIvpliv hadUI onna teaching.poučevanje.

    IfČe wezdružimo combineugotovitve theobeh findingsraziskav, fromlahko bothoblikujemo studies,naslednjo theytrditev: allowdobro usorodje tolahko makepremosti thevrzel, followingdokler statement:je Av goodrokah toolučenca. canVendar bridgeobičajno ane gapdoseže astistih, longki asga itnajbolj ispotrebujejo, in theko handsje ofodstranjeno, aza learner.sabo However,ne itpusti usuallynobenih does not reach those who need it, and leaves no skills behind when it is removed.veščin.

    GermanNemški figurespodatki showkažejo theenak samevzorec. patte.Med Among 1,1.590 youngmladimi, peoplestarimi agedod 14 todo 20,20 let, je 80 percentodstotkov iz najbolj premožnih družin in the most affluent families and 55 percentodstotkov iniz thenajmanj leastpremožnih affluentUI sawvidelo AIkot aspriložnost. anIn opportunity.prepoznavanju Andsledi recognitiontudi is followed by action:dejanje: 70 percentodstotkov ofgimnazijcev grammarje schoolUI pupilsuporabilo usedza AIdomače fornaloge, homework,v comparedprimerjavi withs 58 percentodstotki atna lowernižjih secondarysrednjih schools.šolah.

    Kaj pa strokovni razvoj učiteljev? Anketa med več kot deset tisoč učitelji v Angliji je pokazala, da je 45 odstotkov učiteljev na zasebnih šolah že prejelo usposabljanje o UI. V primerjavi s tem je bilo na javnih šolah takih 21 odstotkov. Znotraj sistema javnih šol so se deleži gibali od 26 odstotkov na najbogatejših šolah do 18 odstotkov na najrevnejših.

    (5) Kar se izboljša pri učenju z UI, ni nujno tisto, kar vidite

    Predstavljajte si tole. V svoje poučevanje uvedete novega klepetalnika, učenci pa so nad uporabo novega orodja navdušeni.

    WhatZakaj aboutje teacherto professionalpomembno. development?Obsežen Apregled surveyje ofugotovil moreštiri thanrazlične tenvidike thousandučenja teachers(poznavanje in Englandrazumevanje, founddelo thatz 45UI, percentangažiranost, ofmotivacija, teacherssamozavest, atsposobnost privatenačrtovanja, schoolsspremljanja hadin alreadypopravljanja receivedlastnega training on AI. This compared with 21 percent at state schools. Within the state school system, the figures ranged from 26 percent at the wealthiest schools to 18 percent at the poorest schools. 

    (5) What improves when learning with AI is not necessarily what you see

    Imagine thisdela). YouZadnji introducese ani newizboljšal, ChatBotostali intopa yourso teaching and the pupils enthusiastic about using the new tool.se.

    WhyPri this matters. A large review found four different things about learning (Knowing and understanding, working with AI, engagement, motivation, confidence, ability to plan, monitor and correct their own work). The last one didn't improve whereas the others did.

      For knowing and understanding generative AI had larger effects than intelligent tutoring systems and adaptive practice software that is already usedpoznavanju in schools.razumevanju Forje applyingimela knowledge,generativna andUI forvečje motivationučinke andkot engagement,inteligentni AItutorski did not perform better that already existing software. For self-steering we do not have enough studies to tell any effect.

      Here’s a quick side note: the greatest impacts are foundsistemi in theprilagodljiva artsvadbena andprogramska humanities,oprema, ratherki thanse v šolah že uporablja.

      Pri uporabi znanja ter pri motivaciji in mathematicsangažiranosti orUI theni naturaldelovala sciences.bolje kot že obstoječa programska oprema.

      Za samousmerjanje nimamo dovolj raziskav, da bi lahko govorili o kakršnem koli učinku.

      Še kratka opomba: največji učinki se kažejo na področju umetnosti in humanistike, ne pa pri matematiki ali naravoslovju.

      GoPojdite deeper.globlje. Yeo, G. H., & Lansford, J. E. (2025). Effects of artificial intelligence on educational functioning: A review and meta-analysis. Educational Psychology Review, 37(37(4), articlečlanek 110. DOI 10.1007/s10648-025-10085-5.