Ubuyobozi

AI muburyo bwo kwerekana amashusho

Kugereranya ibice bya videwo ya AI ikoresha urutoki ruhuza hamwe no kwiga imashini kugirango ibone amashusho yoherejwe cyangwa yerekanwe kumurongo wica amategeko ya platform, mubisanzwe mbere yuko abantu babibona.

Umva iki gitaboAI-byavuzwe · ~8 min

Kuriyi page9 min soma
  1. Incamake
  2. Ibyingenzi byingenzi
  3. Kwibira cyane
  4. Gusoma urubuga rwo gukorera mu mucyo kunenga
  5. Gushyira mu bikorwa Isi
  6. Kazoza ka AI muburyo bwo kwerekana amashusho
  7. Kubishyira mubikorwa
  8. Inkomoko no gusoma
  9. Komeza Ubushakashatsi
  10. Ibibazo bikunze kubazwa

Incamake

Bifite akamaro kubera igipimo: YouTube ivuga ko amasaha arenga 500 ya videwo agera buri munota, kandi mu mpera za 2023 sisitemu zikoresha zabanje kwerekana 96.1% ya videwo yakuyeho.

Ibyingenzi byingenzi

  1. Kugereranya amashusho bihuza hash guhuza ibintu bizwi, ibyiciro kubintu bishya, abasubiramo abantu kubijyanye, hamwe na raporo zabakoresha.
  2. Raporo ya YouTube yo mu Kwakira kugeza Ukuboza 2023 irerekana 96.1% ya videwo 9,012.232 yakuweho bwa mbere yashyizwe ahagaragara na sisitemu zikoresha.
  3. Kwiyumvisha ibintu nka Meta ifungura-isoko ya PDQ na TMK + PDQF ihuza amakopi ya videwo azwi udasangiye dosiye zangiza.
  4. Facebook yahagaritse kopi zirenga miliyoni 1.2 za videwo yibasiwe na Christchurch yoherejwe mu masaha 24, ariko abayashyize mu majwi babuze imbonankubone.
  5. Ikigo gishinzwe kugenzura ibikorwa bya Meta cyasanze icyiciro cyarakuyeho nabi poste yo kumenyekanisha kanseri y'ibere maze isaba kubwira abakoresha igihe automatike ikora.

Kwibira cyane

Video content moderation is the work of finding and acting on videos that break a platform's rules, such as graphic violence, child sexual abuse material, terrorist propaganda, nudity, harassment or dangerous acts. AI in this context means automated systems that inspect videos and their audio, text and metadata, then either take action directly or send the video to human reviewers. Platforms rely on automation because the volume is far beyond human capacity. In its October to December 2023 report filed under California law AB 587, YouTube said more than 500 hours of content are uploaded every minute and that the vast majority of content it reviews and acts on is first detected by automated systems.

There are two main families of technique. The first is matching against known content. A system computes a hash, a short digital fingerprint, of a video or its frames and compares it with a database of fingerprints from material already judged violating. Cryptographic hashes such as MD5 only match exact copies. Perceptual hashes such as Meta's PDQ for images and TMK+PDQF for video are designed so that visually similar files produce similar fingerprints, which catches copies that have been resized, re-encoded or lightly edited. Because only fingerprints are shared, organizations can pool them without exchanging the harmful files themselves.

The second family is classification of new content. Machine learning classifiers are trained on labeled examples to estimate how likely a video is to contain a given kind of violation. A pipeline can sample frames, runs image models on them, analyzes the audio and any speech, reads on-screen text and the title or description, and combines these signals into scores per policy area. Each score is compared with thresholds chosen by the platform: very high-confidence cases may be removed or restricted automatically, uncertain cases go to human moderators, and low scores pass. YouTube's report describes this as a collaborative loop in which human moderators confirm decisions and their feedback is used to retrain the models.

The final stage is the action and the appeal. Actions are not just removal. YouTube's report lists restricting a video, for example age-restricting content unsuitable for all audiences, as well as removal and channel strikes, with three strikes in 90 days able to end a channel. Users are notified and can appeal, and YouTube says appeals go to human review. The results show how early automation acts: in that quarter, YouTube removed 9,012,232 videos, 96.1% of them first flagged by automated systems, 3.4% by ordinary users and 0.6% by organizations in its Priority Flagger program. Just over half of removed videos, 51.51%, had zero views, and another 26.43% had between 1 and 10 views.

Shared databases grew out of specific crises. After a gunman livestreamed the March 2019 attack on two mosques in Christchurch, New Zealand, Facebook reported that the live broadcast was viewed fewer than 200 times and about 4,000 times in total before removal, that the first user report came 29 minutes after the stream began, and that in the following 24 hours it blocked more than 1.2 million copies at upload and removed about 300,000 more. It found more than 800 visually distinct variants and used audio matching to catch edits whose visuals had changed but whose soundtrack had not. That year GIFCT, the Global Internet Forum to Counter Terrorism, expanded its Hash-Sharing Database from content produced by groups on the United Nations Security Council sanctions list to include content from perpetrators who livestream or record attacks, and later added attacker manifestos and URLs. In August 2019 Facebook released PDQ and TMK+PDQF as open source.

Compared with alternatives, each method has a clear role. Human-only review cannot keep pace with hundreds of hours of uploads per minute. User reporting is valuable for context but slow: in the Christchurch case no user reported the live video until after the broadcast ended. Hash matching is precise and cheap but only finds content someone has already identified, and determined uploaders try to evade it with edits. Classifiers can find new material but make more mistakes. In practice, platforms layer all four, using hashes for known material, classifiers for triage, humans for context and appeals, and user reports as an additional signal.

The limitations are significant and documented. Classifiers need many training examples, and Facebook explained that its systems missed the Christchurch livestream partly because such events are rare and partly because similar-looking footage, such as livestreamed video games, would cause floods of false alarms. Classifiers also miss context. In a January 2021 decision, Meta's Oversight Board found that a machine learning classifier had removed a Brazilian user's Instagram post about breast cancer symptoms under the nudity policy, failing to recognize the Portuguese words for breast cancer on the image, even though awareness posts were explicitly allowed. The Board recommended telling users when automation acted, allowing human review of such appeals, and auditing samples of automated removals. Automated systems can also perform unevenly across languages and cultures, and they struggle with satire, news reporting and documentary uses that YouTube's policies explicitly exempt.

In practice, building video moderation requires clear written policies, labeled training data for each policy area, reviewer teams and tooling, appeal processes, and metrics such as the share of removals reversed on appeal. Smaller platforms rarely build all of this themselves. They often join hash-sharing programs such as GIFCT's, which does not dictate how members enforce their own rules, use open-source fingerprinting tools like PDQ and TMK+PDQF, and buy classification from vendors. Livestreams add a time constraint, since harm can occur before review, so platforms combine real-time classifiers with rapid escalation paths. The central design choice is where to set thresholds: lower thresholds catch more violations but remove more legitimate speech, and the right balance differs by harm type, which is why many platforms publish transparency reports so outsiders can examine the trade-offs.

04Urugero rwakazi

Gusoma urubuga rwo gukorera mu mucyo kunenga

  1. Fungura YouTube ya AB 587 raporo yo mu Kwakira kugeza Ukuboza 2023 hanyuma usange umubare wamashusho yakuweho, 9,012,232.

  2. Shakisha imbonerahamwe yo kuvanaho isoko yambere yo kumenya hanyuma umenye ko ibendera ryikora ryikora rifite 96.1%.

  3. Shakisha imbonerahamwe yo kuvanaho ibitekerezo hanyuma wongereho 0-reba na 1-10-reba imigabane, 51.51% wongeyeho 26.43%.

  4. Reba icyo raporo itagaragaza, nkukuntu umubare wimurwa wibeshye cyangwa wahinduwe mubujurire.

  5. Gereranya nicyifuzo cya 2021 cyubuyobozi bushinzwe kugenzura ko urubuga rutangaza umubare wimikorere ikuweho nyuma yo gusuzuma abantu.

  6. Andika ibibazo bijyanye nukuri raporo isubizwa.

Ibyo byerekana

Hafi ya 77.9% ya videwo yakuweho yari ifite ibitekerezo 10 cyangwa bike, byerekana automatike ikora hakiri kare, ariko raporo yonyine ntishobora kwerekana inshuro byari byiza. Reba ibyo wasomye uhuza buri mubare na label yimbonerahamwe muri raporo.

Gushyira mu bikorwa Isi

Raporo ya YouTube yo mu Kwakira kugeza Ukuboza 2023 yerekana videwo 9,012.232 zavanyweho kubera kurenga ku Mabwiriza y’Umuryango, 96.1% zagaragaye bwa mbere n’ibendera ryikora, naho 51.51% zavanyweho mbere yo kubona igitekerezo kimwe.

Nyuma y’igitero cyo muri Werurwe 2019 Christchurch, Facebook yahagaritse kopi zirenga miliyoni 1.2 za videwo y’igitero cyoherejwe mu masaha 24, ikoresheje amashusho n’amajwi bihuye n’ibisobanuro birenga 800 byahinduwe.

Ububikoshingiro bwa HIF-Sharing Database butuma ibigo byabanyamuryango bisangira ibitekerezo byumvikana nka PDQ na TMK byerekana amashusho na videwo by’iterabwoba n’iterabwoba, bityo urubuga rumwe rukaba rushobora gufasha abandi guhagarika kohereza.

Facebook yafunguye-ifoto yayo ya PDQ hamwe na TMK + PDQF yerekana amashusho ya algorithms muri Kanama 2019, bityo andi masosiyete n’imiryango idaharanira inyungu ishobora gutunga urutoki ikoreshwa ry’abana, poropagande y’iterabwoba n’ihohoterwa rishingiye ku gitsina.

Ubujurire nubugenzuzi: nyuma yuko icyiciro cya Instagram cyakuyeho nabi poste yo kumenyekanisha kanseri yamabere mumwaka wa 2020, Ikigo gishinzwe kugenzura ibikorwa bya Meta cyasabye kubwira abakoresha igihe automatike ikora no kohereza abantu bambaye ubusa bambaye ubusa kugirango basuzume abantu mubujurire.

Kazoza ka AI muburyo bwo kwerekana amashusho

Several directions are visible in public sources. Hash-sharing has broadened in scope, with GIFCT's database expanding in 2019 to cover recorded attacks and later to manifestos and URLs, and open-source fingerprinting tools lowering the barrier for smaller platforms. Calls for transparency about automation, such as the Oversight Board's 2021 recommendations to tell users when automation acted and to publish how often automated removals are reversed, push platforms to measure errors as well as removals. Open problems remain hard. Rare, novel events like the 2019 Christchurch livestream still provide few training examples, context-dependent material such as news, satire and medical information remains error-prone, and performance across languages and regions is uneven. Synthetic video made with generative AI raises further questions about how detection and labeling should work alongside moderation. How regulators, platforms and independent researchers will measure moderation quality, rather than just volume, is an open question, and published transparency reports are the best current evidence for judging progress.

Kubishyira mubikorwa

Impamvu ari ngombwa

  • Kubaka amahitamo. Igishushanyo-cy-urwego rugena niba AI itezimbere ibisubizo nyabyo.
  • Itsinda hamwe nakazi. Guhuza ibikorwa byiza bikora umusaruro wunguka abakoresha bashobora kwizera.
  • Ibyago n'umutekano. Gukoresha neza ibibazo bigabanya umunaniro wimpinduka hamwe ningaruka zo gushyira mubikorwa.

Witondere

  • Gutangiza inzira yamenetse birashobora kongera ibibazo bihari.
  • Amakipe arashobora gukora cyane kandi agakuraho ibitekerezo byabantu bikenewe.
  • Ubwiza burashobora gutemba niba ibisubizo bidahwema gusuzumwa.

Intambwe ikurikira

  1. Shushanya ibikorwa byubu hanyuma umenye intambwe-yo guterana hejuru.
  2. Sobanura aho abantu bagenzura mbere yo kwikora byuzuye.
  3. Hugura abakoresha kubisobanuro, inzira zo kuzamuka, hamwe nubuziranenge.
  4. Kurikirana ibisubizo-urwego rwibisubizo kugirango wemeze agaciro karambye.

Inkomoko no gusoma

  1. California Department of Justice (filed by YouTube)YouTube Q4 2023 Raporo, Californiya AB 587
  2. Meta NewsroomA Further Update on New Zealand Terrorist Attack
  3. Meta NewsroomGufungura-Sourcing Ifoto- na Video-Guhuza Ikoranabuhanga kugirango Internet itekane
  4. Global Internet Forum to Counter TerrorismHash-Kugabana Ububikoshingiro
  5. Oversight BoardIbimenyetso bya kanseri y'ibere no kwambara ubusa (2020-004-IG-UA)
  6. YouTubeAmabwiriza ya YouTube
  7. Meta (GitHub)PDQ: perceptual image hashing (ThreatExchange)
  8. Meta (GitHub)TMK+PDQF: video hashing (ThreatExchange)

Komeza Ubushakashatsi

Ibaruwa y'amakuru y'ubuntu

Ibuka amakuru ya buri munsi ya AI

Inkuru eshatu zizewe za AI buri gitondo cy'iminsi y'akazi, zanditse mu rurimi rworoshye. Ubuntu iteka ryose, nta kwamamaza.

Imeri imwe buri munsi y'akazi. Kwiyimura mu buryo bumwe. Ntitwigera tugurisha cyangwa ngo dusangire aderesi yawe.

Isuzume

Fata AI mubibazo bya Video Ibirimo

Ibisubizo ako kanya kuri buri kibazo, n'icyemezo gishobora gusangirwa gifite ID yagenzurwa umaze gutsinda isomo.

Tangira ikibazo

Fasha uburezi bwa AI bw'ubuntu. AI Understanding ni umuryango utegamiye kuri leta wa 501(c)(3) — nta kwamamaza, nta kwishyura, iteka ryose. Tanga impano

Ibibazo bikunze kubazwa

Niki AI muburyo bwo kwerekana amashusho?

Kugereranya ibice bya videwo ya AI ikoresha urutoki ruhuza hamwe no kwiga imashini kugirango ibone amashusho yoherejwe cyangwa yerekanwe kumurongo wica amategeko ya platform, mubisanzwe mbere yuko abantu babibona. Bifite akamaro kubera igipimo: YouTube ivuga ko amasaha arenga 500 ya videwo agera buri munota, kandi mu mpera za 2023 sisitemu zikoresha zabanje kwerekana 96.1% ya videwo yakuyeho.

Kuki kugereranya amashusho bisobanurwa ngo 'multimodal'?

Video ihuza ubwoko butandukanye bwibimenyetso, kuburyo bugereranije bukenera icyerekezo, isesengura ryimikorere, imvugo-ku-nyandiko, na OCR ikorera hamwe.

Nigute gushishoza gushishoza byafasha urubuga kumenya kongera kohereza amashusho azwi?

Amashusho yerekana amashusho yerekana urutoki arashobora kugereranwa nintoki yibikoresho bizwi kugirango umenye kopi hafi-ebyiri; ntigaragaza ibintu bishya byangiza byonyine.

Ni ukubera iki urubuga rukomeza inzira nyinshi kubasuzuma abantu?

AI ikurikirana kandi itanga amanota y'ibirimo, ariko imanza zidasobanutse zikeneye guca imanza kubijyanye n'imigambi n'imiterere byiyongera kubantu.

Nigute imvugo-ku-nyandiko igira uruhare mu gushyira mu gaciro?

Mu mashusho yerekana amashusho, kumenyekanisha imvugo birashobora gutanga inyandiko-mvugo ivuye mu majwi, abatondekanya inyandiko bashobora noneho gusuzuma ibimenyetso bya politiki nko kuvuga urwango cyangwa iterabwoba.

Niki igikumwe cya videwo gikomeye gishobora gufasha kumenya nyuma yo gushushanya cyangwa guhindura amashusho?

Amashusho yerekana amashusho yashizweho kugirango ashyigikire guhuza kopi zisa nkizishobora kubikwa; gukomera biterwa nuburyo bwihariye no guhinduka.