{"CACHEDAT":"2026-05-13 09:35:33","TRANSLATEDAT":"2026-05-13 09:35:33","SOURCESIGNATURE":"7a9e56d2ea3f2c69597ecdc4c37790c3bd22ac955bbb62f6ceb193688b397de6","SLUG":"misinformation-manipulation-WtiyDFyWdl","MARKDOWN":"# Information Disorder\n\nWardle & Derakhshan (2017), in a foundational report for the Council of Europe, introduce *Information Disorder* as the umbrella term for false, misleading, or harmfully shared information in the public information environment. They distinguish three categories along two axes: whether the content is *false* or *true*, and whether it is shared with *intent to harm*.\n\n| Type | Content | Sharer's knowledge of falseness | Intent to harm |\n|------|---------|---------------------------------|----------------|\n| **Misinformation** | false or misleading | does not know | no |\n| **Disinformation** | false or misleading | knows | yes |\n| **Malinformation** | true | (not the criterion) | yes |\n\nThe three categories form an interconnected ecology: the same content item can move between them depending on who shares it and with what intent.\n\n### Fake News\n\nThe term *fake news* is widely used in public debate but analytically imprecise. It usually refers to false or misleading information presented in the style of news reporting. However, because the term can refer to different types of information disorder and is often used politically to discredit unwanted reporting, more precise terms such as *misinformation*, *disinformation*, *fabricated content*, *false context*, or *manipulated content* should be preferred.\n\n## Misinformation\n\nMisinformation is false or misleading information shared by people who do not realise it is false. The sharer believes the content is accurate; there is no intent to deceive or harm.\n\nMisinformation typically arises in fast-moving news situations (early reports with unconfirmed details), in passing along claims without checking accuracy, and in repeating information from trusted sources that turn out to be wrong. It can be as widespread and consequential as disinformation, despite the absence of malicious intent — its sharers are often genuinely motivated and trusted within their networks, which gives the content reach and perceived credibility (Lewandowsky et al., 2017).\n\n### Misinformation Susceptibility Factors\n\n\n:::success\n**1. Cognitive factors** — how information is processed\n\n* **Confirmation bias**: People pay more attention to information that supports what they already believe, and scrutinise opposing information more harshly. — Nickerson (1998)\n* **Illusory truth effect**: Claims can feel more believable simply because people have seen them before, even when they are false. — Unkelbach et al. (2019)\n* **Processing fluency**: Information that is easy to read, familiar in wording, or visually clear feels more trustworthy than information that is harder to process. — Reber & Schwarz (1999)\n* **Low cognitive reflection**: People who rely on quick, intuitive judgements rather than pausing to check are more likely to fall for false claims. — Pennycook & Rand (2019)\n* **Bullshit receptivity**: Some people are inclined to find vague but impressive-sounding statements meaningful, even when they are empty. — Pennycook et al. (2015)\n* **Low numeracy / scientific literacy**: People may struggle to evaluate statistics, graphs, risk statements, or scientific evidence; education alone is no automatic protection. — Kahan et al. (2017)\n\n**2. Affective / motivational factors** — how emotions and identity shape judgement\n\n* **Emotional reasoning / reliance on emotion**: People accept claims because they feel right, frightening, satisfying, or morally urgent, rather than because they have been checked. — Martel, Pennycook & Rand (2020)\n* **Specific emotions** (anger, fear, moral outrage): Strong emotions can lead people to react quickly and share before checking; moral-emotional language is especially powerful in spreading political content. (Belief and sharing are distinct effects with different evidence.) — Brady et al. (2017); Martel et al. (2020)\n* **Identity-protective cognition**: People process information in ways that protect their political, religious, or group identity, and resist evidence that would threaten it. (Closely linked to *political identity* in category 5.) — Kahan (2013, 2017)\n* **Conspiracy mindset**: A general tendency to suspect hidden plots or powerful actors behind events makes conspiracy-style misinformation more plausible. — Douglas, Sutton & Cichocka (2017)\n\n**3. Social factors** — whom and which groups people trust\n\n* **Source credibility / trust in the sender**: How believable a claim feels depends on who appears to be saying it — a friend, an influencer, an expert, a politician, a news outlet, or an anonymous account. — Traberg & van der Linden (2022)\n* **Social proof / popularity cues**: Likes, shares, views, and comments can act as credibility cues, although the effect depends on context. — Avram et al. (2020)\n* **In-group / partisan congruency**: Information from \"people like us\", or aligned with a group's position, is accepted more easily. (Closely linked to *political identity* in category 5.) — Sultan et al. (2024)\n* **Network homophily / echo chambers**: When people mostly encounter information through networks of similar others, certain views appear more widely shared than they really are. (See also D1G5.) — Cinelli et al. (2021)\n\n**4. Contextual / situational factors** — the conditions under which people meet information\n\n* **Time pressure**: When people have to judge information quickly, they distinguish true from false claims less accurately. — Sultan et al. (2022)\n* **Information overload**: When too much information arrives at once, people fall back on shortcuts like headlines, emotion, source labels, or popularity. — Laato et al. (2020)\n* **Distraction / cognitive load**: When attention is drawn elsewhere, people may share without checking accuracy, even if they could otherwise recognise the misinformation. — Pennycook et al. (2020)\n* **Crisis / uncertainty contexts** (pandemics, war, disasters): Urgent need for explanation increases openness to false or premature claims. — Roozenbeek et al. (2020)\n* **Platform design**: Feeds, notifications, recommendation systems, autoplay, and fast scrolling encourage reactive rather than reflective behaviour. (See also D1G1.) — Lorenz-Spreen et al. (2020)\n\n**5. Background and dispositional factors** — who may be more or less vulnerable in specific contexts\n\n* **Age**: Older adults sometimes distinguish true from false headlines better in studies, but share misinformation more often online. — Guess, Nagler & Tucker (2019)\n* **Education**: Not an automatic shield; meta-analyses show no simple effect on misinformation discrimination. — Sultan et al. (2024)\n* **Political identity and prior beliefs**: Shape which information feels plausible, threatening, trustworthy, or worth sharing. (Operates via identity-protective cognition in category 2 and in-group cues in category 3 — not purely a demographic variable.) — Sultan et al. (2024)\n* **Media literacy habits**: People who compare sources, read beyond headlines, and recognise manipulation techniques rely less on emotional or social shortcuts. *(Note: this is the learning goal of SciLMi itself — listed here as a documented protective disposition, not as a prerequisite.)* — Guess et al. (2020)\n\n:::\n\n\n:::info\n* Avram, M., Micallef, N., Patil, S., & Menczer, F. (2020). Exposure to social engagement metrics increases vulnerability to misinformation. *Harvard Kennedy School (HKS) Misinformation Review*, *1*(5). \n* Brady, W. J., Wills, J. A., Jost, J. T., Tucker, J. A., & Van Bavel, J. J. (2017). Emotion shapes the diffusion of moralized content in social networks. *Proceedings of the National Academy of Sciences*, *114*(28), 7313–7318. \n* Cinelli, M., De Francisci Morales, G., Galeazzi, A., Quattrociocchi, W., & Starnini, M. (2021). The echo chamber effect on social media. *Proceedings of the National Academy of Sciences*, *118*(9), e2023301118. \n* Douglas, K. M., Sutton, R. M., & Cichocka, A. (2017). The psychology of conspiracy theories. *Current Directions in Psychological Science*, *26*(6), 538–542. \n* Guess, A., Nagler, J., & Tucker, J. (2019). Less than you think: Prevalence and predictors of fake news dissemination on Facebook. *Science Advances*, *5*(1), eaau4586. \n* Jones-Jang, S. M., Mortensen, T., & Liu, J. (2021). Does media literacy help identification of fake news? Information literacy helps, but other literacies don't. *American Behavioral Scientist*, *65*(2), 371–388. \n* Kahan, D. M. (2013). Ideology, motivated reasoning, and cognitive reflection. *Judgment and Decision Making*, *8*(4), 407–424.\n* Kahan, D. M., Peters, E., Dawson, E. C., & Slovic, P. (2017). Motivated numeracy and enlightened self-government. *Behavioural Public Policy*, *1*(1), 54–86. \n* Laato, S., Islam, A. K. M. N., Islam, M. N., & Whelan, E. (2020). What drives unverified information sharing and cyberchondria during the COVID-19 pandemic? *European Journal of Information Systems*, *29*(3), 288–305. \n* Lorenz-Spreen, P., Lewandowsky, S., Sunstein, C. R., & Hertwig, R. (2020). How behavioural sciences can promote truth, autonomy and democratic discourse online. *Nature Human Behaviour*, *4*(11), 1102–1109. \n* Martel, C., Pennycook, G., & Rand, D. G. (2020). Reliance on emotion promotes belief in fake news. *Cognitive Research: Principles and Implications*, *5*(1), 47. \n* Nickerson, R. S. (1998). Confirmation bias: A ubiquitous phenomenon in many guises. *Review of General Psychology*, *2*(2), 175–220. \n* Pennycook, G., Cheyne, J. A., Barr, N., Koehler, D. J., & Fugelsang, J. A. (2015). On the reception and detection of pseudo-profound bullshit. *Judgment and Decision Making*, *10*(6), 549–563.\n* Pennycook, G., McPhetres, J., Zhang, Y., Lu, J. G., & Rand, D. G. (2020). Fighting COVID-19 misinformation on social media: Experimental evidence for a scalable accuracy-nudge intervention. *Psychological Science*, *31*(7), 770–780. \n* Pennycook, G., & Rand, D. G. (2019). Lazy, not biased: Susceptibility to partisan fake news is better explained by lack of reasoning than by motivated reasoning. *Cognition*, *188*, 39–50. \n* Reber, R., & Schwarz, N. (1999). Effects of perceptual fluency on judgments of truth. *Consciousness and Cognition*, *8*(3), 338–342. \n* Roozenbeek, J., Schneider, C. R., Dryhurst, S., Kerr, J., Freeman, A. L. J., Recchia, G., van der Bles, A. M., & van der Linden, S. (2020). Susceptibility to misinformation about COVID-19 around the world. *Royal Society Open Science*, *7*(10), 201199. \n* Sultan, M., Tump, A. N., Geers, M., Lorenz-Spreen, P., Herzog, S. M., & Kurvers, R. H. J. M. (2022). Time pressure reduces misinformation discrimination ability but does not alter response bias. *Scientific Reports*, *12*(1), 22416. \n* Sultan, M., Tump, A. N., Ehmann, N., Lorenz-Spreen, P., Hertwig, R., Gollwitzer, A., & Kurvers, R. H. J. M. (2024). Susceptibility to online misinformation: A systematic meta-analysis of demographic and psychological factors. *Proceedings of the National Academy of Sciences*, *121*(47), e2409329121. \n* Traberg, C. S., & van der Linden, S. (2022). Birds of a feather are persuaded together: Perceived source credibility mediates the effect of political bias on misinformation susceptibility. *Personality and Individual Differences*, *185*, 111269. \n* Unkelbach, C., Koch, A., Silva, R. R., & Garcia-Marques, T. (2019). Truth by repetition: Explanations and implications. *Current Directions in Psychological Science*, *28*(3), 247–253. \n\n:::\n\n### ☑ Logical Fallacies\n\n\n:::success\n- [ ] Check for hasty generalisations - Conclusion from too little evidence.\n- [ ] Check for false dilemmas - Limiting options to two when more exist.\n- [ ] Check for straw man arguments - Misrepresenting a position to refute it easily.\n- [ ] Check for appeals to ignorance - Claiming truth due to lack of disproof.\n- [ ] Check for appeals to authority - Assuming truth based on authority alone.\n- [ ] Check for red herrings - Distracting from the main issue.\n- [ ] Check for false causes - Confusing correlation with causation.\n- [ ] Check for ad hominem - Attacking the person, not the argument. \n- [ ] Check for ad populum - Arguing truth from popularity.\n- [ ] Check for slippery slopes - Asserting one step leads to extremes.\n- [ ] Check for circular reasoning - Using the conclusion as a premise.\n\n:::\n\n\n:::tip\n* \n* \n* \n\n:::\n\n## Disinformation\n\nDisinformation is false or misleading information shared deliberately, by people who know it is false, with the intent to harm, deceive, or manipulate. Both the sharer's awareness of the falseness and the harmful intent are definitional features.\n\nCommon forms include political influence operations, commercial deception, propaganda, fabricated content, and manipulated media. Disinformation is often produced by organised actors and amplified through *Coordinated Inauthentic / Artificial Amplification* (above).\n\n## Malinformation\n\nMalinformation is *true* information shared with intent to harm. The content itself is accurate, but its release, framing, or timing is calculated to damage a person, group, or institution. The classical pattern is the deliberate movement of private, sensitive, or context-bound information into a public or harmful context.\n\nExamples include the publication of leaked private communications to discredit a target, the release of accurate but stigmatising personal data (e.g. revenge porn), and the strategic disclosure of factually correct but contextually damaging information at moments calculated for maximum impact.\n\nMalinformation is the least-discussed of the three categories because the content is not false, which places it outside fact-checking frameworks. But it forms a significant part of the information-disorder ecology (Wardle & Derakhshan, 2017).\n\n\n:::warning\n**The boundaries are porous.**\n\nThe three categories often overlap or shift in practice:\n\n* A piece of *disinformation*, once shared by people who genuinely believe it, becomes *misinformation* as it spreads further. The same content can sit in different categories depending on who shares it.\n* Content can be partly true and partly false. Wardle (2017) lists seven forms within information disorder — including *misleading content*, *false context*, and *manipulated content* — not all false content is fully fabricated.\n* Determining intent from outside is empirically hard. Without access to the sharer's knowledge state and motivation, the line between misinformation and disinformation often cannot be drawn definitively.\n\nThis is why the framework asks learners to *explain why the boundary between misinformation and disinformation is often difficult to determine in practice*.\n\n:::\n\n\n:::info\n* Wardle, C., & Derakhshan, H. (2017). *Information Disorder: Toward an interdisciplinary framework for research and policy making*. Council of Europe Report DGI(2017)09. \n* Wardle, C. (2017). Fake news. It's complicated. *First Draft News*. \n* Lewandowsky, S., Ecker, U. K. H., & Cook, J. (2017). Beyond misinformation: Understanding and coping with the \"post-truth\" era. *Journal of Applied Research in Memory and Cognition*, 6(4), 353–369. \n* Vraga, E. K., & Bode, L. (2020). Defining misinformation and understanding its bounded nature. *Political Communication*, 37(1), 136–144. \n\n:::\n\n# AI Hallucination / Confabulation\n\nThe systematic tendency of generative AI systems to produce plausible-sounding content that is factually incorrect, fabricated, or unverifiable — including invented references, non-existent studies, misattributed quotations, made-up statistics, and false biographical, historical, or scientific details.\n\nHallucinations occur because generative AI systems produce output by predicting plausible continuations from patterns learned during training, not by retrieving verified information from a knowledge source. The system optimises for fluency and plausibility, not for accuracy. Fabricated content is typically presented with the same confidence as accurate content — there is no internal signal that reliably distinguishes the two.\n\nThe term **confabulation** is sometimes preferred over **hallucination** in research contexts, since confabulation in psychology denotes the construction of false but sincerely believed accounts without intent to deceive — closer to what generative systems actually do. *Hallucination* remains the dominant term in public and technical discourse and is retained here.\n\n**Hallucinations are not occasional bugs.** They are a structural feature of the underlying generation process, not errors that can be reliably eliminated through better prompting (Kalai & Vempala, 2024). They occur across all generative systems — including ☑ RAG AI, where fabrication can also affect how retrieved sources are summarised, quoted, or attributed.\n\n**Didactic positioning:** Hallucinations can be located conceptually as a forth category alongside *misinformation, disinformation*, and *malinformation*. Unlike both, they are not produced by a human source with intent or error in mind, but emerge from the generation process itself — system-generated false content without an intentional source. This three-way distinction is offered here as a pedagogically useful framing rather than as established consensus terminology.\n\n**Common types:**\n\n* fabricated references — citations to articles, books, or studies that do not exist\n* misattributed or fabricated quotations\n* invented statistics or numerical claims\n* inaccurate biographical, historical, or scientific details presented confidently\n* confidently incorrect summaries of real sources\n\n\n:::tip\nTreat any factual claim from a generative AI system as unverified until checked against an independent source. References and quotations are particularly easy to verify — search whether the cited source exists and contains what the AI claimed.\n\n:::\n\n\n:::info\n* Bender, E. M., Gebru, T., McMillan-Major, A., & Shmitchell, S. (2021). On the dangers of stochastic parrots: Can language models be too big? *Proceedings of the 2021 ACM Conference on Fairness, Accountability, and Transparency (FAccT '21)*, 610–623. \n* Ji, Z., Lee, N., Frieske, R., Yu, T., Su, D., Xu, Y., Ishii, E., Bang, Y. J., Madotto, A., & Fung, P. (2023). Survey of hallucination in natural language generation. *ACM Computing Surveys, 55*(12), Article 248. \n* Kalai, A. T., & Vempala, S. S. (2024). Calibrated language models must hallucinate. *Proceedings of the 56th Annual ACM Symposium on Theory of Computing (STOC 2024)*, 160–171. \n* Smith, A. L., Greaves, F., & Panch, T. (2023). Hallucination or confabulation? Neuroanatomy as metaphor in large language models. *PLOS Digital Health, 2*(11), e0000388. \n\n:::\n\n# Content Misrepresentation\n\n| Format | Manipulated | Fabricated |\n|--------|-------------|------------|\n| Text | genuine quote stripped of context, doctored headline | invented quote, fictional article, AI-generated text |\n| Image | retouched photograph, cropped original | AI-generated image, painted fake \"photograph\" |\n| Audio | edited or sped-up original recording | voice clone, AI-generated speech |\n| Video | cheap fake, recut clips | fully synthetic deepfake video |\n| Document | forged letterhead on modified template | invented \"official\" letter |\n\n## Fabricated Content\n\n**Fabricated content** is information — text, image, audio, video, document, or other format — that has been entirely invented, with no genuine source material. It may be produced manually (invented quotes, fictional news articles) or generated synthetically by AI systems (synthetic videos, voice clones, AI-generated images, AI-written text).\n\n## Manipulated Content\n\n**Manipulated content** is geniune information — text, image, audio, video, document, or other format — that has been altered to deceive. The source material is real, but it is modified through editing, selective cropping, speed adjustment, recontextualisation, framing changes, voice substitution, or other transformations that change its meaning or apparent context.\n\n### Cheap Fake\n\nA **cheap fake** is media that has been altered using conventional, widely available tools — image editing software, video editing software, or basic recontextualisation — rather than AI-based generative methods. The term was coined by Paris and Donovan (2019) to draw attention to the fact that the most consequential forms of manipulated media in public discourse are typically not sophisticated AI-generated deepfakes, but simple, low-cost techniques that anyone can apply.\n\nCommon cheap fake techniques include:\n\n* speed manipulation (e.g. the slowed-down Pelosi video, 2019)\n* recutting and selective editing of audio or video\n* recontextualisation (real media presented with a false caption or framing)\n* photo retouching, cropping, or compositing\n* swapped captions and doctored screenshots\n* audio splicing\n\nWhether a cheap fake constitutes *Misinformation*, *Disinformation*, or *Malinformation* depends on the actor's knowledge and intent (→ *Information Disorder*), not on the technique itself. The same edited video can serve satire, fiction, education, or deception.\n\nCompared with *Deep Fakes*, cheap fakes require less technical skill but are not necessarily easier to detect — well-executed recontextualisation or selective editing can be extremely difficult to identify without access to the original source material.\n\n\n:::info\n* Paris, B., & Donovan, J. (2019). *Deepfakes and Cheap Fakes: The Manipulation of Audio and Visual Evidence*. Data & Society Research Institute. \n\n:::\n\n\n:::tip\nCheap Fake Verification Tools: \n\n* \n* \n* \n* \n* \n* \n* \n* \n* \n\n:::\n\n### Deep Fake\n\n### Deep Fake\n\nA **deep fake** is synthetic media generated by artificial intelligence — specifically using deep learning techniques such as generative adversarial networks (GANs), diffusion models, or transformer-based generators. The term is a portmanteau of *deep learning* and *fake*, originating in 2017 in online communities producing AI-generated face-swap videos. The technology has since expanded across media types and is now broadly accessible through consumer-level tools.\n\nCommon forms include:\n\n* **Face swaps**: replacing one person's face with another's in video\n* **Voice cloning**: synthesising a speaker's voice from a small audio sample\n* **Full synthetic video**: AI-generated footage showing people, events, or scenes that never occurred\n* **Synthetic photography**: AI-generated still images of fictional people or events\n* **Text-to-video and text-to-audio**: generating media from written prompts\n\nWhether a deep fake constitutes *Misinformation*, *Disinformation*, or *Malinformation* depends on the actor's knowledge and intent (→ *Information Disorder*), not on the technique itself. Deep fake methods are also used legitimately in film production, accessibility tools (synthetic voices for people who have lost theirs), language dubbing, satire, education, and the arts.\n\nCompared with *Cheap Fakes*, deep fakes require more technical capacity to produce convincingly and can achieve higher visual or auditory realism. They are not, however, the dominant form of media manipulation in public discourse — Paris and Donovan (2019) note that simpler cheap-fake techniques remain more common and often more consequential. Detection is an active research area; current approaches combine artefact analysis, biological inconsistency checks, and provenance verification (e.g. C2PA content credentials).\n\n\n:::info\n* Paris, B., & Donovan, J. (2019). *Deepfakes and Cheap Fakes: The Manipulation of Audio and Visual Evidence*. Data & Society Research Institute. \n* Chesney, R., & Citron, D. K. (2019). Deep fakes: A looming challenge for privacy, democracy, and national security. *California Law Review*, 107(6), 1753–1820.\n* Vaccari, C., & Chadwick, A. (2020). Deepfakes and disinformation: Exploring the impact of synthetic political video on deception, uncertainty, and trust in news. *Social Media + Society*, 6(1), 1–13. \n\n:::\n\n\n:::tip\nDeep Fake Verification Tools: \n\n* \n* \n* \n* \n* \n* \n* \n* \n* \n\n:::\n\n\n##","HTML":"

Informācijas haoss

\n

Wardle & Derakhshan (2017) Eiropas Padomes pamatziņojumā ievieš terminu „informācijas haoss” kā vispārēju apzīmējumu nepatiesai, maldinošai vai kaitīgi izplatītai informācijai sabiedriskajā informācijas vidē. Viņi izšķir trīs kategorijas, balstoties uz diviem kritērijiem: vai saturs ir nepaties vai paties, un vai tas tiek izplatīts ar nolūku nodarīt kaitējumu.

\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n
VeidsSatursDalītāja zināšanas par nepatiesībuNodoms nodarīt kaitējumu
Dezinformācijanepatiesa vai maldinošanezina
Dezinformācijanepatiesa vai maldinošazina
Nepareiza informācijapatiesa(nav kritērijs)
\n

Šīs trīs kategorijas veido savstarpēji saistītu ekosistēmu: viens un tas pats saturs var pārvietoties starp tām atkarībā no tā, kas to dala un ar kādu nolūku.

\n

Viltus ziņas

\n

Termins „viltus ziņas” tiek plaši lietots publiskās debatēs, taču analītiski tas ir neprecīzs. Parasti tas attiecas uz nepatiesu vai maldinošu informāciju, kas pasniegta ziņu reportāžu stilā. Tomēr, tā kā šis termins var attiekties uz dažādiem informācijas traucējumu veidiem un bieži tiek izmantots politiskos nolūkos, lai diskreditētu nevēlamas reportāžas, būtu jāizmanto precīzāki termini, piemēram, „nepatiesa informācija”, „dezinformācija”, „izdomāts saturs”, „nepatiesais konteksts” vai „manipulēts saturs”.

\n

Nepareiza informācija

\n

Nepareiza informācija ir nepatiesa vai maldinoša informācija, ko izplata cilvēki, kuri neapzinās, ka tā ir nepatiesa. Izplatītājs uzskata, ka saturs ir precīzs; nav nolūka maldināt vai nodarīt kaitējumu.

\n

Nepareiza informācija parasti rodas strauji mainīgās ziņu situācijās (sākotnējās ziņas ar neapstiprinātiem faktiem), pārsūtot apgalvojumus, nepārbaudot to pareizību, un atkārtojot informāciju no uzticamiem avotiem, kas izrādās nepareiza. Tā var būt tikpat izplatīta un ietekmīga kā dezinformācija, neskatoties uz ļaunprātīga nolūka trūkumu — tās izplatītāji bieži vien ir patiesi motivēti un uzticami savās tīklos, kas nodrošina satura sasniedzamību un uztveramo ticamību (Lewandowsky et al., 2017).

\n

Faktori, kas veicina nepatiesas informācijas uztveršanu

\n
\n
\n\n

1. Kognitīvie faktori — kā tiek apstrādāta informācija

\n
    \n
  • Apstiprinājuma tendence: cilvēki pievērš lielāku uzmanību informācijai, kas atbalsta to, kam viņi jau tic, un pretēju informāciju izvērtē daudz kritiskāk. — Nickerson (1998)\n
  • \n
  • Iluzorās patiesības efekts: apgalvojumi var šķist ticamāki vienkārši tāpēc, ka cilvēki tos ir redzējuši iepriekš, pat ja tie ir nepatiesi. — Unkelbach et al. (2019)\n
  • \n
  • Apstrādes vieglums: informācija, kas ir viegli lasāma, pazīstama formulējumā vai vizuāli skaidra, šķiet uzticamāka nekā informācija, kuru ir grūtāk apstrādāt. — Reber & Schwarz (1999)\n
  • \n
  • Zems kognitīvais refleksijas līmenis: cilvēki, kuri paļaujas uz ātriem, intuitīviem spriedumiem, nevis apstājas, lai pārbaudītu, ir vairāk pakļauti riskam noticēt nepatiesiem apgalvojumiem. — Pennycook & Rand (2019)\n
  • \n
  • Uzņēmība pret muļķībām: Daži cilvēki ir noslieci uzskatīt neskaidrus, bet iespaidīgi skanošus apgalvojumus par nozīmīgiem, pat ja tie ir tukši. — Pennycook et al. (2015)\n
  • \n
  • Zema matemātiskā / zinātniskā kompetence: cilvēkiem var būt grūti izvērtēt statistiku, grafikus, riska aprakstus vai zinātniskos pierādījumus; izglītība pati par sevi nav automātiska aizsardzība. — Kahan et al. (2017)\n

    2. Afektīvie / motivācijas faktori — kā emocijas un identitāte ietekmē spriedumu

    \n
  • \n
  • Emocionāla loģika / paļaušanās uz emocijām: Cilvēki pieņem apgalvojumus, jo tie šķiet pareizi, biedējoši, apmierinoši vai morāli steidzami, nevis tāpēc, ka tie ir pārbaudīti. — Martel, Pennycook & Rand (2020)\n
  • \n
  • Konkrētas emocijas (dusmas, bailes, morāls sašutums): Spēcīgas emocijas var likt cilvēkiem reaģēt ātri un dalīties ar informāciju, pirms tā ir pārbaudīta; morāli emocionāla valoda ir īpaši spēcīga politiska satura izplatīšanā. (Pārliecība un dalīšanās ir atšķirīgi efekti ar atšķirīgiem pierādījumiem.) — Brady et al. (2017); Martel et al. (2020)\n
  • \n
  • Identitāti aizsargājoša kognīcija: Cilvēki apstrādā informāciju tādā veidā, kas aizsargā viņu politisko, reliģisko vai grupas identitāti, un pretojas pierādījumiem, kas to apdraudētu. (Cieši saistīts ar politisko identitāti 5. kategorijā.) — Kahan (2013, 2017)\n
  • \n
  • Sazvērestības domāšana: vispārēja tendence aizdomāties par slēptiem plāniem vai ietekmīgiem dalībniekiem notikumu aizkulisēs padara sazvērestības tipa dezinformāciju ticamāku. — Douglas, Sutton & Cichocka (2017)\n

    3. Sociālie faktori — kam un kurām grupām cilvēki uzticas

    \n
  • \n
  • Avota uzticamība / uzticēšanās sūtītājam: Tas, cik ticams šķiet apgalvojums, ir atkarīgs no tā, kas to saka — draugs, ietekmīgā persona, eksperts, politiķis, ziņu aģentūra vai anonīms konts. — Traberg & van der Linden (2022)\n
  • \n
  • Sociālais pierādījums / popularitātes pazīmes: „Patīk”, „dalīties”, skatījumi un komentāri var kalpot kā uzticamības pazīmes, lai gan to ietekme ir atkarīga no konteksta. — Avram et al. (2020)\n
  • \n
  • Piederība grupai / partijas vienotība: informācija no „cilvēkiem, kas ir tādi kā mēs”, vai kas atbilst grupas nostājai, tiek pieņemta vieglāk. (Cieši saistīts ar politisko identitāti 5. kategorijā.) — Sultan et al. (2024)\n
  • \n
  • Tīkla homofilija / rezonanses telpas: Kad cilvēki galvenokārt sastopas ar informāciju caur līdzīgu cilvēku tīkliem, noteikti viedokļi šķiet plašāk izplatīti, nekā tie patiesībā ir. (Skatīt arī D1G5.) — Cinelli et al. (2021)\n

    4. Kontekstuāli / situatīvi faktori — apstākļi, kādos cilvēki sastopas ar informāciju

    \n
  • \n
  • Laika spiediens: Kad cilvēkiem informācija jāizvērtē ātri, viņi mazāk precīzi atšķir patiesus apgalvojumus no nepatiesiem. — Sultan et al. (2022)\n
  • \n
  • Informācijas pārslodze: Kad vienlaikus tiek saņemts pārāk daudz informācijas, cilvēki izmanto īsākus ceļus, piemēram, virsrakstus, emocijas, avotu marķējumus vai popularitāti. — Laato et al. (2020)\n
  • \n
  • Uzmanības novirzīšana / kognitīvā slodze: ja uzmanība tiek novirzīta uz citu lietu, cilvēki var dalīties ar informāciju, ne pārbaudot tās pareizību, pat ja citādi viņi spētu atpazīt dezinformāciju. — Pennycook et al. (2020)\n
  • \n
  • Krīzes / nenoteiktības situācijas (pandēmijas, karš, katastrofas): steidzama vajadzība pēc skaidrojuma palielina atvērtību nepatiesiem vai pāragriem apgalvojumiem. — Roozenbeek et al. (2020)\n
  • \n
  • Platformas dizains: Ziņu plūsmas, paziņojumi, ieteikumu sistēmas, automātiskā atskaņošana un ātra ritināšana veicina reaģējošu, nevis pārdomātu uzvedību. (Skatīt arī D1G1.) — Lorenz-Spreen et al. (2020)\n

    5. Fona un rakstura faktori — kas konkrētos apstākļos var būt vairāk vai mazāk neaizsargāti

    \n
  • \n
  • Vecums: pētījumos vecāki cilvēki dažkārt labāk atšķir patiesas no nepatiesām virsrakstiem, bet tie biežāk izplata dezinformāciju tiešsaistē. — Guess, Nagler & Tucker (2019)\n
  • \n
  • Izglītība: nav automātisks aizsargājošs faktors; metaanalīzes liecina, ka tai nav vienkāršas ietekmes uz nepatiesas informācijas atšķiršanu. — Sultan et al. (2024)\n
  • \n
  • Politiskā identitāte un iepriekšējie uzskati: nosaka, kāda informācija šķiet ticama, draudīga, uzticama vai dalīšanās vērta. (Darbosies caur identitāti aizsargājošu kognīciju 2. kategorijā un grupas iekšējām norādēm 3. kategorijā — nav tikai demogrāfiskais mainīgais.) — Sultan et al. (2024)\n
  • \n
  • Mediju lietotprasmes ieradumi: cilvēki, kuri salīdzina avotus, lasa ne tikai virsrakstus un atpazīst manipulācijas paņēmienus, mazāk paļaujas uz emocionāliem vai sociāliem īsajiem ceļiem. (Piezīme: tas ir pats SciLMi mācību mērķis — šeit minēts kā dokumentēta aizsargājoša tendence, nevis kā priekšnoteikums.) — Guess et al. (2020)\n
  • \n
\n
\n
\n
\n
\n\n
    \n
  • \nAvram, M., Micallef, N., Patil, S., & Menczer, F. (2020). Saskare ar sociālās iesaistīšanās rādītājiem palielina neaizsargātību pret dezinformāciju. Harvard Kennedy School (HKS) Misinformation Review\n, 1\n(5). https://doi.org/10.37016/mr-2020-033\n
  • \n
  • \nBrady, W. J., Wills, J. A., Jost, J. T., Tucker, J. A., & Van Bavel, J. J. (2017). Emocijas ietekmē morāli orientēta satura izplatīšanos sociālajos tīklos. Proceedings of the National Academy of Sciences\n, 114\n(28), 7313–7318. https://doi.org/10.1073/pnas.1618923114\n
  • \n
  • \nCinelli, M., De Francisci Morales, G., Galeazzi, A., Quattrociocchi, W., & Starnini, M. (2021). Eho kameras efekts sociālajos tīklos. Nacionālās Zinātņu akadēmijas raksti\n, 118\n(9), e2023301118. https://doi.org/10.1073/pnas.2023301118\n
  • \n
  • \nDouglas, K. M., Sutton, R. M., & Cichocka, A. (2017). Sazvērestības teoriju psiholoģija. Current Directions in Psychological Science\n, 26\n(6), 538–542. https://doi.org/10.1177/0963721417718261\n
  • \n
  • \nGuess, A., Nagler, J., & Tucker, J. (2019). Mazāk, nekā jūs domājat: viltus ziņu izplatības izplatība un prognozējošie faktori Facebook. Science Advances\n, 5\n(1), eaau4586. https://doi.org/10.1126/sciadv.aau4586\n
  • \n
  • \nJones-Jang, S. M., Mortensen, T., & Liu, J. (2021). Vai mediju pratība palīdz atpazīt viltus ziņas? Informācijas pratība palīdz, bet citas prasmes — nē. American Behavioral Scientist\n, 65\n(2), 371–388. https://doi.org/10.1177/0002764219869406\n
  • \n
  • Kahan, D. M. (2013). Ideoloģija, motivēta loģika un kognitīvā refleksija. Judgment and Decision Making, 8(4), 407–424.\n
  • \n
  • \nKahan, D. M., Peters, E., Dawson, E. C., & Slovic, P. (2017). Motivēta skaitīšanas prasme un apgaismota pašpārvalde. Behavioural Public Policy\n, 1\n(1), 54–86. https://doi.org/10.1017/bpp.2016.2\n
  • \n
  • \nLaato, S., Islam, A. K. M. N., Islam, M. N., & Whelan, E. (2020). Kas veicina neapstiprinātas informācijas izplatīšanu un kiberhondriju COVID-19 pandēmijas laikā? European Journal of Information Systems\n, 29\n(3), 288–305. https://doi.org/10.1080/0960085X.2020.1770632\n
  • \n
  • \nLorenz-Spreen, P., Lewandowsky, S., Sunstein, C. R., & Hertwig, R. (2020). Kā uzvedības zinātnes var veicināt patiesību, autonomiju un demokrātisku diskusiju tiešsaistē. Nature Human Behaviour\n, 4\n(11), 1102–1109. https://doi.org/10.1038/s41562-020-0889-7\n
  • \n
  • \nMartel, C., Pennycook, G., & Rand, D. G. (2020). Paļaušanās uz emocijām veicina ticību viltus ziņām. Cognitive Research: Principles and Implications\n, 5\n(1), 47. https://doi.org/10.1186/s41235-020-00252-3\n
  • \n
  • \nNickerson, R. S. (1998). Apstiprinājuma tendence: visur sastopama parādība daudzās izpausmēs. Vispārīgās psiholoģijas pārskats\n, 2\n(2), 175–220. https://doi.org/10.1037/1089-2680.2.2.175\n
  • \n
  • Pennycook, G., Cheyne, J. A., Barr, N., Koehler, D. J., & Fugelsang, J. A. (2015). Par pseidodziļdomīgu muļķību uztveršanu un atklāšanu. Spriedums un lēmumu pieņemšana, 10(6), 549–563.\n
  • \n
  • \nPennycook, G., McPhetres, J., Zhang, Y., Lu, J. G., & Rand, D. G. (2020). Cīņa pret COVID-19 dezinformāciju sociālajos tīklos: eksperimentāli pierādījumi par mērogojamu precizitātes veicināšanas intervenci. Psychological Science\n, 31\n(7), 770–780. https://doi.org/10.1177/0956797620939054\n
  • \n
  • \nPennycook, G., & Rand, D. G. (2019). Lazy, not biased: Susceptibility to partisan fake news is better explained by lack of reasoning than by motivated reasoning. Cognition\n, 188\n, 39–50. https://doi.org/10.1016/j.cognition.2018.06.011\n
  • \n
  • \nReber, R., & Schwarz, N. (1999). Uztveres plūstamības ietekme uz patiesības vērtējumiem. Consciousness and Cognition\n, 8\n(3), 338–342. https://doi.org/10.1006/ccog.1999.0386\n
  • \n
  • \nRoozenbeek, J., Schneider, C. R., Dryhurst, S., Kerr, J., Freeman, A. L. J., Recchia, G., van der Bles, A. M., & van der Linden, S. (2020). Susceptibility to misinformation about COVID-19 around the world. Royal Society Open Science\n, 7\n(10), 201199. https://doi.org/10.1098/rsos.201199\n
  • \n
  • \nSultan, M., Tump, A. N., Geers, M., Lorenz-Spreen, P., Herzog, S. M., & Kurvers, R. H. J. M. (2022). Laika spiediens samazina spēju atšķirt nepatiesu informāciju, bet nemaina atbilžu tendenciozitāti. Scientific Reports\n, 12\n(1), 22416. https://doi.org/10.1038/s41598-022-26209-8\n
  • \n
  • \nSultan, M., Tump, A. N., Ehmann, N., Lorenz-Spreen, P., Hertwig, R., Gollwitzer, A., & Kurvers, R. H. J. M. (2024). Tieksme uzsūkties tiešsaistes dezinformācijai: sistemātiska metaanalīze par demogrāfiskajiem un psiholoģiskajiem faktoriem. Proceedings of the National Academy of Sciences\n, 121\n(47), e2409329121. https://doi.org/10.1073/pnas.2409329121\n
  • \n
  • \nTraberg, C. S., & van der Linden, S. (2022). Līdzīgi domājošie tiek pārliecināti kopā: uztveramā avota uzticamība ietekmē politiskās tendences ietekmi uz uzņēmību pret dezinformāciju. Personība un individuālās atšķirības\n, 185\n, 111269. https://doi.org/10.1016/j.paid.2021.111269\n
  • \n
  • \nUnkelbach, C., Koch, A., Silva, R. R., & Garcia-Marques, T. (2019). Patiesība, to atkārtojot: paskaidrojumi un sekas. Pašreizējās tendences psiholoģijas zinātnē\n, 28\n(3), 247–253. https://doi.org/10.1177/0963721419827854\n
  • \n
\n
\n
\n

☑ Loģiskās kļūdas

\n
\n
\n\n
    \n
  • Pārbaudiet, vai nav pārsteidzīgi vispārinājumi – secinājumi, kas balstās uz pārāk maziem pierādījumiem.\n
  • \n
  • Pārbaudiet, vai nav viltus dilemmu – iespēju ierobežošana līdz divām, ja to ir vairāk.\n
  • \n
  • Pārbaudiet, vai nav izmantoti „salmu vīra” argumenti – pozīcijas nepareiza interpretācija, lai to viegli atspēkotu.\n
  • \n
  • Pārbaudiet, vai nav izmantota neziņas argumentācija – patiesības apgalvojums, pamatojoties uz to, ka nav pierādījumu par pretējo.\n
  • \n
  • Pārbaudiet, vai nav atsauces uz autoritāti – patiesības pieņemšana, balstoties vienīgi uz autoritāti.\n
  • \n
  • Pārbaudiet, vai nav novirzes no temata – novēršot uzmanību no galvenā jautājuma.\n
  • \n
  • Pārbaudiet, vai nav viltus cēloņu – sajaucot korelāciju ar cēloņsakarību.\n
  • \n
  • Pārbaudiet, vai nav ad hominem – uzbrukums personai, nevis argumentam.\n
  • \n
  • Pārbaudiet, vai nav izmantots arguments ad populum – patiesības pierādīšana, balstoties uz popularitāti.\n
  • \n
  • Pārbaudiet, vai nav „slidena nogāze” – apgalvojums, ka viens solis noved pie galējībām.\n
  • \n
  • Pārbaudiet, vai nav apļveida loģika – secinājuma izmantošana kā premisa.\n
  • \n
\n
\n
\n\n

Dezinformācija

\n

Dezinformācija ir nepatiesa vai maldinoša informācija, ko apzināti izplata cilvēki, kuri zina, ka tā ir nepatiesa, ar nolūku kaitēt, maldināt vai manipulēt. Gan izplatītāja apzināšanās par nepatiesību, gan ļaunprātīgs nolūks ir definīcijas pazīmes.

\n

Bieži sastopamas formas ietver politiskās ietekmes operācijas, komerciālu maldināšanu, propagandu, izdomātu saturu un manipulētus medijus. Dezinformāciju bieži rada organizētas personas, un tā tiek pastiprināta ar koordinētu neautentisku/mākslīgu pastiprināšanu (sk. iepriekš).

\n

Kļūdaina informācija

\n

Nepareiza informācija ir patiesa informācija, kas tiek izplatīta ar nolūku nodarīt kaitējumu. Saturs pats par sevi ir precīzs, bet tā publicēšana, pasniegšana vai laiks ir aprēķināts tā, lai nodarītu kaitējumu personai, grupai vai iestādei. Klasisks piemērs ir apzināta privātas, jutīgas vai konteksta atkarīgas informācijas pārnešana publiskā vai kaitīgā kontekstā.

\n

Piemēri ietver noplūdušas privātas sarakstes publicēšanu, lai diskreditētu mērķi, precīzu, bet stigmatizējošu personas datu publiskošanu (piemēram, atriebības porno) un faktiski pareizas, bet kontekstā kaitīgas informācijas stratēģisku atklāšanu brīžos, kas aprēķināti, lai panāktu maksimālu ietekmi.

\n

Nepareiza informācija ir vismazāk apspriestā no šīm trim kategorijām, jo saturs nav nepaties, tādējādi tā neietilpst faktu pārbaudes sistēmās. Taču tā veido nozīmīgu daļu no informācijas haosa ekosistēmas (Wardle & Derakhshan, 2017).

\n
\n
\n\n

Robežas ir caurlaidīgas. Praksē šīs trīs kategorijas bieži pārklājas vai mainās:

\n
    \n
  • Dezinformācija, ko reiz izplatījuši cilvēki, kuri tai patiesi tic, kļūst par malinformāciju, tālāk izplatoties. Vienam un tam pašam saturam var būt dažādas kategorijas atkarībā no tā, kas to izplata.\n
  • \n
  • Saturs var būt daļēji paties, daļēji nepaties. Wardle (2017) uzskaita septiņas informācijas haosa formas — tostarp maldinošu saturu, nepatiesu kontekstu un manipulētu saturu — ne viss nepatiesais saturs ir pilnībā izdomāts.\n
  • \n
  • No ārpuses ir empīriski grūti noteikt nolūku. Ja nav pieejama informācija par dalītāja zināšanu līmeni un motivāciju, robežu starp neprecīzu informāciju un dezinformāciju bieži vien nevar noteikt viennozīmīgi.\n

    Tāpēc šajā sistēmā mācību dalībniekiem tiek lūgts izskaidrot, kāpēc praksē bieži ir grūti noteikt robežu starp nepatiesu informāciju un dezinformāciju.

    \n
  • \n
\n
\n
\n
\n
\n\n
\n
\n
\n

AI halucinācijas / konfabulācija

\n

Ģeneratīvo AI sistēmu sistemātiska tendence radīt ticami skanošu saturu, kas faktiski ir nepareizs, izdomāts vai neapstiprināms — tostarp izdomātas atsauces, neeksistējoši pētījumi, nepareizi attiecinātas citātas, izdomātas statistikas dati un nepatiesas biogrāfiskas, vēsturiskas vai zinātniskas detaļas.

\n

Halucinācijas rodas tāpēc, ka ģeneratīvās AI sistēmas rada rezultātus, prognozējot ticamus turpinājumus no apmācības laikā apgūtajiem modeļiem, nevis iegūstot pārbaudītu informāciju no zināšanu avota. Sistēma optimizē tekošumu un ticamību, nevis precizitāti. Izdomāts saturs parasti tiek pasniegts ar tādu pašu pārliecību kā precīzs saturs — nav iekšēja signāla, kas uzticami atšķirtu abus.

\n

Pētniecības kontekstā terminu „konfabulācija” dažkārt izmanto biežāk nekā „halucinācija”, jo psiholoģijā konfabulācija nozīmē nepatiesu, bet patiesi ticamu stāstu veidošanu bez nolūka maldināt — kas ir tuvāk tam, ko faktiski dara ģeneratīvās sistēmas. Halucinācija joprojām ir dominējošais termins sabiedriskajā un tehniskajā diskursā, un tas tiek saglabāts arī šeit.

\n

Halucinācijas nav gadījuma rakstura kļūdas. Tās ir pamatā esošā ģenerēšanas procesa strukturāla iezīme, nevis kļūdas, kuras var droši novērst, izmantojot labākus norādījumus (Kalai & Vempala, 2024). Tās rodas visās ģeneratīvajās sistēmās — tostarp ☑ RAG AI, kur izdomājumi var ietekmēt arī to, kā atrastie avoti tiek apkopoti, citēti vai attiecināti.

\n

Didaktiskais pozicionējums: Halucinācijas konceptuāli var iedalīt kā ceturto kategoriju līdzās dezinformācijai, nepatiesai informācijai un kļūdainai informācijai. Atšķirībā no abām pārējām, tās nerada cilvēks ar nolūku vai kļūdu, bet rodas no paša ģenerēšanas procesa — sistēmas radīts nepatiesas saturs bez apzināta avota. Šo trīskāršo iedalījumu šeit piedāvājam kā pedagoģiski noderīgu ietvaru, nevis kā vispārpieņemtu terminoloģiju.

\n

Izplatītie veidi:

\n
    \n
  • izdomātas atsauces — citāti no rakstiem, grāmatām vai pētījumiem, kas nepastāv\n
  • \n
  • nepareizi attiecinātas vai izdomātas citātas\n
  • \n
  • izdomāta statistika vai skaitliskas apgalvojumi\n
  • \n
  • neprecīzi biogrāfiskie, vēsturiskie vai zinātniskie dati, kas tiek pārliecinoši pasniegti\n
  • \n
  • pārliecinoši nepareizi reālo avotu kopsavilkumi\n
  • \n
\n
\n
\n\n

Jebkuru faktu apgalvojumu no ģeneratīvās AI sistēmas uzskatiet par nepārbaudītu, kamēr tas nav pārbaudīts neatkarīgā avotā. Atsauces un citāti ir īpaši viegli pārbaudāmi — meklējiet, vai citētais avots pastāv un satur to, ko apgalvoja AI.

\n
\n
\n
\n
\n\n
    \n
  • \nBender, E. M., Gebru, T., McMillan-Major, A., & Shmitchell, S. (2021). Par stochastisko papagailu briesmām: vai valodas modeļi var būt pārāk lieli? 2021. gada ACM konferences par taisnīgumu, atbildību un pārredzamību (FAccT '21) materiāli\n, 610–623. https://doi.org/10.1145/3442188.3445922\n
  • \n
  • \nJi, Z., Lee, N., Frieske, R., Yu, T., Su, D., Xu, Y., Ishii, E., Bang, Y. J., Madotto, A., & Fung, P. (2023). Pētījums par halucinācijām dabiskās valodas ģenerēšanā. ACM Computing Surveys, 55\n(12), 248. raksts. https://doi.org/10.1145/3571730\n
  • \n
  • \nKalai, A. T., & Vempala, S. S. (2024). Kalibrētiem valodas modeļiem ir jāhalucinē. 56. ikgadējā ACM simpozija par skaitļošanas teoriju (STOC 2024) materiāli\n, 160–171. https://doi.org/10.1145/3618260.3649777\n
  • \n
  • \nSmith, A. L., Greaves, F., & Panch, T. (2023). Halucinācijas vai konfabulācija? Neiroanatomija kā metafora lielos valodas modeļos. PLOS Digital Health, 2\n(11), e0000388. https://doi.org/10.1371/journal.pdig.0000388\n
  • \n
\n
\n
\n

Satura nepatiesa attēlošana

\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n
FormātsManipulētsIzdomāts
Tekstsautentisks citāts, izrauts no konteksta, viltota virsrakstsizdomāts citāts, fiktīvs raksts, ar mākslīgo intelektu ģenerēts teksts
Attēlsretušēta fotogrāfija, apgriezts oriģinālsAI ģenerēts attēls, gleznota viltus "fotogrāfija"
Audiorediģēts vai paātrināts oriģinālais ierakstsbalss klons, ar AI ģenerēta runa
Videolēti viltoti, pārgriezti klipipilnībā sintētisks deepfake video
Dokumentsviltots veidlapas galvenais uz modificēta paraugaizdomāta "oficiāla" vēstule
\n

Izdomāts saturs

\n

Izdomāts saturs ir informācija — teksts, attēls, audio, video, dokuments vai cits formāts —, kas ir pilnībā izdomāta un kurai nav reāla avota. To var izveidot manuāli (izdomāti citāti, fiktīvi ziņu raksti) vai sintētiski ģenerēt ar AI sistēmām (sintētiski video, balss kloni, AI ģenerēti attēli, AI rakstīts teksts).

\n

Manipulēts saturs

\n

Manipulēts saturs ir patiesa informācija — teksts, attēls, audio, video, dokuments vai cits formāts —, kas ir mainīta, lai maldinātu. Avota materiāls ir reāls, bet tas ir modificēts, izmantojot rediģēšanu, selektīvu apgriešanu, ātruma pielāgošanu, konteksta maiņu, kadru izmaiņas, balss aizstāšanu vai citas transformācijas, kas maina tā nozīmi vai acīmredzamo kontekstu.

\n

Lēts viltojums

\n

Lēts viltojums ir medijs, kas ir mainīts, izmantojot tradicionālus, plaši pieejamus rīkus — attēlu rediģēšanas programmatūru, video rediģēšanas programmatūru vai pamata konteksta maiņu — nevis uz AI balstītas ģeneratīvās metodes. Terminu ieviesa Paris un Donovan (2019), lai pievērstu uzmanību tam, ka visietekmīgākās manipulēta medija formas publiskajā diskursā parasti nav sarežģīti, ar AI ģenerēti deepfakes, bet gan vienkāršas, zemas izmaksas metodes, kuras var izmantot ikviens.

\n

Bieži izmantotas lētu viltojumu metodes ietver:

\n
    \n
  • ātruma manipulācija (piemēram, palēninātais Pelosi video, 2019)\n
  • \n
  • audio vai video pārgriešana un selektīva montāža\n
  • \n
  • rekontekstualizāciju (reālu mediju attēlošanu ar nepatiesu aprakstu vai ietvaru)\n
  • \n
  • foto retušēšana, apgriešana vai kompozīcija\n
  • \n
  • apmainīti apraksti un viltoti ekrānuzņēmumi\n
  • \n
  • audio savienošana\n
  • \n
\n

Tas, vai lēts viltojums ir dezinformācija, malinformācija vai nepatiesa informācija, ir atkarīgs no autora zināšanām un nolūkiem (→ informācijas haoss), nevis no paša paņēmiena. Viens un tas pats montēts video var kalpot kā satīra, fikcija, izglītojošs materiāls vai maldināšana.

\n

Salīdzinot ar dziļajiem viltojumiem, lēti viltojumi prasa mazākas tehniskās prasmes, bet tos ne vienmēr ir vieglāk atklāt — labi izpildītu rekontekstualizāciju vai selektīvu montāžu var būt ārkārtīgi grūti identificēt, ja nav pieejams oriģinālais avota materiāls.

\n
\n
\n\n
\n
\n
\n\n

Deep Fake

\n

Deep Fake

\n

Deepfake ir sintētisks medijs, ko ģenerē mākslīgais intelekts — konkrēti, izmantojot dziļās mācīšanās metodes, piemēram, ģeneratīvās pretinieku tīklojumus (GAN), difūzijas modeļus vai transformatoru bāzētus ģeneratorus. Termins ir vārdu “deep learning” (dziļā mācīšanās) un “fake” (viltus) apvienojums, kas radās 2017. gadā tiešsaistes kopienās, kurās tika veidoti ar mākslīgo intelektu ģenerēti sejas apmaiņas video. Kopš tā laika tehnoloģija ir izplatījusies visos mediju veidos un tagad ir plaši pieejama ar patērētāju līmeņa rīkiem.

\n

Biežāk sastopamās formas ir:

\n
    \n
  • Sejas apmaiņa: vienas personas sejas aizstāšana ar citas seju video\n
  • \n
  • Balss klonēšana: runātāja balss sintēze no neliela audio parauga\n
  • \n
  • Pilnībā sintētisks video: ar AI radīts materiāls, kurā redzami cilvēki, notikumi vai ainas, kas nekad nav notikušas\n
  • \n
  • Sintētiskā fotogrāfija: ar mākslīgo intelektu radīti nekustīgi attēli ar izdomātiem cilvēkiem vai notikumiem\n
  • \n
  • Teksta pārvēršana video un audio: mediju ģenerēšana no rakstītiem norādījumiem\n
  • \n
\n

Tas, vai dziļais viltojums ir nepatiesa informācija, dezinformācija vai ļaunprātīga informācija, ir atkarīgs no izpildītāja zināšanām un nolūkiem (→ informācijas haoss), nevis no paša paņēmiena. Dziļo viltojumu metodes tiek likumīgi izmantotas arī filmu ražošanā, pieejamības rīkos (sintētiskās balsis cilvēkiem, kuri zaudējuši savu balsi), valodu dublēšanā, satīrā, izglītībā un mākslā.

\n

Salīdzinot ar lētiem viltojumiem, dziļajiem viltojumiem ir nepieciešamas lielākas tehniskās iespējas, lai radītu pārliecinošu rezultātu, un tie var sasniegt augstāku vizuālo vai audiālo reālismu. Tomēr tie nav dominējošā mediju manipulācijas forma publiskajā diskursā — Paris un Donovan (2019) norāda, ka vienkāršākas lētu viltojumu metodes joprojām ir izplatītākas un bieži vien rada lielākas sekas. To atklāšana ir aktīva pētniecības joma; pašreizējās pieejas apvieno artefaktu analīzi, bioloģisko neatbilstību pārbaudes un izcelsmes pārbaudi (piemēram, C2PA satura apliecības).

\n
\n
\n\n
    \n
  • \nParis, B., & Donovan, J. (2019). Deepfakes and Cheap Fakes: The Manipulation of Audio and Visual Evidence\n. Data & Society Research Institute.https://datasociety.net/library/deepfakes-and-cheap-fakes/\n
  • \n
  • Chesney, R., & Citron, D. K. (2019). Deepfakes: draudīgs izaicinājums privātumam, demokrātijai un valsts drošībai. California Law Review, 107(6), 1753–1820.\n
  • \n
  • \nVaccari, C., & Chadwick, A. (2020). Deepfakes un dezinformācija: sintētisko politisko video ietekmes izpēte uz maldināšanu, nenoteiktību un uzticēšanos ziņām. Social Media + Society\n, 6(1), 1–13. https://doi.org/10.1177/2056305120903408\n
  • \n
\n
\n
\n","UPDATEDAT":"2026-05-11T20:49:09.513Z","LANG":"lv","ID":"3a2e66d9-94a9-482d-a20b-3c977e896fa3","TITLE":"Nepareiza informācija un manipulācija","SOURCELANG":"en"}