推 vandervaal: 看到逆模因還以為看到SCP的文 09/08 11:00
→ skyhawkptt: PLA嘿美日常... 09/08 16:56
DARPA unleashes anti-meme militia to fight deepfakes & ‘polarizing’ viral
content
DARPA反模因軍對抗深度造假和極化的網路內容
原文：RT.com
https://www.rt.com/usa/467836-darpa-deepfakes-disinfo-malicious-dissent/
The Pentagon has unveiled an initiative to fight ‘large-scale, automated
disinformation attacks’ by unearthing deep-fakes and other polarizing
content – with the eventual goal of rooting out so-called ‘malicious intent
’ entirely.
五角大樓公佈了一項倡議，旨在通過發掘深度造假和其他兩極分化內容來打擊“大規模自
動化虛假信息攻擊”--- 最終目標是完全消除所謂的“惡意企圖”。
The Defense Advanced Research Projects Agency (DARPA) is seeking software
capable of churning through a test set of half a million news stories,
photos, and audio/video clips to target and neutralize potentially viral
information before it spreads. In DARPA jargon, the aim is to “automatically
detect, attribute, and characterize falsified multi-modal media to defend
against large-scale, automated disinformation attacks.” “Polarizing viral
content,” however, includes inflammatory truths, and the program’s ultimate
goal seems to be to stamp out dissent.
美國國防部高級研究計劃局（DARPA）正在尋求一種軟體，能夠經由測試50萬個新聞報導
，照片和音頻/視頻剪輯來定位潛在的移動網路信息，並使其失效。在DARPA術語中，目的
是“自動檢測，歸因和描述偽造的多模式媒體，以抵禦大規模自動化的虛假信息攻擊。”
然而，“極化移動網路內容”裡面包括具有煽動性的事實，而該計劃的最終目標似乎是要
消除反對意見。
The Semantic Forensics program will scan news stories and social media posts
with a barrage of algorithms in the hope of identifying inconsistencies that
could mark a story as fake. The desired program will not just identify a meme
as inauthentic – it will identify the source of that meme, the alleged
intent behind it, and predict the impact of its spread.
語義取證程式將使用一系列算法掃描新聞報導和社交媒體帖子，以期找出不一致性，而能
標記出假故事。理想的程式不僅僅會將模因辦識為假---它將識別模因的來源，其背後的
所謂意圖，並預測其傳播的影響。
To hear them tell it, the Pentagon just wants to even the playing field
between the ‘good guys’ – the fake-hunters pursuing the cause of truth in
media – and the ‘bad guys’ sowing discord one slowed-down Nancy Pelosi
speech at a time. But the Pentagon’s targets aren’t limited to deepfakes,
the bogeyman-of-the-month being used to justify this unprecedented military
intrusion into the social media and news realm, or fake news at all. If the
program is successful after four years of trials, it will be expanded to
target all “malicious intent” – a possibility that should send chills down
the spine of any journalist who’s ever disagreed with the establishment
narrative.
據五角大樓說，他們只想確保“好人”與“壞人”間的公平競爭，這裡“好人”是指在媒
體上打假追求真相的人，而“壞人”是指那些傳播慢速佩洛西講話的人[註1]。然而，五
角大樓的目標並不僅限於發掘深度造假。這項駭人的技術被用來給前所未有地對社交媒體
和新聞領域進行軍事入侵提供正當性，儘管這也包括虛假新聞。如果該計劃在經過四年的
試驗後獲得成功，它將擴大到針對所有的“惡意意圖”---這種可能性會讓任何對主流
敘事有異議的記者感到寒意。
[註1]：佩洛西說話的視頻曾被故意造假成慢速播放，使其聽起來像是個醉鬼。來源：
https://tinyurl.com/y4p49fb5
To adequately test the program, the Pentagon has to spike its array of
500,000 test stories with 5,000 convincing fakes, some of which could
conceivably make their way into the “live” news stream - although the
mainstream media has not exactly had trouble generating false stories on its
own in recent weeks. MSNBC’s wholly unverified and still
incompletely-retracted “Russian cosigners” fiction and the scare story that
the Trump administration would end birthright citizenship for the children of
US service members born overseas both took social media by storm before the
fact-checkers could boot up their computers.
為了充分測試這個程式，五角大樓得用5000個確信的假消息來摻入其50萬個測試故事。不
難想像，其中一些假消息進入了“即時”新聞串流---儘管最近幾週主流媒體靠自己造假
故事並無困難。 MSNBC完全未經證實且仍未完全撤回的“俄羅斯擔保人”小說 [註2]，以
及川普政府將終止美國海外服務人員子女公民身份的恐慌故事，都在社交媒體風靡一時，
而這些都發生在事實查核者發生作用之前。
[註2]：MSNBC於8/28/2019報導川普向德國金融機構貸款的擔保人為普京的朋友。來源：
https://tinyurl.com/y2krknho
And the government itself, including the Pentagon, has an extensive history
of running fake social media profiles to collect data on persons of interest,
including through the NSA's JTRIG information-war program revealed in the
Snowden documents. Agents regularly deploy reputational attacks against
dissidents using false information. Fake identities are used to cajole
unsuspecting individuals into collaborating in fake FBI "terror" plots, a
phenomenon which might once have been called entrapment but is merely
business as usual in post-9/11 America.
包括五角大樓在內的美國政府，本身擁有廣泛運行社交媒體假檔案的歷史，以收集有關人
士的數據，包括通過史諾登文件中所披露的國家安全局的JTRIG信息戰計劃 [註3]。特工
們使用虛假信息定期對異議人士進行聲譽攻擊。偽造的身份被用來將毫無戒心的個人騙入
FBI假造出的”恐怖威脅”中，使其與FBI進行合作。這種現象曾被稱為” 釣魚執法”，
但在9/11後的美國祇是日常工作。
[註3]：JTRIG的任務範圍包括使用“骯髒的技巧”打擊敵人聲譽，植入錯誤信息，關閉敵
方的通信，因而摧毀、否定、羞辱、與破壞敵人。見
https://en.wikipedia.org/wiki/Joint_Threat_Research_Intelligence_Group
Which begs the question – how will DARPA handle the malicious falsehoods
generated by “friendly” media? This, it would seem, is where the “impact”
and “intent” fields come in – fakes from “trusted sources” will be let
through, while fakes (and real stories) designed to “undermine key
individuals and organizations” will be terminated before they have an
impact. When “disinformation” is redefined to include all potentially
polarizing stories that don’t conform to the establishment narrative,
reality is discarded as so much fake news and replaced with Pentagon-approved
pablum.
這引出了一個問題--- DARPA將如何處理“友方”媒體產生的惡意謊言？看起來這就是簡
報中提及“影響”和“意圖” 這個範疇的理由---來自“可靠來源”的假信息將允許通過
，而旨在“破壞關鍵個人和組織”的假信息（和真實故事）將在他們產生影響之前被擋下
。當“虛假信息”被重新定義為包括所有不符合主流敘事的潛在極化故事時，真實將被當
作假新聞丟棄，而由五角大樓批准的“簡單精神口糧”取而代之。
***
補充：
DARPA官方新聞
https://www.darpa.mil/news-events/2019-09-03a
DARPA簡報本文
https://www.darpa.mil/attachments/SemanticForensics-IndustryDay-2019-08-12.pdf
心得：DARPA簡報信息量豐富，值得仔細研究。第8頁的目標若能達成，會是一種非常可怕
的控制力量，對國際政治也會有很大影響。
--
※ 發信站: 批踢踢實業坊(ptt.cc), 來自: 129.110.242.26 (美國)
※ 文章網址: https://www.ptt.cc/bbs/IA/M.1567897990.A.935.html