東華大學圖書館 |

Language: English

Help

回圖書館首頁

手機版館藏查詢

Back

Switch To: Labeled | MARC Mode | ISBD

Reverse Turing Test in the Age of De...

Uchendu, Adaku,

Linked to FindBook

Google Book

Amazon

博客來

Reverse Turing Test in the Age of Deepfake Texts /

Record Type:	Electronic resources : Monograph/item
Title/Author:	Reverse Turing Test in the Age of Deepfake Texts // Adaku Uchendu.
Author:	Uchendu, Adaku,
Description:	1 electronic resource (158 pages)
Notes:	Source: Dissertations Abstracts International, Volume: 85-05, Section: B.
Contained By:	Dissertations Abstracts International85-05B.
Subject:	Authorship. -
Online resource:	https://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=30720650
ISBN:	9798380725361

Reverse Turing Test in the Age of Deepfake Texts /
Uchendu, Adaku,

Reverse Turing Test in the Age of Deepfake Texts /Adaku Uchendu. - 1 electronic resource (158 pages)

Source: Dissertations Abstracts International, Volume: 85-05, Section: B.

As Artificial Intelligent (AI) technologies become ubiquitous, humans will have to contend with many benefits and disadvantages of these advancements. Particularly, in recent years, Natural Language Generation (NLG) methods have massively improved in the generation of well-written human-quality texts, leading to sophisticated Large-scale Language Models (LLMs) such as ChatGPT, GPT-4, and LLaMA. Naturally, these newer LLMs can generate texts that can be easily misconstrued as human-written, thus causing the problem of "deepfake texts" and exaggerating and exacerbating the potential for misuse (e.g., the generation of misinformation at scale).In this dissertation, thus, we strive to understand the phenomenon surrounding deepfake texts better and develop solutions to tackle the problem. We first frame the detection of deepfake texts using the Reverse Turing Test (RTT) concept, which entails automatically distinguishing AI-generated texts from human-written ones. Then, we address 3 Research Questions (RQs): (1) Can one detect subtle linguistic differences between AI-generated texts and human-written ones? (2) Can one build an accurate deepfake text detector to distinguish between human vs. deepfake text authorship? (3) How can one improve human performance in detecting deepfake texts?RQ1 is addressed by building an interpretable linguistic model to capture the subtle linguistic patterns that distinguish human-written from deepfake texts. RQ2 is addressed by several types of deepfake text detectors-e.g., stylometric, deep learning based, and a hybrid model (i.e., an ensemble of a Transformer-based model and Topological Data Analysis (TDA) techniques). RQ3 is addressed by studying how English experts vs. English non-experts perform at detecting deepfake texts in an individual and collaborative setting.

English

ISBN: 9798380725361Subjects--Topical Terms:

529425
Authorship.
Subjects--Index Terms:

Large-scale Language Models

Reverse Turing Test in the Age of Deepfake Texts /
LDR:03333nmm a22004573i 4500 001 2437656
005 20251124054953.5
006 m o d
007 cr|nu||||||||
008 261011s2023 miu||||||m |||||||eng d
020 $a 9798380725361
035 $a (MiAaPQD)AAI30720650
035 $a (MiAaPQD)PennState_21659azu5030
035 $a AAI30720650
040 $a MiAaPQD $b eng $c MiAaPQD $e rda
100 1 $a Uchendu, Adaku, $e author. $3 3828719
245 1 0 $a Reverse Turing Test in the Age of Deepfake Texts / $c Adaku Uchendu.
264 1 $a Ann Arbor : $b ProQuest Dissertations & Theses, $c 2023
300 $a 1 electronic resource (158 pages)
336 $a text $b txt $2 rdacontent
337 $a computer $b c $2 rdamedia
338 $a online resource $b cr $2 rdacarrier
500 $a Source: Dissertations Abstracts International, Volume: 85-05, Section: B.
500 $a Advisors: Lee, Dongwon Committee members: Bardzell, Jeffrey; Zhang, Rui; Hosseini, Hadi; Silverman, Justin.
502 $b Ph.D. $c The Pennsylvania State University $d 2023.
520 $a As Artificial Intelligent (AI) technologies become ubiquitous, humans will have to contend with many benefits and disadvantages of these advancements. Particularly, in recent years, Natural Language Generation (NLG) methods have massively improved in the generation of well-written human-quality texts, leading to sophisticated Large-scale Language Models (LLMs) such as ChatGPT, GPT-4, and LLaMA. Naturally, these newer LLMs can generate texts that can be easily misconstrued as human-written, thus causing the problem of "deepfake texts" and exaggerating and exacerbating the potential for misuse (e.g., the generation of misinformation at scale).In this dissertation, thus, we strive to understand the phenomenon surrounding deepfake texts better and develop solutions to tackle the problem. We first frame the detection of deepfake texts using the Reverse Turing Test (RTT) concept, which entails automatically distinguishing AI-generated texts from human-written ones. Then, we address 3 Research Questions (RQs): (1) Can one detect subtle linguistic differences between AI-generated texts and human-written ones? (2) Can one build an accurate deepfake text detector to distinguish between human vs. deepfake text authorship? (3) How can one improve human performance in detecting deepfake texts?RQ1 is addressed by building an interpretable linguistic model to capture the subtle linguistic patterns that distinguish human-written from deepfake texts. RQ2 is addressed by several types of deepfake text detectors-e.g., stylometric, deep learning based, and a hybrid model (i.e., an ensemble of a Transformer-based model and Topological Data Analysis (TDA) techniques). RQ3 is addressed by studying how English experts vs. English non-experts perform at detecting deepfake texts in an individual and collaborative setting.
546 $a English
590 $a School code: 0176
650 4 $a Authorship. $3 529425
650 4 $a Taxonomy. $3 3556303
650 4 $a Linguistics. $3 524476
650 4 $a Reproducibility. $3 3683754
650 4 $a Information technology. $3 532993
650 4 $a Information science. $3 554358
653 $a Large-scale Language Models
653 $a Deepfake texts
653 $a Reverse Turing Test
653 $a Deep learning
653 $a ChatGPT
690 $a 0290
690 $a 0489
690 $a 0800
690 $a 0723
710 2 $a The Pennsylvania State University. $e degree granting institution. $3 3765934
720 1 $a Lee, Dongwon $e degree supervisor.
773 0 $t Dissertations Abstracts International $g 85-05B.
790 $a 0176
791 $a Ph.D.
792 $a 2023
856 4 0 $u https://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=30720650