《PYMNTS:如何分辨AI生成内容和人类创作的内容成为难题(2023)(英文版)(16页).pdf》由会员分享,可在线阅读,更多相关《PYMNTS:如何分辨AI生成内容和人类创作的内容成为难题(2023)(英文版)(16页).pdf(16页珍藏版)》请在三个皮匠报告上搜索。
1、September 2023Generative AI Tracker SeriesAUGUST 2023Generative AI Tracker SeriesRead the previous editionIs That Content Generated by AI or Humans?Hard to Tell Companies Face the Dilemma of Labeling AI-Generated Text Appropriately P.10 Academia,Research and Publishing Adopt Content Scouts P.12 Tack
2、ling AI with AI May Not Be All It Is Cracked Up to Be P.16 Generative AI Renews Longtime Privacy Concerns P.242|Generative AI Tracker SeriesGenerative AI Tracker Series|3 2023 PYMNTS All Rights Reserved 2023 PYMNTS All Rights ReservedAcknowledgmentThe Generative AI Tracker Series is produced in coll
3、aboration with AI-ID,and PYMNTS is grateful for the companys support and insight.PYMNTS retains full editorial control over the following findings,methodology and data analysis.Whats Inside04 AI or Human?It Can Be Hard to TellSome studies find that more than half of individuals cannot accurately ide
4、ntify content made by AI chat-bots such as ChatGPT.10 Companies Face the Dilemma of Labeling AI-Generated Text AppropriatelyGenerative AI tools are promising but not foolproof.Their level of accuracy can be difficult to calculate and varies widely.12 Academia,News,Publishers:Early Adopters of AI-Gen
5、erated Content DetectionThe effort to identify and label AI content is critical for industries that rely on authentic,trustworthy work to protect their integrity.16 Tackling AI with AI May Not Be Ideal.Not Yet,at Least Countermeasures to keep tabs on artificially gener-ated content are AI tools them
6、selves.They come with drawbacks.24 Generative AI Renews Longtime Privacy ConcernsArtificially generated content,including deceptive and fraudulent content,is at our doorstep.Humans and companies cannot rein it in entirely.30 AboutInformation on PYMNTS and AI-ID4|Generative AI Tracker SeriesGenerativ
7、e AI Tracker Series|5 2023 PYMNTS All Rights Reserved 2023 PYMNTS All Rights ReservedBackgroundAI or Human?It Can Be Hard to TellHow do we decide whether a piece of text was created by a human,artificial intelligence(AI)or a combination of the two?Some stud-ies find that more than half of individual
8、s cannot accurately identify content made by AI chatbots.That percentage jumps to 66%with content generated by GPT-4,the latest language model powering ChatGPT.The implication for regulators,businesses and the public at large is that many are susceptible to AI manipulation without realizing it.This
9、can have troubling and profound consequences for society,as AI content has the potential to influence large swaths of the population based on false premises.Factors influencing ChatGPTs accuracy when generating responses:Training dataContextUser inputLanguage complexityBiasGenerative natureHuman eva
10、luation6|Generative AI Tracker SeriesGenerative AI Tracker Series|7 2023 PYMNTS All Rights Reserved 2023 PYMNTS All Rights ReservedBackgroundThe tools to spot AI-generated content are less than reliable.As generative AI grows,the countermeasures being developed to detect artificially generated conte
11、nt are close behind.Companies such as Originality.ai,Turnitin and Copyleaks are designing tools that discern between unique content and content compiled from online material.These tools are promising but not infallible,as their accuracy can be difficult to measure and varies widely.In some cases,the
12、se solutions can be highly unreliable.OpenAI recently made news for retiring its AI classifier due to its“low rate of accuracy”in detecting artificial content.ChatGPTs level of accuracy can fluctuate significantly over time,in what experts call“drift.”ChatGPTs level of accuracy in March when identif
13、ying prime numbers ChatGPTs level of accuracy in June for the same task8|Generative AI Tracker SeriesGenerative AI Tracker Series|9 2023 PYMNTS All Rights Reserved 2023 PYMNTS All Rights ReservedBackgroundIt is in the public interest to distinguish human-from AI-generated content.As AI models grow i
14、n size and complexity,the ability to distinguish AI-produced content from human-sourced material becomes increas-ingly difficult.The same can be said for fraud:The more sophisticated the model,the more challenging it is to spot fraudulent activity.However,exposing artificial content is worth the eff
15、ort and is criti-cal to the integrity and survival of industries and organizations that depend on authentic,trustworthy information.10|Generative AI Tracker SeriesGenerative AI Tracker Series|11 2023 PYMNTS All Rights Reserved 2023 PYMNTS All Rights ReservedCompanies of NoteEmerging Companies Are Ta
16、ckling the Challenge of Spotting AI-Generated ContentA host of startups are cropping up to tackle detecting AI-generated content,and global organizations are fast on their heels.The emerg-ing companies highlighted here represent just a few of the key players offering AI-generated content detection s
17、olutions.That list will only grow as the technology advances and new firms enter the market.More and more universities,publishing companies,research organi-zations and financial firms are showing an interest in AI-generated content detection tools to ensure that the work of their students,writers,re
18、searchers and analysts is authentic,original and reliable.GPTZero secured seed funding in the spring to offer AI-generated content detection tools using existing generative AI models.Enterprises and educators can integrate these tools into their pro-cesses using application programming interfaces(AP
19、Is).Similarly,Copyleaks obtained seed funding in 2022 for its AI-generated con-tent detection tool,a Chrome extension that identifies synthetically generated content and plagiarism.Originality.ai is another content detection tool that seeks to recog-nize plagiarism and paraphrasing.Turnitin offers a
20、 similar solution for academic settings with the promise of delivering academic integ-rity by spotting plagiarism.AI-generated content detection tools are far from failsafe.They will require iterative efforts to improve accuracy.74%Proportion of AI-generated text that OpenAI said its classifier tool
21、 missed detecting68%Share of employees who do not disclose their use of generative AI to their employers#10How AI-generated content detection ranked in educational challenges in 2022#1How AI-generated content detection ranked in educational challenges in 202312|Generative AI Tracker SeriesGenerative
22、 AI Tracker Series|13 2023 PYMNTS All Rights Reserved 2023 PYMNTS All Rights ReservedInnovation and Use CasesAcademia,News,Publishers:Early Adopters of AI-Generated Content Detection Just as financial organizations prioritize fraud detection,industries and institutions that depend on authentic,trust
23、worthy content to preserve their integrity are seeking technologies that can correctly and comprehensively identify AI-generated content.These sectors include academia,research and news organizations and publishers.Schools and teachers want to make sure their students are doing their own work by det
24、ermining who or what produced it.Research organizations value the integrity of original content submitted by their researchers and graduate students.News organizations pride them-selves on producing accurate and original content.Such use cases are examples of how AI-generated content detectors can b
25、e applied to combat fraud and distinguish authenticity from counterfeits.14|Generative AI Tracker SeriesGenerative AI Tracker Series|15 2023 PYMNTS All Rights Reserved 2023 PYMNTS All Rights ReservedInnovation and Use CasesAI can be used to produce inauthentic content or commit fraud.It can also be
26、used to stop it.AI is not inherently good or bad.It can,however,assume a good or bad face or serve helpful or harmful causes depending on who is wielding its power and for what purpose.Guiding princi-ples and regulations such as the European Unions General Data Protection Regulation(GDPR)can provide
27、 valuable context to assess where they fall.Just as generative AI can be used to replace human creativity in the industries where the latter is most essential,fraudsters can use generative AI to tap personal data,access financial funds,create fake identities or conduct fraudulent transac-tions.Gener
28、ative AI models lack transparency and governance guidelines,giving fraudsters the very shadows in which they seek to hide.However,businesses,academics,fraud examiners and research-ers can also leverage AI to boost their creativity,improve customer experience,automate processes,augment transaction mo
29、nitoring and spot artificial content,fraudulent transactions and fraudsters.Financial institutions such as Visa,Bank of America and PayPal use generative AI as part of their efforts to stop illicit transactions and malicious actors.The bigger and more powerful the model,the harder it is to build AI
30、models to detect what text is written by a human and what isnt.IRENE SOLAIMAN Policy directorSource:Joe Marshall:As AI cheating booms,so does the industry detecting it:We couldnt keep up with demand.The Guardian.July 5,2023.https:/ August 202316|Generative AI Tracker SeriesGenerative AI Tracker Seri
31、es|17 2023 PYMNTS All Rights Reserved 2023 PYMNTS All Rights ReservedIssues and ChallengesTackling AI with AI May Not Be Ideal.Not Yet,at LeastAs generative AI grows by leaps and bounds,countermeasures to keep tabs on artificially generated content are emerging.The chal-lenge for companies developin
32、g these solutions is not only that their tools are new,unproven and offer mixed success but also that these tools capabilities will be increasingly stretched as AI-generated con-tent grows and becomes more humanlike.Leading AI developers and technology companies including Amazon,Google,Meta,Microsof
33、t and OpenAI have made voluntary commitments to the Biden-Harris Administration to ensure that their generative AI content tracker tools are effective.Whether those busi-nesses honor their commitments,and to what degree their promises hold substance,remains to be seen.You cant trust anything that yo
34、u see or hear regarding the U.S.elections,thanks to AI.ERIC SCHMIDT Former CEOSource:Bhaskar Chakravorti:The AI Regulation Paradox.Foreign Policy.August 4,2023.https:/for- August 2023.18|Generative AI Tracker SeriesGenerative AI Tracker Series|19 2023 PYMNTS All Rights Reserved 2023 PYMNTS All Right
35、s ReservedIssues and ChallengesContent scouts are AI tools themselves.Generative AI tools use keywords related to a topic as input and scour the web for content with similar keywords.They then rewrite that content autonomously.This is where content scouts come in.These tools employ AI to analyze the
36、 text to detect certain patterns and characteristics indicative of AI-created content.No foolproof methods to detect and expose artificial content exist today.The current crop of content scouts AI tools themselves have a spotty track record in terms of accu-racy and consistency.The reality is that t
37、here may never be a perfect tool,especially as AI-generated content becomes more sophisticated and humanlike.Nevertheless,content scouts work to identify generative AI content based on clues in the text,such as a lack of typos or misspellings,information with no accompanying sources or shorter,flat-
38、sounding sentences all dead giveaways.Repeated words or phrases and the absence of analysis or commentary offer additional hints.The FTC welcomes innovation,but being innovative is not a license to be reckless.We are prepared to use all our tools,including enforcement,to challenge harmful practices
39、in this area.SAMUEL LEVINE Director Source:Cat Zakrzewski:FTC investigates OpenAI over data leak and ChatGPTs inaccuracies.Washington Post.July 13,2023.https:/ August 2023.20|Generative AI Tracker SeriesGenerative AI Tracker Series|21 2023 PYMNTS All Rights Reserved 2023 PYMNTS All Rights ReservedIs
40、sues and ChallengesWill voluntary commitments be enough?If history is any indication,the initial measures technology compa-nies and regulators have taken may not be enough.The preventive measures that technology companies offer today,either with good intentions or to generate goodwill,may also be in
41、adequate.This is partly because of commercial interests and partly because of unfore-seen societal implications of generative AI.Going forward,companies that use generative AI will draw scru-tiny from regulators.Regulators are likely to cite concerns around consumer or public manipulation,the spread
42、 of misinformation or dissemination of harmful content.In response,companies leveraging generative AI will most likely emphasize the benefits of these tech-nologies,downplay troubling scenarios as outliers and seek ways to circumvent regulation.22|Generative AI Tracker SeriesGenerative AI Tracker Se
43、ries|23 2023 PYMNTS All Rights Reserved 2023 PYMNTS All Rights ReservedCybersecurity dons a layer of AI.Cybersecurity breaches are a primary concern across many indus-tries,prompting organizations to address the threat by enhancing their data security protocols to block unauthorized access and stop
44、account takeovers.To this end,companies are using AI to integrate fraud monitoring,identity verification databases,data encryption,virtual private networks(VPNs)and antivirus software into their busi-ness systems and processes.While the use of generative AI is currently in its early stages,it will n
45、ot be long before AI companies incorporate generative AI capabili-ties into their solutions.The repercussions of cybersecurity breaches for these industries include theft of trade secrets,leaks in confidential information and the proliferation of synthetic identities that can lead to the creation of
46、 fake accounts or takeover of valid,existing accounts.They can also hurt the companys reputation and brand in the public eye.56%Share of business leaders incorporating AI into their cybersecurity protocols Issues and Challenges67%Share of business leaders planning to invest in AI technologies to boo
47、st their cybersecurity protocols24|Generative AI Tracker SeriesGenerative AI Tracker Series|25 2023 PYMNTS All Rights Reserved 2023 PYMNTS All Rights ReservedPolicy and RegulationAI Detection Tools Can Be a Blessing and a Curse Governing and regulating AI across borders and businesses have long-term
48、 and far-reaching implications some unintended,some yet unknown.These regulations must not only address fraud,data pri-vacy and security breaches but also balance the protection of free speech while simultaneously preventing misinformation.In the near future,regulators will likely focus on more tang
49、ible con-cerns,such as consumer data protection,intellectual property law violations and financial fraud.In the long term,they will need to expand their efforts to deal with broader matters,including setting guardrails to protect global democratic institutions.26|Generative AI Tracker SeriesGenerati
50、ve AI Tracker Series|27 2023 PYMNTS All Rights Reserved 2023 PYMNTS All Rights ReservedPolicy and RegulationRegulatory and business disclosures are coming soon.Until AI detection tools are perfected and future-proofed,the public will have to rely on self-disclosures,voluntary initia-tives and new le
51、gislation to ensure the accuracy,credibility and authenticity of AI systems.Watermarking AI content is another tool that can help.Government regulation,of course,will go a long way toward safeguarding consumers and the public,but it remains a work in progress.Regulatory efforts such as the REAL Poli
52、tical Advertisements Act,introduced in the U.S.House of Representatives in May,offers an example of how lawmakers are trying to rein in AI.If passed,the law will require advertisers to disclose any AI-generated content used in political ads.This is particularly relevant ahead of the 2024 presidentia
53、l elections.Given the many risks of using personal data collected by AI tools,employers should vet the tools they use and negotiate service agreements to reduce their risk.In these service agreements,employers should consider requiring assurances that the generative AI service complied with applicab
54、le data protection laws when collecting and processing personal data.ZOE ARGENTO Co-chair,privacy and data security practice Source:Zoe Argento:Data protection issues for employers to consider when using generative AI.IAPP.August 9,2023.https:/iapp.org/news/a/data-protection-issues-for-employers-to-
55、consider-when-us-ing-generative-ai/.Accessed August 2023.28|Generative AI Tracker SeriesGenerative AI Tracker Series|29 2023 PYMNTS All Rights Reserved 2023 PYMNTS All Rights ReservedSeveral AI laws and investigations,both in the U.S.and abroad,are in the works.The Federal Trade Commission is invest
56、igating OpenAI to determine whether ChatGPT violates customer protection laws or misuses personal data,as well as generates misleading or harmful information.At issue are OpenAIs data models.How these investigations amount to regulation,fines and conse-quences remains a question mark.How and which e
57、ntity will have the authority to oversee whether the models are retrained properly also remains to be seen.Policy and Regulation$8.8B Total fraud losses in U.S.in 20221.65B Fines issued by European regulators for GDPR and data violations related to AI 30|Generative AI Tracker SeriesGenerative AI Tra
58、cker Series|31 2023 PYMNTS All Rights Reserved 2023 PYMNTS All Rights ReservedThe Generative AI Tracker Series may be updated periodically.While reasonable efforts are made to keep the content accurate and up to date,PYMNTS MAKES NO REPRESENTATIONS OR WARRANTIES OF ANY KIND,EXPRESS OR IMPLIED,REGARD
59、ING THE CORRECTNESS,ACCU-RACY,COMPLETENESS,ADEQUACY,OR RELIABILITY OF OR THE USE OF OR RESULTS THAT MAY BE GENERATED FROM THE USE OF THE INFORMATION OR THAT THE CONTENT WILL SATISFY YOUR REQUIREMENTS OR EXPECTATIONS.THE CONTENT IS PROVIDED“AS IS”AND ON AN“AS AVAILABLE”BASIS.YOU EXPRESSLY AGREE THAT
60、YOUR USE OF THE CONTENT IS AT YOUR SOLE RISK.PYMNTS SHALL HAVE NO LIABILITY FOR ANY INTERRUPTIONS IN THE CONTENT THAT IS PROVIDED AND DISCLAIMS ALL WARRANTIES WITH REGARD TO THE CONTENT,IN-CLUDING THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE,AND NON-INFRINGEMENT AND
61、 TITLE.SOME JURISDICTIONS DO NOT ALLOW THE EXCLUSION OF CERTAIN WARRANTIES,AND,IN SUCH CASES,THE STATED EXCLUSIONS DO NOT APPLY.PYMNTS RESERVES THE RIGHT AND SHOULD NOT BE LIABLE SHOULD IT EXER-CISE ITS RIGHT TO MODIFY,INTERRUPT,OR DISCONTINUE THE AVAILABILITY OF THE CONTENT OR ANY COMPONENT OF IT W
62、ITH OR WITHOUT NOTICE.PYMNTS SHALL NOT BE LIABLE FOR ANY DAMAGES WHATSOEVER,AND,IN PARTICULAR,SHALL NOT BE LIABLE FOR ANY SPECIAL,INDIRECT,CONSEQUENTIAL,OR INCIDENTAL DAMAGES,OR DAMAGES FOR LOST PROFITS,LOSS OF REVENUE,OR LOSS OF USE,ARISING OUT OF OR RELATED TO THE CONTENT,WHETHER SUCH DAMAGES ARIS
63、E IN CONTRACT,NEGLIGENCE,TORT,UNDER STATUTE,IN EQUITY,AT LAW,OR OTHERWISE,EVEN IF PYMNTS HAS BEEN AD-VISED OF THE POSSIBILITY OF SUCH DAMAGES.SOME JURISDICTIONS DO NOT ALLOW FOR THE LIMITATION OR EXCLUSION OF LIABILITY FOR INCIDENTAL OR CONSEQUENTIAL DAMAGES,AND IN SUCH CASES SOME OF THE ABOVE LIMIT
64、ATIONS DO NOT APPLY.THE ABOVE DISCLAIMERS AND LIMITATIONS ARE PROVIDED BY PYMNTS AND ITS PARENTS,AFFILIATED AND RELATED COMPANIES,CONTRACTORS,AND SPONSORS,AND EACH OF ITS RESPECTIVE DIRECTORS,OFFICERS,MEMBERS,EMPLOYEES,AGENTS,CONTENT COMPONENT PROVIDERS,LICENSORS,AND ADVISERS.Components of the conte
65、nt original to and the compilation produced by PYMNTS is the prop-erty of PYMNTS and cannot be reproduced without its prior written permission.The Generative AI Tracker Series is a registered trademark of Whats Next Media&Analytics,LLC(“PYMNTS”).Disclaimer 30PYMNTS is where the best minds and the be
66、st content meet on the web to learn about“Whats Next”in payments and commerce.Our interactive platform is reinventing the way in which companies in payments share relevant information about the initiatives that shape the future of this dynamic sector and make news.Our data and analytics team include
67、s economists,data scientists and indus-try analysts who work with companies to measure and quantify the innovation that is at the cutting edge of this new world.AI-ID is a leading-edge provider of AI and Generative AI Platform Management,Regulation,Governance and Output control capabili-ties and sol
68、utions.With purpose-built solutions including the AIGX File,AI-ID Platform and AI Registry components,the patented AI-ID technology suite and services give Generative AI platforms,providers,their customers and all regulators and stakeholders the trust and data provenance needed in a rapidly advancin
69、g AI world.We are a company with a primary mission:solving the key chal-lenges facing the AI marketplace today while providing strong integrated foundational capabilities for its long-term future suc-cess.Learn more at www.ai-id.io.We are interested in your feedback on this report.If you have ques-tions or comments,or if you would like to subscribe to this report,please email us at .About