Sense Of Ownership And Authorship For Content Authored Using Large-scale Language Models

3 main points
✔️ Examine how individuals' sense of ownership is affected by large-scale languagemodelstothe content generated
✔️ Suggests that there is a psychological conflict between sense ofownership andsense of authorship to the content generated by language models
✔️ Raises new questions about the concepts of ownership and authorship in the digital age

LLMs as Writing Assistants: Exploring Perspectives on Sense of Ownership and Reasoning
written by Azmine Toushik Wasi, Mst Rafia Islam, Raima Islam
(Submitted on 20 Mar 2024 (v1), last revised 22 Apr 2024 (this version, v3))
Comments: Published on arxiv.
Subjects: Human-Computer Interaction (cs.HC); Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Computers and Society (cs.CY); Machine Learning (cs.LG)

code：

The images used in this article are from the paper, the introductory slides, or were created based on them.

Summary

In writing,it is said that thesenseof ownership (=sense of being one's own artifact)is strengthened byour thoughtful time andeffort, creating an attachment to the work. However, the use of writing assistants, such as the large-scale language models that have been discussed in recent years, can cause conflicts as to whether some content can be called one's own work, since it is no longer our own creation.

For example, if you use a large-scale language model for a particularly creative task,peoplewill appreciate the contribution of the large-scale language model,even though all tasks are equal in effort, regardless of whether they are creative or notfor the large-scale language model.Also,as an author, you are free to claim rights to thecontent generated by thelarge-scale languagemodel, although you are unlikely to claim thecontent as completely your own (sense ofownership).

Through research, this paper delves into these issues and seeks to better understand human-computer interaction and improve writing support systems through the task of writing.

The sense of ownership is deeply rooted in our cognition and behavior from an early age. It implies that an individual has authority over a particular thing, content, asset, or right. It also means a sense of responsibility that one feels as an owner or a member of a team, project, or organization. Beyond the legal or formal aspects of this is the psychological and subjective experience of the individual. In writing and creative tasks, the degree of originality and effort required to complete a task is directly related to a sense ofownershipor authorship.

These senses in writing are more than simply holding pen and paper. It is a deep connection between thought and words that invests us with ourselves and fosters a sense of party to our own ideas. This is a process that fosters a relationship with our creations that goes beyond simply claiming credit. The stroke of a pen gives us the power to express our identity on the page and share a story that communicates our essence.

Large-scale language models such as ChatGPT, Gemini, and Microsoft Copilot are increasingly being used, and they are being used effectively in education and business. They have been effective in a variety of writing tasks, including creative narratives and essays, academic writing, and legal writing.By utilizinglarge-scale languagemodels, the writing process becomes a collaborative one, blending human input with machine-generated content. This facilitates the exploration of diverse perspectives and styles and may change traditional notions of content creation.

In creative tasks, we humansoften feel astrong sense of attachment and a strongsense of ownership andauthorship, but how does the use of a large-scale language model change this sense? And how does a large-scale language model-based writing assistantaffect the perception of a sense of ownership andauthorship? These are especially important in light of the widely recognized problem of "plagiarism" (stealing text, figures, words, phrases, story lines, ideas, etc., in whole or in part from the work of others and using them as one's own in one's own work).Understandinghow individualsperceive a sense of ownership andauthorship ina context supported by a large-scale language modelcan help address issues of uniqueness and attribution in content creation.

This paper firstconsiders the question of whetherparty consciousness (i.e.,thestrength of the sense ofownership andsense ofauthorship) varies across content typeswhen using a large-scale language model as a writing assistant.For creative and non-creative tasks, the degree of involvement in idea generation and story formationhas a significant impact onsense of ownership. When a large-scale language model completes a task automatically, we tend to feel less credit for the outcome. In contrast, for routine, noncreative tasks, we may have a stronger sense of contribution from working with a large-scale language model.

Next,we examine how theuse of alarge-scale languagemodel-based writing assistantaffects the perception of a sense of ownership andauthorship. Technically,with the support of alarge-scale languagemodel, an AI can become the author of content, but may not feel the same emotional connection as if it had created the content independently. In fact, one may self-identify as the author of AI-generated content but not feel a sense of complete ownership. This discrepancy between the sense of ownership and the sense of authorship can lead to a psychological dilemma.

Understanding these nuances is critical in the development of effective writing assistants. In this paper, we surveyed 35 participants to comprehensively explore these issues and analyze their perspectives. Our goal is to develop this understanding in order to create tools that meet the needs of users and enhance the writing experience. Furthermore,it is hoped thatnavigating these complexitieswill be key to usinglarge-scale languagemodels more efficiently inreal lifeand to making them serve as a valuable aid in the creative process.

Survey Summary

Theexperimentinvolves a short-term survey of35 participantsabout theirsenseofownership andauthorshipwhen using a writing assistant.The survey asked participants the following questions

• If you use ChatGPT for non-creative writing (e.g. assignments), do you think it contributes to the content?
• If you used ChatGPT to create a story or poem (creative task),do you think it contributes to the content?
• Would you like to submit the content generated by the large-scale language model somewhere under your name? (assignments, newspaper articles, etc.)
• Do you believe that the content that ChatGPT has created for you is yours?
• Given the content generated by ChatGPT for a particular prompt, there is a complex interaction between the prompt and the resulting text. The prompt acts as a catalyst for the generation of the AI model and can determine the direction and nature of the response. Given that content is prompted and guided by human input,do you thinkthe content ChatGPT creates foryou is yours?
• While the content is initially generated by ChatGPT, subsequent modifications reflect human intervention and can be shaped to make the text more appropriate to the context or to convey a specific message. At this time,do you thinkChatGPT's modified responsesare your own?

Participantsconsisted primarily of college students (94.3%) between the ages of 18 and 24, with a minority (5.7%) in the 24-30 age group. Gender is predominantly male (65.7%), with females accounting for 34.3% of participants. Furthermore, the majority of participants (88.6%) have a background in the sciences and a minority (8.6%) have a background in the arts. Participants also rated themselves as having a high level of awareness of technology, scoring a value of 3 to 5 on a 5-point scale (5 being indicated as "technology enthusiasts" and 1 as lacking awareness of technology).

Results (of a study)

Thestudy found that when participants perceive less intervention (contribution) by the AI, such as in non-creative content such as simple assignments, theyexhibit astronger sense of partyhood (sense ofownership andauthorship). In contrast, for creative content such as poems, stories, and birthday wishes, participants felt a greater contribution by the AI and tended to show less sense of partyhood.

Thegraph belowshows the results ofparticipants'responses to the extent to which they felt they were party todifferent types of large-scale language model-based writingassistance(1 being very low and 5 being very high). In creative content, human contributions tend to be valued less and more credit is given to large-scale language models.

In creative tasks,there is a tendency to treatlarge-scale language modelsasroughlyequivalent tohuman contributors.Sense of ownership and sense of authorship involve originality, contribution, andresponsibility, butresponsibility for the content produced bylarge-scale languagemodels is less clear. On the other hand, tasks that do not require creativitytend tohave less of this problem andgive less credit tolarge-scale language modelsbecause the value of originality and contribution is not as high.Thisillustrates the difference inhow theuse oflarge-scale languagemodelsaffects thesense of partyhood (the sense of being one's own artifact) innoncreative and creative tasks.

Additionally, the study examined perceptions of sense of ownership and sense of authorship for content generated by large-scale language models. Interestingly, even though many participants in the survey did not claim ownership of content generated by large-scale language models (51.5% "no" and 31.4% "maybe"), they indicated their willingness to submit the same content as their own work (28.6% "yes" and 48.6% "maybe"). ).

This psychological conflictsuggests that there is a complex relationship between copyright and identity that needs to be explored further,as individuals do not claim ownership, but allow their names to be used.

In addition, reminding participants of the role they play in providing the prompts also indicates an increase in participants claiming ownership of the content.("Yes" from 17.1% to 28.6%, "Maybe" from 31.4% to 48.6%, and "No" from 51.4% to 22.9%). This may be a recognition that individuals are actively involved in shaping the direction of the content, prompting a re-evaluation of their sense of ownership. After making them aware that they had edited their responses to the large-scale language model, the sense of authorship skyrocketed, with 62.9% of participants claiming ownership of the content.

Insights from this study confirm the importance of collaboration in content creation, which has a marked impact on the perception of individual sense of ownership andauthorship. This understanding can enhance the effectiveness of human-computer collaboration.

Discussion

This paper reveals a psychological conflict between the sense of ownership and the sense of authorship. While people seek recognition for their work, theymay be hesitant to claim ownership. This dilemma may stem from the perception that authorship has meaning beyond mere ownership. However,when made aware oftheir own active involvement in the creation and editing of content (e.g., suggesting ideas in prompts, editing text, etc.), they begin to perceive themselves more strongly as owners and authors. This may indicate that acknowledging individual contributions is key to feeling a sense of ownership andauthorship over a work.

Furthermore, the extensive use of large-scale language models carries a number of risks, such as obscuring authorship, introducing bias, difficulty in explanation, loss of autonomy, lack of critical thinking, and outsourcing of thinking. These issues also underscore the importance of clearly distinguishing between individual contributions and those made by large-scale language models.

Although this paper addresses non-creative and creative tasks separately, a greater variety of content types should be explored. And it is hoped that this will lead to improved and more comprehensive understanding of writing support systems.

Increasing transparency and accountability of the content generated by large-scale language models and providing clear guidelines and standards will help individuals make informed decisions regarding ownership. A collaborative approach between humans and AI could also help reinforce a sense of ownership andauthorship among individuals involved in creative tasks.

The sense of ownership is also deeply tied to specific proprietary rights and ethical considerations, and has implications for legal frameworks and jurisdictions. the treatment of AI-generated content varies from country to country, and in the United States, as in the case of Thaler v. Perlmutter, a human work is required to claim ownership In the U.S., as in Thaler v. Perlmutter, a human author is required to claim ownership. In China, on the other hand, it has been decided that AI-generated content should also be protected by copyright law. Legal decisions will shape the values of copyright and ownership in the AI age. This poses new challenges to the concept of ownership in the digital age.

Summary

This paper identifies two main mental dilemmas that can be addressed by writing with the aid of large-scale language models. First, how the type of contentinfluences the individual's sense of ownership of the evaluations attributed to thelarge-scale languagemodel,and second,the link between the sense of ownership and the authorship sense claims for the content generated bythe large-scale language model. Through analysis of the survey data, we show how these dilemmas are related to the thought processes behind them. It is hoped that addressing these dilemmas will lead to a better understanding of human-computer interaction in writing tasks and promote the development of effective writing support systems.

• 
• 
• 

Categories related to this article

• Large Language Models
• Machine Learning
• Deep Learning
• Language Processing
• Natural Language Processing
• Artificial Intelligence