tech

Recently, a doctoral student at Pennsylvania State University, Lu Yao, and his collaborators, have conducted a study on generative AI.

The study's origins can be traced back to the rapid advancements in generative artificial intelligence (Generative AI) technology in recent years.

As a new frontier in AI development, generative AI has demonstrated astonishing creative abilities, capable of generating realistic images, videos, text, audio, and other rich multimedia content based on simple prompts or instructions.

Among them, large language models (such as GPT-3) and text-to-image/video generation models (such as DALL-E, Stable Diffusion) are undoubtedly the most dazzling "stars."

They have broken through the limits of human cognition, allowing AI systems to possess "imagination" and give birth to creative works beyond imagination.At the same time, the community of content creators (such as YouTubers) is actively embracing and mastering these generative AI tools, integrating them into their creative practices.

Advertisement

With the assistance of AI, people can greatly enhance the efficiency of creation, and even surpass their own capabilities, generating high-quality and distinctive content works.

To elaborate, this study began with a peer exchange in 2023.

At that time, Lv Yao and the second author of this paper, He Zhang, the third author, Jie Cai, as well as Dr. Shuo Niu from Clark University in the United States, attended the Conference on Computer Supported Cooperative Work and Social Computing (CSCW, ACM Conference on Computer-Supported Cooperative Work and Social Computing).

During the exchange, Shuo Niu mentioned an emerging phenomenon: content creators are increasingly trying to incorporate generative AI tools into their creative practices.However, what specific attempts were made? What were the effects? What opportunities and challenges exist? In this conversation, they all believed that this is a topic with great potential and expressed their interest in it.

Based on such doubts, they began to consult a large number of literature, collecting various viewpoints and first-hand materials.

Through several months of continuous attention, they gradually established the core theme of this research: a systematic exploration of the application of generative AI in content creation.

To fully understand the practice of content creators using generative AI, they designed a search strategy and downloaded a large number of related video materials from the video platform YouTube.

There were also some technical challenges in this process, such as how to filter irrelevant videos, how to obtain more diversified examples, etc. After several adjustments to the strategy, they gradually overcame these difficulties.After initial screening, they ultimately narrowed down to 68 high-quality analytical samples, covering multiple different creative fields.

These videos demonstrate the actual interaction process between creators and generative AI tools, which is crucial for the subsequent data analysis.

For this valuable first-hand material, they used classic qualitative research methods such as open coding and affinity diagram grouping to conduct a detailed analysis of the video data.

During this period, they repeatedly watched and understood each video, identifying and marking all elements related to the research question.

The biggest challenge in the process was: how to accurately understand and name new terms and new things, because the application of generative AI is a cutting-edge field, lacking a mature theoretical framework and terminology system.Due to the novelty and cutting-edge nature of generative AI applications, they often encounter disagreements and disputes over the definition of concepts when classifying and summarizing video materials.

Some AI tools, such as Synthesia, can generate both images and videos. Should they be classified as "image processing tools" or "video processing tools"?

In response to this question, team members have engaged in heated debates in the meeting room.

Some believe that the classification should be based on the final output form of the tool; however, others argue that this is too one-sided and overlooks the role and contribution of the tool in the creative process.

Ultimately, through several rounds of discussion and collision, they decided to start from the actual "behavior" of the tool in the creative process, code the AI usage presented in each video material, and then classify and name it accordingly.This seemingly simple methodological debate actually reflects their profound contemplation on the essence of AI creative practice.

Another discussion point that left a deep impression on Lu Yao was the philosophical proposition about whether "AI-generated content" truly belongs to the category of "creation" in the present.

Since AI only produces results based on prompts and rules, can it be described as "creative"?

This questioning not only sparked their reflection on the boundaries between intelligent beings and intelligence, but also demonstrated the new dimensions that generative AI brings to the traditional concept of "creation."

Although they did not provide an ultimate conclusion, each of them had their own unique insights into this philosophical proposition.After completing the preliminary coding, they refined a series of valuable findings from a vast amount of raw data through focused discussion and summarization, and then corresponded them one by one to the research questions set.

For example, on the question of "What AI tools do content creators use?", they summarized six major types of tools, such as language models, image processors, and video processors, and listed specific tool examples under each type in detail.

On the question of "What role do AI tools play in creation?", they summarized eight main types of AI contributions from the videos, such as "generation," "upgrade," and "suggestion."

Finally, the relevant paper was published at the top conference on human-computer interaction, ACM SIGCHI CHI Conference on Human Factors in Computing Systems, with the title "A Preliminary Exploration of YouTubers' Use of Generative-AI in Content Creation" [1].

If this result can be noticed and recognized by the industry, it can bring some positive value to the relevant fields.The following advancements are expected to be promoted:

Firstly, the optimization of content creation tools.

Based on the research team's meticulous analysis of creative practices, developers of generative AI tools can more accurately grasp user needs and develop integrated solutions that conform to the habits of creators.

For example, seamlessly integrating various AI capabilities such as visual generation, audio generation, and video editing into the same creative platform, thereby simplifying the creative process and improving work efficiency.

In addition, through this study, the team has gained a deeper understanding of the specific roles and contributions of generative AI in creation.This can provide a basis for optimizing the human-computer interaction interface of the tool, further enhancing the collaborative efficiency between humans and AI.

For example, more freedom for feedback and correction can be added to the tool interface, as well as a more intelligent hint mechanism, to help creators adjust and control AI output results efficiently.

Secondly, promote artificial intelligence education.

Generative AI is quietly changing the rules of the content creation game. Compared with traditional manual creation, the AI-assisted creation model is undoubtedly more efficient and has a lower threshold.

In the future, how to efficiently use these emerging tools to ensure that they are not marginalized in fierce human-computer competition will be a practical issue that content creators urgently need to address.This study can provide a reference for designing AI education and training courses, helping creators gain the necessary AI literacy.

Including skills for efficiently calling and orchestrating various AI capabilities, standards for judging the quality of AI output, and human-computer division of labor modes in creation, thus smoothly integrating into the new normal of human-computer collaboration.

Once again, improve content review and legal policies.

With generative AI rapidly permeating various aspects of content creation, the demand for reviewing the quality of AI-generated content will also become increasingly urgent.

This study on the specific use scenarios and output forms of generative AI in current content creation can provide a basis for formulating relevant review standards and methodologies for identifying AI-generated content.Based on the conclusions of this study, in the field of video creation, YouTube creators generally use AI tools to generate virtual character images, which are then used to record "character" videos.

During the review process, it is possible to identify whether the characters in the video are real people based on detailed features such as character movements and speech.

On the other hand, the influx of large-scale AI-generated content also poses new challenges to the existing intellectual property legal system.

Who owns the copyright of AI-generated works? Is the creator's contribution sufficient to obtain copyright? Can the algorithm itself enjoy some rights?

Such legal disputes will reveal more and more gaps, which are urgently needed to be standardized and clarified through judicial precedents and the supplementation of laws and regulations.This study may provide some empirical support to clarify copyright disputes and explore relevant legal responsibilities, contributing to the formulation of scientific and reasonable AI content governance policies.

Finally, it gives birth to new business models.

Generative AI is undoubtedly a kind of productivity. While empowering content creators, it will also give birth to new business models.

Relying on the power of AI, a large number of new creators will be able to join the tide of content creation more conveniently, thereby intensifying the "decentralization" trend in this field.

Under this trend, innovative profit models may emerge. It can be anticipated that to meet the needs of the AI-assisted creative process, new emerging business forms such as a market for trading creative materials, a supply chain for auxiliary creative services, and intermediate products for "adding value" and selling generated content may emerge.For example, there was a YouTuber who once demonstrated how to create novel and interesting video content based on AI material templates through secondary editing, and generate revenue by selling these contents.

This study may accumulate theoretical foundations for this kind of "disintermediation" new entrepreneurship, and point out the direction for finding new profit models.

In addition, Lv Yao added: "I want to emphasize again the profound impact of generative AI on the field of content creation. This impact is not only reflected in the transformation of creative tools and processes, but also gradually permeates into multiple aspects such as content ecology, business models, and legal policies."

Generative AI provides content creators with a brand new "intelligent assistant", which is expected to liberate the productivity of creators and give birth to more creative works.

At the same time, the proliferation of AI-generated content may also have a certain impact on the content quality and intellectual property rights of the platform.How to strike a balance between innovation and control, to make good use of AI's advantages while avoiding its potential risks, is an urgent issue that needs to be addressed.

Therefore, Lu Yao hopes that this study can inspire more thinking and discussion among peers on this topic.

At the same time, generative artificial intelligence and its application in the field of content creation are developing and evolving rapidly, and Lu Yao believes that this achievement is just the beginning of a long journey.

In the future, they will continue to closely follow this cutting-edge hot topic and strive to make new breakthroughs in the following directions:

Firstly, expand the scope of research.At present, they have only analyzed YouTube videos, but the use of generative AI extends far beyond that.

Therefore, their plan is to expand the scope of research to other mainstream content platforms, such as TikTok, Instagram, X, etc., in order to gain a more comprehensive understanding of the trends in content creators' practices.

Secondly, track technological development.

The emergence of AI innovations such as the second generation of large language models and visual question-answering models is endless, and will further change the rules of content creation.

Therefore, it is necessary to keep up with these developments in a timely manner and analyze their impact on creative practices and the content ecosystem.Thirdly, delve into the creator's experience.

Currently, they mainly focus on the aspect of creative output, but they lack a deep understanding of the actual experience and feelings of content creators.

In the future, Lv Yao and others plan to conduct qualitative research, collecting first-hand data through interviews and other methods, to fully grasp the users' personal feelings when using generative AI.

Fourthly, formulate design principles.

On the basis of understanding creative practice, they hope to summarize some design principles and best practices to provide guidance for tool developers and content platforms, so that AI-assisted creative tools can better fit the needs of the use scenarios.Chapter Five, discusses ethics and policy.

Generative AI, while driving content innovation, may also bring about ethical and legal issues such as copyright disputes and intellectual property controversies. It is also expected to conduct relevant research on this to contribute ideas and strategies to address these challenges.