ChatGPT was launched by OpenAI on November 30, 2022. Five days later, the number of users exceeded one million, and by January 2023, the user count surpassed 100 million, making it the fastest-growing consumer application. This company has been deeply involved in AI for eight years, and they are just one of the many companies in the vast AI ecosystem, which has yet to fully surface.
It can be said that GPT-4 heralds the dawn of the AI ecosystem era.
In the gaming field, the application possibilities of AI in gaming, its commercial value, user value, and the value to developers are worth pondering.
Application of AI in the Gaming Ecosystem
Game PME Flywheel
In gaming, the complexity can be simplified into three major modules: production, marketing outreach, and the final service target: player experience, forming a flywheel cycle.
In the PME triad, we can further subdivide and analyze the specific application scenarios of AI in each module from the current mainstream job function perspective. The current state of AI and gaming ecosystem content is shown in the figure below:
AI Ecosystem Applications in Gaming PME
1. P Production: Efficiency Enhancement, Art as the First Breakthrough
Ball Playing Girl (Ray) – MidJourney
In terms of production, art is currently the first breakthrough. The AIGC art wave is comparable to the invention of the camera. Currently, AIGC’s artistic resources excel in intention drawing and concept design, such as Midjourney, which helps game artists try different abstract content in seconds. This not only helps artists draw inspiration but also provides an easy path for non-artistic planners and other roles to create intention drawings and quickly align with art design colleagues, significantly reducing upfront design and alignment costs.
However, regarding fully standardized commercial use, stability and variability still have obvious shortcomings.
StableDiffusion, with its relatively open parameter exposure, performs better. The current main production methods are:
(1) Simple image generation: text-to-image, image-to-image (line art), combining different main models to produce drafts, followed by hand-drawn corrections;
(2) Pose generation: tools like OpenPose draw poses, ControlNet accurately outputs shapes based on poses, combined with text prompts to determine content tendencies, and fine-tuned with Lora for details like facial features, clothing, and accessories, or corrected with local redrawing tools, finally outputting with SD’s built-in or external super-resolution tools;
(3) Professional hand-drawn image generation: precise hand-drawn line outlines, with ControlNet coloring, obtaining drafts and then adjusting poses and details using Photoshop with large color blocks, continuously iterating with image-to-image methods;
(4) Style-switching image generation: MidJourney generates concept images, imported into StableDiffusion for redrawing in a specified style, or StableDiffusion generates reference images, imported into MidJourney for diverging different styles.
Overall, the common art resource part has preliminarily established the capability for AIGC pipeline production, with room for improvement and opportunities.
Ball Playing Girl (Gold) – StableDiffusion
In terms of audio, I have tested multiple AI technology tools. Except for voice-changing tools like Voice AI, AI-generated audio, such as Amper Music and MuseNetAI for background music, is still immature, with a strong patchwork feel, unable to convey emotions well through music. Due to information limitations and space constraints, I will not elaborate further.
AI Writing Game Document Outlines
In planning and programming, there are standardized areas with successful cases and materials that can replace “reinventing the wheel” work. Here, ChatGPT 3.5 performs average, while GPT-4 shows good performance after multiple inputs to refine context and boundaries. For example, basic numerical design, worldview framework design, and gameplay scheme design can yield reasonable and logical proposals, but creativity is relatively lacking. In basic programming code, it can act as a good multi-language “translation tool” and assist in coding tasks, saving a lot of character input costs in small module functionality. Testing shows that there are still many bugs, but similar to planning cases, returning problematic parts to GPT for correction can generally yield correct solutions.
GPT4 Assisting Code Generation
2. M Marketing: Automated Analysis, Lowering Barriers
Traditional public opinion analysis and information collection require crawling vast amounts of data and processing it through NLP sentiment analysis, clustering analysis, etc., which has relatively high barriers and costs. Large language models like ChatGPT greatly optimize and integrate the above processes. When fully networked with plugins, multilingual, and multi-channel public data analysis will efficiently assist in gaining insights into player and market conditions, and provide effective automated report analysis for public opinion and anomalies.
Additionally, AI’s powerful summarization and information extraction capabilities yield significant results in this area. Non-public data, when fed for analysis, report generation, and predictions, will see greatly enhanced efficiency in extraction and summarization. At the same time, we must remain vigilant about information leakage and ensure the security of confidential information.
3. E Player Experience: AI Promotes the Emergence of New Gameplay, Narrative, UGC, and Open World AI Gameplay Will Be the First to Benefit
Disco Elysium: Dialogue Exploration Puzzle Game
AI bot technologies, including adversarial AI and social ecological AI, have always been key areas of research and application in the gaming industry, now more prominently exposed to the public. Human information decision-making and “wisdom” largely depend on the reception, analysis, and processing of natural language, and large language models will promote AI bots to present more vivid performances in multiple dimensions. For example, changing the way players interact, allowing players to convey more open information through natural language interactions, which can be processed by AI, enhancing the information collection capabilities of AI bots. AI bots can provide feedback through language output, breaking the binary interaction choices for players.
With further technical advancements, this could serve as a basis for AI bot GOAP goal establishment, with principles similar to HuggingGPT, where a central driver propels AI’s behavior to be more intelligent, providing players with a more vivid AI experience.
Hyperparameter GAEA (Image Source: Hyperparameter Official Public Information)
The current stage case, such as the latest release from Hyperparameter company GAEA, focuses on creating AI NPCs that have “life” and establishing feedback mechanisms. I believe the establishment of feedback mechanisms relies on three pathways: environmental labels and values, interaction behavior labels and values (PvP/PvE/EvE), and the labels and values of language interactions from large language models (LLM) as they mature. Once this holistic system is coupled, the final effect may far exceed our expectations.
4. PME Integration: The Comprehensive Integration of PME and AI Will Rewrite Game Production Methods and Potentially Promote Changes in Production Relationships
With the combination of various parts of PME and AI, there will be significant opportunities for improvement in production efficiency and standardization, promoting the emergence of new gameplay and providing players with new experiences. This may be the greatest boost before the widespread adoption of new device carriers (XR), and the comprehensive integration of PME and AI could rewrite game production methods, potentially driving changes in production relationships.
For example, 20 years ago, the construction field primarily relied on hand-drawn blueprints and construction drawings, whereas now it has largely achieved full digitization and informatization. The current game industry production methods may be comparable to the hand-drawn era of 20 years ago. This progress and change significantly lower the entrance barriers and production costs, enhancing the proportion of innovation and thinking abilities. UGC’s participation in gaming may transform from a “play style” to a fusion of “production style + play style,” with the implementation of individual value advocated by web3 (decentralization + ownership) having concrete carriers, reshaping the definition of UGC.
Specific Applications of AI Tools in Game Production
Existing AI Tools in the Production Module
1. Planning: Invisible Assistant
The enhancement of AI in planning currently focuses on two aspects: first, the integration as a basic computational tool, and second, helping to inspire ideas and provide a large framework design.
The former refers to the complex process of integrating multiple software and data, which can be handled by AI, such as adjusting certain parameters in a complex numerical table. Many computational experiments require repeated adjustments of various values to achieve “balance.” Inputting overall data into ChatGPT can clearly outline numerical adjustment needs, allowing AI to provide different calculation and adjustment plans as multiple previews, significantly saving time costs, which can then be manually corrected to gradually meet practical needs.
The latter, such as copywriting/narrative planning, can involve providing keywords to one of the large models like ChatGPT, Wenxin Yiyan, or Tongyi Qianwen, allowing AI to assist in writing and obtaining inspiration, and then adjusting and expanding based on that, even extracting keywords from each chapter to generate prompts for MidJourney to create corresponding illustrations, forming a well-rounded planning document.
2. Art: Embracing Until It Becomes a Part of Us – The Development Stages of AIGC
AI’s application in art was partially discussed in previous sections, but regarding overall development, I believe there are several stages:
(1) Technical breakthroughs as the main focus, art effects: “Not unusable”;
(2) “Player co-creation”, pioneers running into the field, iterating and improving art effects – “Sè sè is the first productivity”;
(3) Professional personnel entering the public eye, “Prompt/AIGC engineers”, multiple technologies merging, plugins springing up like bamboo shoots after rain, accelerating the overall enhancement of technology and quality;
(4) Deep integration, self-training AI models for art becoming standard, shifting from “Please provide your portfolio” to “Please provide your SD model set”;
(5) The chain is connected, with AI 2D and 3D tools being perfected, diverging into two paths: one being extremely convenient natural language input generating various conceptual, non-standard art content, and the other being extremely complex, with massive parameter exposure for the professional development path of “parametric art”.
Currently, it appears (April 2023) we are at stage 2.5, focusing on explaining the training principles of stage 4:
We can compare the growth training process of an artist. For example, an artist needs to practice the head, hands, feet, and torso in a “modular practice” and “overall combination practice,” and strengthen the important parts of the facial features and hands. Once the basic drawing functions are developed, they gradually undergo “generalization training” on different characters’ postures and expressions, moving from shape writing to spirit writing, ultimately reaching a stage of rote memory through massive data and training before engaging in artistic creation. This process undoubtedly takes years, while AI technology’s advantage lies in inputting the results of drawing each “part” once for many groups, and leaving the combination and generalization of different subjects to AI, greatly reducing trial and error and repetition costs during collaborative processes – we draw the “imagery” in our minds and synchronize with the “client”. If the client believes it does not match their mental “imagery,” the cost of repeated adjustments is significantly reduced.
It is evident that the integration of the AI ecosystem has largely conducted a broad screening of the art position. The deeper the drawing foundation and the more mature the artistic style, or the more unique the style, the greater the amplification of one’s value, and many opportunities exist in more complex scene domains. By leveraging AI, productivity can be expanded from having “only one pair of hands,” while merely acting as a drawing labor force faces iteration.
3. Client: I Defeat Myself
Although AI currently cannot fully replace programmers in writing complex game clients, it can assist programmers in generating some simple code or expanding based on existing code. Using AI to assist in writing game client code has shown good results, such as the GitHub Copilot tool, with the following methods and steps:
GitHub Copilot is an AI-based code completion tool that can assist in writing game clients. GitHub Copilot is trained on OpenAI’s large code repository and can understand various programming languages and frameworks, such as Python, JavaScript, Unity, and Unreal Engine.
(1) Prepare the environment and install GitHub Copilot, which will integrate into the programming environment (e.g., Visual Studio Code);
(2) Write game logic: When you encounter areas needing assistance during writing, you can leverage GitHub Copilot. Simply input some relevant keywords or comments, and GitHub Copilot will generate corresponding code suggestions based on its understanding;
(3) Optimize the code: AI-generated code may not fully meet our needs, so it needs to be checked, corrected, and adjusted to the desired solution. No code is 100% perfect; review and validation remain critical processes;
(4) Test and iterate: Run the client for appropriate testing. If issues arise, fix errors and optimize the code. If the newly written part has problems, you can try submitting it to ChatGPT for help in checking.
Currently, tools like GitHub Copilot that assist in writing game client code are flourishing. While they cannot yet complete writing tasks independently, AI assistance will play an increasingly important role in the game development field as technology advances.
4. Gameplay Updates Brought by AI and Gaming
|
|
|
|
|
AI dialogue will shape the world perception in RPGs more clearly and specifically, for example, enhancing NPC relationship development, NPC dialogue understanding of the world, and the diversity of NPC interaction behaviors. RPG worlds will become more vibrant, potentially extending to deeper trading, socializing, and combat gameplay with NPCs, such as in Hogwarts and Dream of the Red Chamber.
|
|
|
Party games like Goose Duck Kill may have AI players become possible due to the high threshold for gathering enough real players, and too many players may diminish the experience. The emergence of AI players resolves such issues and may lead to more complex gameplay variants.
|
|
|
For instance, Tom Cat + ChatGPT, Genshin Impact characters + ChatGPT, character and animal nurturing gameplay gains new interaction and emotional expression through AI dialogue, spawning “portable” independent gameplay and gameplay updates.
|
|
|
Text-based puzzle games will become more open in participation, solving methods, and reward mechanisms, potentially generating countless open endings. For example, Disco Elysium-type games.
|
|
|
Dungeons and Dragons already has an initial GPT version integrated to act as the DM. It can be anticipated that the evolution of AI in generating artistic resources will transform traditional text-based running groups into visually rich, emotionally resonant experiences, with player AI changing voices for immersive role-playing, elevating the gaming experience to a new level.
|
|
|
Nurturing paths in games like Uma Musume may be generated by AI, providing different interactive gameplay, where narrative text is generated based on players’ foundational data and continuous input, combining fixed and dynamic content, truly allowing players to nurture their own “idols”.
|
Outlook on AI and Gaming Development
(1) Copyright of AIGC:
a. United States: AIGC-generated content is not protected by copyright, and in cases of low similarity, it is not considered infringement of source material.
On February 21, 2023, the copyright registration of the AIGC comic “Zarya of the Dawn” was permitted, but only for non-AIGC parts, namely the selected and coordinated arrangements. The copyright of this work was initially approved but later rejected by the U.S. Copyright Office upon learning it was AIGC content. After further consideration, the scope of copyright protection was divided. Overall, policies are still in a transitional phase. We need to focus on two aspects: first, copyrights only protect human creative works, which can be understood as belonging to “anyone” in the absence of protection; second, the risk of infringement still exists. AI-generated training sets typically use millions of images, and currently, there are no infringement issues based solely on similarity judgments, but large-scale usage cannot be ignored, which may change the definition of infringement in the future.
b. European Union: A four-step method determines whether AIGC content qualifies as a “work”.
The European Commission’s 2020 report proposed a “four-step test” to judge whether AIGC meets the qualification of a “work”:
Step 1 – Literary, artistic, and scientific fields;
Step 2 – Human intellectual activity;
Step 4 – Expression.
Currently, AIGC works generally do not meet the second and third criteria and cannot obtain copyright permission and corresponding protective rights.
(2) Data Security and Cultural Security of Large AI Models
a. Italy: On March 31, the Italian National Privacy Authority officially ordered the ban of ChatGPT, accusing OpenAI of “illegally collecting personal data”.
It is worth noting that this “temporary ban” will remain in effect until OpenAI complies with the EU’s landmark privacy law – the General Data Protection Regulation (GDPR).
b. China: On April 11, the Cyberspace Administration of China issued a notice soliciting public opinions, mainly stating that AI content providers should bear the responsibilities of content producers and personal information protection obligations. Before providing services to the public using generative AI products, it is necessary to report security assessments to the national cyberspace department and bear responsibility for content review.
2. Outlook on AI Ecosystem and Gaming Development
The current overwhelming scene of the AI ecosystem may still be in the single-point explosion phase—on the eve of AGI. I believe the development of the AI ecosystem will undergo at least four stages:
(1) Single-point explosion;
(2) Large-scale vertical applications and multi-modal applications;
(3) Functional integration and coupling, giving rise to new functions and experiences;
(4) Industry reshuffling, with leading monopolies and segmented leaders.
Currently, we are transitioning from stage I to II, as we can see that large language models like ChatGPT, MJ, and SD have ignited user groups after sufficient single-point explosions, sparking a wildfire in various vertical fields. For example, the Microsoft Office suite combined with GPT will continue to refine until it fully adapts and addresses specific pain points.
GPT-4 has already shown preliminary capabilities in multi-modal image processing, and in subsequent versions 5/6/7, it will comprehensively cover audio and video processing capabilities, taking a new step in processing mixed “information” and integrated inputs in various carrier forms. It is evident that the gaming industry, as a frontier of general entertainment and integrated cutting-edge technology, will be the best to lead the industry and the most significantly disrupted industry.
In the AI ecosystem + gaming industry, the three ends of PME will gradually welcome large-scale applications in vertical segmentation, with each part of P, M, E, such as production, planning, client, and art, quickly covering from production to updates in player experiences.
Overview of HuggingGPT (Image Source: arXiv Paper HuggingGPT…)
The framework connecting large language models (LLM) like ChatGPT with various AI models in the machine learning community (such as Hugging Face) to solve AI tasks has begun to show results, as seen in the above image of Microsoft’s Asia Research Institute’s HuggingGPT, which can cover numerous complex AI tasks across different modalities and fields. Additionally, many integration tools such as BabyAGI and AutoGPT are emerging, hinting at the early signs of AGI.
Currently, the stage of scattered AI technologies flourishing and competing is primarily focused on extending and transforming first-stage technologies. After experiencing sufficient explosions, it will gradually converge. New technologies and functions will combine in ways that have never been systematically designed, leading to emergent innovations, similar to the breakthroughs brought by the integration of CV, NLP, etc. These breakthroughs will bring new functions and gaming experiences, not only providing convenience to creators but also genuinely impacting every player.
3. The Individual in the AI Era
As a new era arrives, individual attitudes and choices become particularly important. I suggest two relatively integrative attitudes based on personal circumstances:
a. Maintain an Open Mindset: Embracing the AI era requires an open and optimistic mindset without fear of the challenges posed by new technologies. AI is a great tool; “A gentleman uses tools wisely.” From taming animals, creating carriages, bicycles, to fuel cars, and new energy vehicles, we have always kept pace with the times. As the saying goes: “The car era has arrived; what we should do is get a driver’s license and learn to drive it.”
b. Update Knowledge Structure: In the AI era, individuals need to continuously update their knowledge structures, learning interdisciplinary knowledge, especially in AI-related fields (such as computer science, data analysis, statistics, etc.). Acquiring knowledge has become extremely convenient, and we can actively understand or master AI-related skills (such as basic programming, machine learning algorithms, application development, etc.) according to our circumstances.
c. Follow the Trend of the Times, Think Deeply, and See Through the Essence: On one hand, continuously pay attention to the dynamics and trends in the AI field to keep pace with the times; on the other hand, remain alert to chasing fleeting trends, thinking deeply about the essence of needs, the essence of problems, and the essence of value chains, viewing the substance behind trends from first principles. In addition to logical and scientific thinking and skills, non-logical insights, empathetic thinking, and cultural literacy will become more precious. We should avoid defining and cutting in with disciplines, keeping “people-oriented” and thinking from perspectives of care and love; perhaps we will discover and gain something different.
(2) Discover Application Scenarios, Solve Specific Problems, and Achieve Personal and Social Value
a. Immerse and Understand the Industry: Conduct in-depth research and understanding of the pain points and needs in the industry, discovering specific problems that AI technology can solve. Individuals can leverage their interests and strengths to realize the application and breakthroughs of AI technology in their areas of focus, such as considering whether to train a vertical GPT model for RPG categories to bring value to companies and players.
b. Drive Innovation and Development: By excavating new application scenarios for AI technology, such as ControlNet developed by Chinese PhD student Lvmin Zhang at Stanford University, we can explore better applications in the gaming field to promote overall industry innovation and development while also enhancing personal value.
c. Popularize AI, Be Both Teacher and Friend: As a user in the AI field, once familiar with it, one can choose to help more people understand and use AI technology to solve practical problems in life. This will also be very helpful for our own enhancement; “In a group of three, one must be my teacher.”
This article briefly discusses the applications and future development prospects of the AI ecosystem in the gaming industry, providing a panoramic analysis of the applications in the PME triad of the gaming industry, sharing the current technical applications of AI in the P end, and finally discussing the regulations and risks of AI, offering insights into the development of the AI ecosystem + gaming industry, along with some suggestions for individuals to respond to the AI wave, for reference only.
I hope this aids gaming industry practitioners in gaining a comprehensive understanding. Perhaps this is a seed, helping you grow a small sprout for future research on AI + gaming; or perhaps this is the eagle eye of Assassin’s Creed, allowing you to survey the vast landscape from a broader perspective; or maybe this is an “investment report,” attracting a “planter” to the AI ecosystem + gaming, watering it and nurturing the industry’s development, bringing value to society and players.
#Written in April 2023, the above content is for reference only.
Game Grape is hiring content editors,
Click “Read the original text” for details
Industry Recruitment Season | 2023 Talent Trends | AI and Layoffs
Starry Sky Railway | A Harsh Year for ACG | Industry Traps
Exclusive Interview with Shen Li | Teaching History with Games | AI Art Pricing
Click the card below to set the public accountas a star (Process as shown in the image above)
to receive notifications and the complete cover in real-time!
