The author of this article: Waylon, original title: "Game Ecological Panorama Analysis in the AI Era - Challenges and Opportunities, Applications and Development Prospects", the head image is from: "Disco Elysium" game screenshot
ChatGPT is on fire, some people experience it first, and some people know it later. ChatGPT was launched by OpenAI on November 30, 2022. Five days later, the number of users exceeded one million, and more than two months later (January 2023), the number of users of ChatGPT exceeded 100 million, making it the fastest growing consumer app. This company has been deeply engaged for 8 years, and they are only one of the companies in the AI ecosystem, and the behemoth of the AI ecosystem has not really surfaced.
Perhaps it can be said that GPT4 indicates that the era of AI ecology has slowly opened.
In the field of games, what are the possible applications of "AI + games"? What is the value to business value, user value, production developers? It's worth thinking about.
First, the application of AI in the game ecosystem
Picture: Game PME Flywheel
The game aspect can be abstracted into three modules, the first two modules are production, marketing reach, and the third final service object - player experience, which constitutes a flywheel cycle.
At the three ends of PME, we can further subdivide and analyze the specific application scenarios of AI in each module from the perspective of current mainstream job functions, and the content of AI and game ecology at this stage is shown in the figure below:
Figure/Application of AI ecology on the game PME side
(1) P Production end: efficiency enhancement is the mainstay, and art is the first breakthrough
Photo: Bomb Girl (Ray) - MidJourney
From the perspective of the production part, the current art is the first breakthrough, and this wave of AIGC art is no less than the invention of the camera.
At present, AIGC's art resources are outstanding in intentional drawing and concept design, such as Midjourney, which helps game art designers try different abstract content in a few seconds, on the one hand, it can help art designers draw inspiration, on the other hand, it can provide other positions without art skills to easily make intentional maps, quickly align with art design colleagues, and greatly reduce the cost of early design and alignment.
In terms of complete standardized commercial use, stability and variability are still obviously insufficient. StableDiffusion's more open parameter exposure is relatively better. At present, the main production methods are:
- Simple drawing: Wen Sheng diagram, Tusheng diagram (line draft), combined with different main model output, get the first draft, and then hand-drawn correction;
- Pose drawing: Open pose and other tools to draw the pose, Controlnet accurately outputs the shape according to the pose, combines the text Prompt to determine the content tendency, cooperates with Lora to fine-tune details such as faces, clothing, jewelry, etc., or combined with local redraw tools to correct, and finally outputs with SD's built-in or external super-resolution tools;
- Professional hand-drawn drawing: accurate hand-drawn line sketch outline, Controlnet color filling, after getting the first draft, use Ps to paint large color blocks to adjust the posture and details, and return to use the picture to continue iteration;
- Style switching map: MidJourney produces a concept map and imports StableDiffusion to redraw to a specified style, or StableDiffusion takes out a reference map and imports MidJourney to diverge different styles;
In general, the art general resources part initially has the ability of AIGC pipeline production, and the problem is to improve space and opportunities.
Photo: Explosive Girl (Gold) - StableDiffusion
In terms of audio, the author tested a variety of AI technology tools, in addition to the more practical of voice change (such as Voice AI), AI-generated audio, such as Amper Music, MuseNetAI generated background music, etc., the results are not mature, the sense of patchwork is strong, can not well convey emotions through music, considering the information limitations, and the reason for length is no longer expanded.
Figure/AI to write the game document outline
In terms of planning and procedures, in some areas with a high degree of standardization and already having successful cases and materials, it can replace the work of "reinventing the wheel". Here ChatGPT3.5 performs average, and GPT4 performs well after multiple inputs, refining context and boundaries. For example, basic numerical design, world view framework design, and gameplay scheme design can get a good and logical solution, but the creativity is relatively insufficient.
Similar in terms of basic program code, it can be used as a better multi-language implementation "translation tool", as well as code assistance work, in the function of small modules can save a lot of character input costs, there will still be more bugs after testing, but similar to the plan, the problematic part returned to GPT prompt correction, basically can get the correct solution.
Figure/GPT4 assists in code generation
(2) M marketing side: automated analysis, lower the threshold
Traditional public opinion analysis and information collection need to collect massive information through crawlers and other methods, and then process it through NLP sentiment analysis, cluster analysis, etc., and summarize and interpret the content, which has a relatively high threshold and cost.
Large language models such as ChatGPT greatly optimize the integration of the above processes, and after being fully networked with plug-ins and other methods, multilingual and omni-channel public data analysis will efficiently assist in the insight of players and market conditions, and effectively analyze public opinion and changes.
In addition, AI's powerful ability to summarize and refine language information is effective in this aspect, and the efficiency of non-public data will be greatly improved in analysis, report presentation, prediction, and refinement and summary after feeding. At the same time, we need to be vigilant about information leakage and pay attention to the security of confidential information.
(3) E Player experience: AI promotes the generation of new gameplay, and narrative, UGC, and big world AI gameplay will benefit first
Photo: "Disco Elysium" Dialogue exploration puzzle game
Confrontational AI and social ecological AI in AI bot technology have always been the focus of research and application in the game industry, and this time it has also been exposed to a greater extent to the public.
Human information decision-making and "wisdom" are to a large extent the reception, analysis, and processing of natural language, and the big language model will promote AIbot to present more vivid performance in multiple dimensions. For example, by changing the way players interact, players will be able to convey more and more open information through natural language interaction, which can be received and processed by AI to improve the information collection ability of AI bots. And the AI bot uses language output as feedback, breaking the player's choice of either/or interaction.
After further technical improvement, it can be used as the basis for the establishment of AI bot GOAP goals, similar to HuggingGPT in technical principle, and the behavior of AI driven by the hub is more intelligent, bringing players a more vivid AI experience.
Figure/Hyperparameter GAEA (Image from the public information of the Hyperparameter official website)
The current cases, such as Hyperparameters' newly released GAEA, are focused on creating "living" AI NPCs and establishing feedback mechanisms.
The author believes that the establishment of the feedback mechanism depends on three ways: the label and value of the environment, the label and value of the interaction behavior (PvP/PvE/EvE), and the label and value of the language interaction after the maturity of the large language model (LLM), once the coupling of this holistic system is completed, the final effect may far exceed our expectations.
(4) PME integration: The full integration of PME and AI will rewrite the production mode of games, which may further promote the transformation of production relations
After combining pan-AI on the PME side, there will be significant opportunities to improve production efficiency and standardization, and promote the generation of new gameplay and bring players new experiences. It may be the biggest boost before the popularity of new device carriers (XR), and the full integration of PME and AI will likely rewrite the way games are produced, further promoting changes in production relations.
For example, 20 years ago, the construction field was basically dominated by hand-drawn paper and construction drawings, but now it has basically achieved comprehensive digitalization and informatization, and the current production method of the game industry may be equivalent to the hand-drawn era 20 years ago. And this progress and change, so that the entry threshold and production costs are greatly reduced, more enhance the proportion of innovation and thinking ability, UGC game participation mode may change from a "way of play" to a "mode of production + way of play" integration, web3 advocated by the realization of individual value (de-centric + ownership) has a specific carrier, the definition of UGC will be reshaped.
Second, the specific use of AI tools in game production
Figure: Existing AI tools at the production end
(1) Planning: invisible assistant
The improvement of AI in planning is currently mainly focused on two aspects: one is the integration as a basic computing tool, and the other is to help drive inspiration and give a large framework design.
The former refers to the complex multi-software, multi-data processing and integration process, which can be handled by AI, such as the need to adjust certain parameters of a complex numerical table. Among them, there are many calculation experiments, such as the need to adjust the values dozens of times to achieve "balance". The overall data can be input into ChatGPT, the numerical adjustment requirements can be clearly delineated, and different calculation and adjustment schemes can be given by AI as a multi-scheme preview, which greatly saves time and costs, and then gradually meets the landing requirements through manual correction.
The latter, such as copywriting/narrative planning, can be given keywords, handed over to ChatGPT, Wen Xin Yiyan, one of the general meaning of one of the big models, written by AI, get inspiration, and then adjust and expand on this basis, and also give each chapter to it to refine keywords and generate Prompt, input into MidJourney to generate corresponding illustrations, and form a plan with pictures and texts.
(2) Fine Arts: Embracing until it becomes a part - the development stage of AIGC
The application of AI and art has mentioned part of the pipeline process in the previous article, and for the overall development, the author believes that there are several stages:
- The first stage: technical breakthrough is the mainstay, supplemented by artistic effects.
- The second stage: "player co-creation", the striker runs to enter, iteratively perfecting the art effect.
- The third stage: professional personnel enter the public vision, "Prompt/AIGC engineers", multi-technology integration, plug-ins have sprung up, and technology and quality improvement have been fully accelerated.
- The fourth stage: deep integration, art self-training AI model becomes standard, "please provide your portfolio" becomes "please provide your SD model set".
- The fifth stage: the chain is opened, AI 2D and 3D tools are perfected, and it is divided into two paths, one is extremely convenient natural language input to generate various conceptual and non-standard art content, and the other is the extremely complex and massive parameter exposure of the "parametric art" professional development path.
At present, it is in the 2.5 stage (April 2023), focusing on the model training principle of stage 4:
We can compare the growth training process of artists, taking the human body as an example, artists need to carry out "sub-module exercises" and "overall combination exercises" on the head, hands, feet, and torso, and "intensive training" on the facial features and important parts of the hands, after the basic drawing of the body function is trained, gradually "generalized training" on the posture and demeanor of different characters, gradually upgraded from writing to writing God, and finally after massive data and training, you can achieve silent writing, and then carry out artistic creation.
This process is undoubtedly carried out in "years", and the advantage of AI technology is that we can input the drawing results of each "part" into multiple groups at a time, and then the combination and generalization of different objects are handed over to AI operations, which greatly saves the trial and error and repeated costs in the collaborative process - we draw the "image" in our mind and synchronize it with "Party A", and "Party A" thinks that it is inconsistent with the "image" in its mind, and the cost of repeated adjustments.
Obviously, the integration of AI ecology, to a large extent, the position of art has been screened on a large scale, the deeper the painting skills, the more mature the art style controlled, or the more unique the style, the greater the value of itself, and there are many opportunities in the field of more complex scenes, through the leverage of AI will only "a pair of hands" of the top productivity expansion. Manpower, who simply acts as a mapping laborer, faces iteration.
(iii) Client: I beat myself
Although AI cannot completely replace programmers to write complex game clients, it can help programmers generate a part of simple code or extend it based on existing code. The use of AI to assist in writing game client code has had good results, such as the GitHub Copilot tool, and the usage and steps are as follows:
GitHub Copilot is a code completion tool developed based on AI technology that can be used to assist in writing game clients. Trained on OpenAI's large codebase, GitHub Copilot understands a wide range of programming languages and frameworks such as Python, JavaScript, Unity, and Unreal Engine.
The steps to use are:
- Prepare the environment and install GitHub Copilot, which will be integrated into the programming environment (such as Visual Studio Code).
- Writing game logic: When you encounter something you need help writing for, you can use GitHub Copilot. Simply enter a few relevant keywords or comments, and GitHub Copilot will generate code suggestions based on their understanding.
- Optimized code: AI-generated code may not be exactly what we need, so it needs to be reviewed, corrected, and adjusted to the desired solution. Without 100% perfect code, review and validation is still an important process.
- Test & Iterate: Run the client to test appropriately. If something goes wrong, fix the error and optimize the code. If there is a problem with this part that you just wrote, you can try submitting it to ChatGPT for help checking.
At present, there are many tools similar to GitHub Copilot to assist in writing game client code, although it is currently impossible to complete the writing task completely alone, but with the development of technology, AI assistance will play an increasingly important role in the field of game development.
(4) Gameplay updates brought about by the integration of AI and games
- RPG/Big World Gameplay:
RPG's world perception will be shaped more clearly and specifically through AI dialogue, such as the cultivation of relationships with NPCs, NPC dialogue's understanding of the world, and the diversity of NPC interaction behaviors, and the RPG world will be more vivid, which may extend deeper trading, socialization, confrontation and other gameplay with NPCs. For example, Hogwarts, Fantasy Westward Journey.
- Party Social Gameplay:
Goose and duck killing social party, AI dialogue makes AI players possible, because the traditional number of real people has a high threshold, and the number of players makes the experience may be reduced. The emergence of AI players can solve such problems and may extend to more complex gameplay variants.
- Virtual human theme:
For example, Tom Cat + ChatGPT, Genshin Character + ChatGPT, character and animal development gameplay due to AI dialogue gives new interaction and emotional expression, giving birth to "portable" independent gameplay and gameplay updates.
- Puzzle Class:
Puzzle solving based on dialogue texts will become more open in terms of participation, puzzle solving, and rewards, and may produce an open ending with thousands of faces. For example, a game of the Disco Bliss genre.
- Script running group category:
Dungeons and Dragons, which currently has a preliminary access to the GPT version, plays the role of DM, and it is foreseeable that the evolution of AI for the generation of art resources will make the original text running group become rich in pictures and rich in sound. Player AI changes voice, immersive play, and the game experience will reach a new level.
- Cultivation class:
The cultivation routes of the cultivation category such as horse racing girls may be generated by AI, which provides different interactive gameplay, and the plot text is generated based on the player's basic data and continuous input, and the fixed and dynamic content are combined to truly cultivate the player's own "idol".
Third, the development prospect of AI and games
(1) Regulatory related
1. Copyright of AIGC
- United States: Content generated by AIGC is not copyrighted, and if the similarity is too low, it is not considered to be the source of infringing material
On February 21, 2023, the copyright registration of AIGC's comic "Zarya of the Dawn" was granted, but only for the non-AIGC part, that is, the part of selection and coordination arrangement, and the copyright of the work was directly reviewed and approved, and then the US Copyright Office rejected it after learning that it was AIGC content, and finally after further consideration, the scope of copyright protection was divided.
In general, the current policy is still in a period of change, and there are two aspects that need our attention: first, copyright can only protect the product of human creativity, which can be understood as belonging to "anyone" in the case of unprotected conditions, and second, the risk of infringement still exists, the training set generated by AI diffusion usually uses more than 10 million images, and at present, there is basically no infringement problem from the similarity, but it cannot be ignored that large-scale use will change the definition of future infringement methods.
- EU: Four-step approach to determining AIGC content as a "work"
The European Commission's 2020 report proposed a "four-step test" to determine whether AIGCs qualify as "works": 1) literary, artistic, and scientific fields; 2) human intellectual activity; 3) originality; 4) Expression.
At present, AIGC works basically do not meet the second and third points, and cannot be protected by copyright permission and corresponding authorization.
2. Data security and cultural security of .AI large models
- Italy
On March 31, Italy's national privacy regulator formally banned ChatGPT and accused OpenAI of "illegally collecting personal data."
Notably, this "temporary injunction" will remain in effect until OpenAI is able to respect the EU's landmark privacy law, the General Data Protection Regulation (GDPR).
- China
On April 11, the Cyberspace Administration of China issued the Measures for the Administration of Generative Artificial Intelligence Services (Draft for Comments), which mainly states that AI content providers should bear the responsibilities and personal information protection obligations of producers of the content generated by the product. Before using generative AI products to provide services to the public, they must report a security assessment to the national internet information department in accordance with regulations, and be responsible for content review.
(2) AI ecology and game development prospects
The current scene of the AI ecosystem may only be on the eve of the single-point outbreak stage - AGI. The author believes that the development of AI ecology will go through at least 4 stages:
I. Single point burst;
II. vertical large-scale applications, as well as multimodal applications;
III. Function integration and coupling, emerging new functions and experiences;
IV. Industry reshuffle, head monopoly + subdivision leader.
At present, we are in the stage of transition from I to II, and we can see that ChatGPT and other large language models, MJ, and SD have sparked in various vertical fields after the full single-point outbreak ignited the user group, such as the Microsoft office family bucket combined with GPT, and this vertical category in the office field will continue to be subdivided until it is fully adapted and specific pain points are opened.
GPT4 has initially demonstrated the image processing function in multi-modality, and in the subsequent 5/6/7 version, it will fully cover the processing power of audio and video, and the processing power of mixed "information" and integrated input in the form of multiple carriers has reached a new level, it is obvious that games, the field of pan-cultural entertainment and integrated cutting-edge technology, will be the best leading industry, and the largest subversion of the industry.
In the AI ecosystem + game industry, the three ends of PME will gradually usher in the large-scale application of vertical segmentation, and in each part of P, M, and E, such as the production side, the planning, client, and art departments will soon be covered, from production to update the player experience.
Overview of HuggingGPT. ）
The framework here to connect various AI models in the machine learning community (such as Hugging Face) to solve AI tasks with large-language models LLM (such as ChatGPT) has begun to bear fruit, such as Microsoft Research Asia's HuggingGPT in the figure above, which can cover complex AI tasks in many different modes and domains. As well as many recent integration tools, such as BabyAGI, AutoGPT, etc., AGI is beginning to appear.
And scattered AI technology in a hundred flowers, a hundred schools of thought contention stage, is still the first stage of technology extension, transformation, after a full outbreak will gradually converge, new technology, new functions in the way that has never been designed by the system after combination, will produce emergent innovation, this process such as CV, NLP and other integration brought by the emergence of innovation, such breakthroughs will bring new features and new experience of the game, not only to the convenience of the producer, but also to really affect every mass player.
(3) Individuals in the AI era
At the dawn of a new era, personal attitude choices are particularly important. The author suggests that according to personal circumstances, there are two relatively integrated attitudes:
1. Embrace and learn
- Keep an open mind: Embracing the AI era first requires maintaining an open and optimistic mindset and not being afraid of the challenges posed by new technologies. AI is a good tool, "gentlemen are good at fakes", we have been keeping pace with the times from taming animals, making carriages, bicycles, to fuel vehicles, new energy vehicles, as the saying goes: "The era of cars is coming, what we should do is to get a driver's license and learn to drive it." ”
- Update knowledge structure: In the era of AI, individuals need to constantly update their knowledge structure and learn interdisciplinary knowledge, especially in AI-related fields (such as computer science, data analysis, statistics, etc.). At present, knowledge acquisition 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 own situation.
- Follow the trend of the times, think deeply, and see through the essence: on the one hand, continue to pay attention to the dynamics and trends in the field of AI, so that you can always keep pace with the times, on the other hand, be vigilant and catch the wind, think deeply about the nature of demand, the essence of the problem, and the essence of the value chain, and look at the essence behind the trend with first principles. In addition to science and engineering thinking and skills, non-logical insight thinking, empathy, and humanistic qualities will become more precious, here we better not use disciplines to define and cut, keep "people-oriented", think more from the perspective of care and love, maybe there will be different discoveries and gains.
2. Discover application scenarios, solve specific problems, and realize self-and social value
- Immerse and deeply understand the industry in which you are located: in-depth study and understanding of the pain points and needs of the industry, discover the specific problems that AI technology can solve, and individuals can give full play to their advantages according to their own interests and specialties to achieve the application and breakthrough of AI technology in their own areas of concern, for example, practitioners who are deeply engaged in RPG categories can consider whether they can train pendant GPT models to bring value to enterprises and players.
- Drive innovation and development: By exploring new application scenarios of AI technology, such as ControlNet's development for Lvmin Zhang, a Chinese Ph.D. student majoring in computer science at Stanford University, can we better apply it in the field of games, promote the innovation and development of the entire industry, and also achieve personal value enhancement.
- Popularize AI, also a teacher and a friend: As a user in the field of AI, after being familiar with and mastering, you can choose to help more people understand and use AI technology to solve real-life problems, which is also very helpful for our own improvement.
IV. Summary
This paper discusses the application and future development prospects of AI ecology and game industry in a shallow way, provides a panoramic analysis of the three-terminal application of PME in the game industry, shares the specific technical application of AI on the P-end at this stage, and finally discusses the supervision and risk items of AI, looks forward to the development of AI ecology + game industry, and gives a little suggestion around the response of individuals in the wave of AI, for reference only.
I hope to help the game industry practitioners understand the overall situation, maybe this is a seed to help you grow a small seedling of AI + game research in the future, maybe this is the eagle eye of Assassin's Creed soaring, help you get a bird's eye view of a wide scene from a global perspective; Perhaps this is an "investment report" that will help AI ecology + games attract a "tree planter", irrigate a pouring of water, nourish the development of the industry, and bring value to society and players.
This content is the author's independent opinion and does not represent the position of Tiger Sniff. It may not be reproduced without permission, please contact [email protected] for authorization
People who are changing and want to change the world are in the Tiger Sniff App