The Future of Gaming and AI: Penetrating the shimmering light of thought inspiration to the other side of the emergence activation
China Game Developers Conference (CGDC) is one of the most forward-looking, international and professional game industry R&D technology conferences in China, and one of the industry frontier summits that can best represent the direction of world game development.
With the key meaning of sharing the latest achievements in game development technology and discussing new trends in development, the conference attracts tens of thousands of eyes from the game and cultural industries and gathers thousands of the most pioneering game industry practitioners from home and abroad every year. Many development technologies that have profoundly influenced the shape of the industry have taken this as a starting point and will gradually enter every corner of game design in the coming years.
On July 30, rct AI CEO Yuheng Chen gave a speech titled “The Future of Gaming and AI” at CGDC 2021. He spent 40 minutes to introduce more possibilities of the future by combining AI and gaming, and vividly depicted rct AI’s views on technology implementation, application design, interactive art and virtual worlds, which received a lot of praise from the developers on site.
This article mainly describes the content of Yuheng Chen’s live speech, hoping to bring inspiration and ideas to more fellow solo travelers under the star field.
Hi everyone, I’m Yuheng Chen, CEO of rct AI.
I’m excited to have the opportunity to share our understanding of games and AI at CGDC, and to share what we’ve been working on. I will try to bring you a new perspective on the value of AI in games and how game play, development and design will evolve in the future after AI is further integrated into games.
From the first day of video games, games and AI have been inseparable, and the first game in human history, Pong, launched by Atari in 1972, was centered on the interaction between humans and virtual computer people in the form of simulated ping pong.
Later, with the further development of information technology, we saw the goblin enemies in Zelda and the “Kanojirushi” in Love Plus; at the same time, a few years ago, robots with massive computing power showed amazing results, beating the best human players in StarCraft 2 and DOTA 2, people finally admitted that AI can surpass human in terms of learning and execution efficiency in some fields.
But that’s not relevant to us, today I want to show you: another way AI can be used in games.
The game pictured below is the 1996 CyberLife simulation game Creatures, which was the beginning of the ultimate human thinking about AI in the digital world and the inspiration for the Bionic Man concept in Detroit: Become Human. Detroit: Become Human is a tribute to Creatures, and the bionic company in the game is called CyberLife.
Creatures is a nurturing game, it has a lot of today’s still very advanced gameplay: players can adopt a small creature in the game, and then teach it how to talk, what toys to play with, what to do, like teaching a child, different players cultivated small creatures will therefore show different behavior, different personality.
In 1996, such a behaviorist-based reinforcement learning framework revolutionized the experience for players at the time; although computers were far less common then than they are today, the game community could have at most 5 million small creatures growing at the same time. It was an extremely impressive achievement and a far-reaching human social experiment.
Today, the creator of Creatures, Steve Grand, a British computer scientist who is now over 100 years old, is still working on the answer to the proposition proposed by Creatures. It is a proposition that we have seen in many science fiction works: virtual artificial life.
Virtual artificial life, is life, not machines. Here is a question we need to think about together first: What do you think is the biggest difference between life and machines?
I think it lies in emergence. In the perspective of emergence, our earth, our civilization, our life, everything is the product of emergence; if there is no emergence, everything only runs according to the predetermined fixed rules, then there is no way to talk about innovation, and the world will not have such a rich diversity as this time.
At the same time, I also think game designers are the ones who can understand emergence best.
The meaning of emergence is that when a system is complex enough, that is, when there are enough parameter states that can be set, enough interactive elements that are rich enough, and enough granularity that is abstract enough, the various parts or modules in the system can create new but recognizable and stable effects among themselves at a higher level.
Both the designer and the experiencer need not focus on the high-level final experience, but rather on the low-level basic elements and basic rules that make up the entire game world, and the final output is often eye-opening and then enlightening.
At some point, when the results come out, people may think: Well, if I write fixed rules at a higher level, the same results can occur. But in reality, the internal structure of the system is completely different, and the magnitude of possibilities that can be generated are also different.
When a designer thinks in this way, he can never design a truly emergent system; on the contrary, it is only when he really breaks the fixed way of thinking at the beginning and focuses on the bottom of the design that he can create a true emergence.
So, for emergent systems, there is a process of quantitative to qualitative change in the middle of the development process.
In games, there are many emergent scenarios that can reflect the classics, such as The Sims, GTA, Minecraft, where players can get different feedback when communicating with the game world in different interactive ways, thus reaping a different gaming experience each time; then there are games like Rimworld, The Scroll Of Taiwu or Dwarf Fortress that are a bit more hardcore, which are made a very complex system based on large-scale scripts and behavior trees, predefined modular rules, and character environment attributes, thus creating a very emergent experience for players.
Façade, born in 2003, is the originator of a very dynamic game experience that allows players to interact with the game content through open-ended interactions, dubbed by gamers as a “ couple quarrel simulator “. It uses NLP technology based on very simple rule matching, where the game designer collaborates with the AI, or rather works better with the AI based on a deeper understanding, to achieve a series of amazing experiences.
The difference in the technology used, in turn, shows a difference in the degree of intelligence, which in turn leads to a difference in the degree of emergence and the complexity of the emergent system that can be carried.
When I was a kid I used to play a game called Neighbours from Hell . In that game, players can trigger or set up some different mechanisms in different rooms to mess with the neighbor. At that time, I was very keen to explore which items could be interacted with, trying to put different items together to expect extra effects, and feeling curious about everything.
But now looking back at the game, I can say that today I may be tired of the game in half an hour of playing it, because the degree of freedom available for interaction in the game is actually very low. Because I have played The Sims, Minecraft, GTA, and other games that are a hundred times more open than it in my years of growing up, the threshold for the complexity of the emergent system needs has long been drawn up.
In contrast, our next generation, whose adaptability to digital content and experiences has been increasing, will also demand a higher level of richness than the current one when they enter the virtual world to explore. AI shaped emergence can gradually meet this demand.
At this year’s ChinaJoy, we rct AI also brought a self-developed mini-game “BABY NOT BAD” to the exhibition, which is our attempt to explore the AI emergent gameplay, inspired by the earlier Neighbours from Hell.
The story takes place in a fictional world where the player is given the task of babysitting a couple’s child while they are away from home. Children are always curious about everything in the world, but there are a lot of hazardous existence in the household for the child. However, “he’s just a kid”, and in the face of his exploration and curiosity, players have to make sure that their blood pressure is stable throughout the game, and avoid having an emotional breakdown.
Every object in the game can interact with each other, and they go behind by a unified way of definition, such as fire meeting combustible objects, water meeting electricity, etc. These ways of definition dictate that we can create a great deal of game content in a much more efficient way, without having to consider the possibility of the upper level players actually playing with it, but just from creating the basic rules and defining the objects themselves.
Currently the game is only in the initial demo stage, but already has a great playability. As the game is perfected in the future, it may translate into unexpected commercial performance.
At the same time, the main character of our story, the baby, is also AI-driven. What he goes to open, what he uses, and what he throws away at each moment, which make up every event the player experiences, are based on AI models to make decisions. Even at the time of creation, our AI has the ability to tell the creator: how much this newly defined rule or object of yours has improved the degree of emergence of the whole system. Thus, the planners do not need to waste their efforts to fill the game with too many waste cases that do not show good results.
The essence of applying AI here is that after populating the game with content in some low-code way, we use AI to intelligently and organically combine that content to make up a different experience for the player each time. AI plays the role of a supporting human, based on a complex system with its own emergent features, to generate a thousand different possibilities of interactive experiences.
Most people trying to understand the future of AI in games are inevitably influenced by historical and existing general public understanding of games. That’s not a bad thing; it’s the nature of business, and as a profit-seeking company, it’s natural to need to think about existing markets and audiences.
However, when we look at the long term and into the future, it is clear to all of us that the future of games, whether it be graphics, hardware, gameplay or the production process behind it, will definitely be iterative and will definitely be changed.
The games I just mentioned were the pioneers of their time. It was their appearance that changed the user’s perception, allowing players to touch the novel experience they never had before, and thus led the development of the whole industry forward.
So, from this perspective, the future of games is emergence. According to the nature of emergence, when all the underlying elements and mechanisms in your game are defined in a unified framework, such a scenario is also the best scenario for us to promote AI implementation. We are not giving players a placebo, a substitute, but handing them a new space to explore and the possibility of more tactile extension of content.
Why is the future of AI emergent? Contemporary AI, whether it is neural network-based deep learning or reinforcement learning, has an inescapable Achilles’ heel behind it, and that is the uninterpretability of the training process or model convergence process. This uninterpretability has brought an industrial revolution beyond expectations to academia and industry, allowing AI to produce efficiency or accuracy that far exceeds that of humans in many task completion scenarios.
However, when un-interpretability meets content creation, a natural conflict arises: that is, the irreconcilable conflict between the control needed by human creators and the un-interpretability of algorithms.
Yet there are two sides to everything. What can we get if we take advantage of this black box? Emergence.
The model does not have a logical inference process that is inherent to us humans or can be completely exhausted by us, it has its own set of reasoning based on mathematical and probabilistic calculations, which provides us with just enough basis for emergent properties.
In the game scenario, when we can achieve a high degree of data structure, when the system complexity increases exponentially compared to previous games, it provides a very suitable ground for the emergence of algorithms. Since humans cannot exhaust all possibilities in the system, let the algorithm do so, and let the algorithm pick what it thinks is the most reasonable path. People only need to understand the possibilities generated by the algorithm a posteriori, and this is emergence.
In addition to the black box, many creators have a question about AI: How do you make sure the feedback is interesting?
The Witcher 3, for example, in the eyes of most players, is a very high quality work with unpredictable gameplay and a dramatic plot, which shapes a fascinating world, and the premise of this world can not be achieved without the strength of the team behind the plot content.
Assuming that characterized AI avatars make up the majority of The Witcher 3’s NPCs and that the AI writes the story, would it be possible to achieve the level of The Witcher 3 now? Our answer is, maybe, maybe not, because the interactions that each player will have before entering the game are uncertain, and he may experience some raw grass in the course of the game, but at the same time, he cannot exclude the possibility of experiencing a masterpiece plot.
In fact, in our experiments with our own Chaos Box algorithm-driven plot continuation or avatar dialogue generation, we have seen AI generate too many surprising, up-and-down, and remarkable stories.
Today’s AI is more than capable of generating a convincing text in terms of continuation. Even if you do not use it as the core of the gameplay, but only as a ghostwriting tool to generate plot text with one click, AI can improve the efficiency of the developer greatly, and can be used as mature content in the game after subsequent partial optimization and modification by the developer.
In addition, the AI-driven game may not be as beautifully rich and epic as The Witcher 3. But with more freedom and plot expansion, it can also make a different kind of marvelous journey from The Witcher 3.
For players, if The Witcher 3 is a “life of Jarot” experience in an alien world, then playing an AI-driven game is probably closer to Re:Zero. For the world, the player is a destabilizing factor that is added to it, and the world can continue to function even without the presence of the player.
In such a premise, your otherworldly journey may be similar to the existing plot of Re:Zero, accidentally helping a young girl and then embarking on a struggle to become involved in the royal family to seize power.
In this sense, I think the best setting for Re:Zero is perhaps not the return of death, but if the male lead did not like white hair so much or did not meddle in the beginning of the story, the whole story will be completely turned upside down.
This “openness” is completely different from the current “open world” in the “definition of space” where you can just run around on the map. That’s why so many people are looking forward to the development of Re:Zero, which is the real “freedom” in our definition.
Today, I’m not going to talk much about the main business we are doing at rct AI, but rather to share with you our thoughts and understanding about emergence and AI applications accumulated in the process of working with customers and developers in the game industry, including in the process of making our own games that we like. I would like to take this opportunity to share with you what we see and believe about the future of game experience from the perspective of gamers and independent game developers.
In the end, I would like to briefly talk about our current main business and core technologies. In the past year, although we are still a very young team, and although we are still at a very small scale and very early stage of development, we have now expanded more than 20 client projects and served more than 200 million gamers worldwide with our very strong execution and technical capabilities.
We provide our collaborating customers with a variety of solutions covering the full life cycle of games in their respective scenarios. The core behind them is all based on our self-developed multimodal interaction capabilities incorporating various technologies such as semantic understanding, text generation, deep reinforcement learning and recommendation algorithms with personification and emotional interaction.
In the second half of this year, we will soon launch our standardized Meta Being creation platform for small and medium-sized game teams and even individual developers, so that more developers can enjoy the ability and advantage of obtaining rich content and emergent nature at low cost.
