Text/Lin Zhi

Recently, Grape Jun chatted with Zhou Dong, the head of Optillusion Studio, about his new game, Pick Me Pick Me.

After graduating in 2019, he set up a studio in New York and has been doing his own research and development. The team has been expanded twice and now has 16 people and an office in Shenzhen, China.

Zhou Dong's first work was "Dream in the Cage", which won the award that year, and friends who pay attention to independent games should not be unfamiliar with this game.

Swipe left and right to view

From a puzzle game to an AI party game, the transition may seem huge, but Zhou Dong said that they will only choose to set up a project if they can't find a product with a high degree of overlap on the market.

Zhou Dong told me that after coming into contact with AI, he felt that AI could inject fresh blood into the dull game industry, which is why he made "Pick Me Pick Me".

Pick Me Pick Me is a two-player online battle arena.

Upon entering the game, the two players will start a "three-person date" with the AI character.

The game is turn-based, with two players taking turns having a conversation with the AI. Players win the game by speaking smartly, funny, or catering to the AI's preferences.

The game's main action page is a chat box with the player's avatar on the left

If the player is very good at the whole conversation, the whole conversation will be more fun.

For example, my friend tried to play "Hello special ......" with Hamlet.

There are two types of actions for players in the game, one is to speak on their own, and the other is to use card skills.

While waiting for the opponent's input, the player can use the cards. Cards are broadly divided into two categories:

One is to debuff the opponent. For example, reversing the meaning of a sentence to confuse the logic of the other party's speech; Or a random embarrassing incident where an opponent takes the wrong cup of water and drinks water with dentures......

SOME DEBUFF CARDS IN THE GAME

The other type is to hang a buff on yourself. For example, "mind reading" can see through what the AI is thinking now, so that his speech can be more favorable; Or "Escort!" Isolate yourself from negative influences and be as steady as an old dog.

In addition to randomly drawn cards, AI characters also make a big difference in the experience of each game, as they will have their own preferences and unique personalities. And because the game uses advanced large language models, you can talk to the AI about new topics for a long time.

Ada, for example, comes across as a negative personality at first, getting bored no matter what the player says.

But she actually likes Cyberpunk 2077 and Two Cities and doesn't like the dull school life. If you tell her that we skip class together and go to Night City, she will be very excited, and when you ask about "Two Cities", she will say that she admires Jinx's personality and so on.

The developers also posted some character posters on Discord, and you can feel the difference between the characters just by the character profiles: a smart three-no girl who loves to play chess (Dongxiang one-two-three?). ), the cool brother who likes dark fantasies but loves to drink milk tea with more pearls, and Hamlet, who inspires Shakespeare......

Character likes and dislikes poster, swipe left and right to view

A game lasts more than ten minutes, and the pace is very fast. Although the AI will provide a lightweight reward, the concept of winning or losing is not strong, and being favored by the AI only adds a little sense of achievement, and the core is to let the player enjoy the process of chatting.

At the end of each round, the AI character who just chatted with will judge who speaks better

You can feel that this is indeed a relatively fresh product. For Zhou Dong, this product is their team's first attempt in the field of AI+ games.

During the conversation, he talked about how production in the AI era should be "cause and effect inversion", where developers have to find the optimal solution through mass generation and screening, rather than presupposing a clear goal at the beginning - which is the opposite of the traditional linear process from concept to implementation.

He encouraged each team member to use AI to find shortcuts to their work, and hoped that the team could follow the AI way of thinking to design content and solve problems, and finally form a team or company in the AI era.

How should we understand the future of AI + gaming? Will AI change the possibilities of game storytelling? Will the role of the developer be completely reshaped in the age of AI?

Perhaps his idea can give some answers to friends who want to get involved in AI+ games.

01

I want players to regain the "mystery" of the game

Grape-kun: Why did you want to make such a game?

Zhou Dong: I've observed that a lot of people buy games and don't play them, and I think it's because it's hard for us to get really excited about new games. After all, there are only so many categories of games, and it is very difficult to create new categories.

As developers who work with games on a daily basis, we experience similar burnouts. A lot of times you think maybe casual games are the best place to go? There is this myth, for example, many people may want to play match 3 at the end of the game. (laughs).

So we wanted to make a product that would allow players to rediscover the "mystery" of playing games.

Grape King: Mystery?

Zhou Dong: When I was a child, I played those games that didn't have a Chinese version, and I couldn't read Japanese, so I had to keep playing by guessing.

At that time, people didn't immediately look for the guide, which made the game form a "magic circle". Players explore in this circle and will not easily jump out. This uncertainty instead brings with it a strong sense of mystery.

But nowadays, many games are designed to teach you all the steps by step, just like working part-time. We want to get back to an experience where the game itself is engaging, rather than an engagement.

For example, when the open-world game first came out, people didn't know its routine, and it was very attractive. Later, when everyone became familiar with the modular design of the open world, the novelty disappeared. Even playing The Legend of Zelda, you can anticipate its mission structure.

Traditional games, especially mobile games, have changed from the initial freshness to the later routines, and everyone is "fighting wits and courage" with operators, and the sense of mystery is becoming less and less.

Grape Jun: But the gameplay of "Pick Me Pick Me" is not complicated, what if players repeat the experience in a short period of time and lose the freshness?

Dong Zhou: We'll be thinking about adding more cards to add more drama to the mix. We're also exploring other modes, such as multiplayer or simultaneous input.

Moreover, the AI in the game plays the role of the host, which in itself has a lot of room for imagination.

Traditional social games can basically be done, such as integrating the gameplay of truth or dare, and letting the AI as the host drive the whole atmosphere.

Grape Jun: It sounds like you're pursuing cutting-edge technology while also retaining a kind of back-to-basics fun? Do you think you have this tendency?

Zhou Dong: A little bit. This feeling will be more noticeable especially after playing a lot of games.

At first, I thought the more fun I had, but then I realized that it was more fun to only play the games that I really enjoyed.

Personally, I like to play galgame, and I like to play some old games. The creation of old games is not influenced by factors such as the market, and can bring some unique experiences that you can't find in modern games.

For example, some old galgames have the setting of jumping time and space, and your save bit itself is a resource. Saving isn't just about saving progress, it's part of the game's mechanics.

I feel like games need this kind of experience, and I believe AI can do it, and that's why we've been exploring this area.

02

A "fancy talking" game

Grape Gentleman: What is unique about Pick Me Pick Me?

Zhou Dong: At this stage, most of the game products that AI can land are "persuasive".

The essence of this type of game is to convince the AI to achieve a goal, such as convincing the AI to open the door and let you in and out, convincing the AI to provide clues, or to make the AI not angry, etc.

As a vampire player, you'll need to convince the AI to open the door. As a successful case of AI game commercialization, Suck Up! Let Zhou Dong see the potential commercial value and user base of AI games.

We've added social interaction to this foundation to make communication between players a core fun. The whole mechanic revolves around chatting, which is essentially a party game of "fancy talking".

Grape Jun: What's the difference between this game and text-only chats like Character.ai?

Dong Zhou: The biggest difference is the interaction between the players. Pure AI chat is limited by technology and cost, and whether the experience is good or not depends on whether the AI is smart or not.

Pick Me Pick Me, on the other hand, pits players against each other, with AI acting as a moderate. The fun of the game depends on whether the players you interact with have fun or not, rather than relying solely on AI.

Grape King: How were the characters of the game designed and confirmed?

Dong Zhou: We initially used AI to generate concept maps, and then we looked at the pictures to design characters.

We prefer one-sided, face-painted characters, such as mean, gentle, middle-of-the-road, domineering president, and so on. The main thing is to let each character bring different topics and chat styles to create a more joyful and chaotic party atmosphere. In the future, the ability for players to customize their characters will also be added.

Grape King: How is the interaction between AI and players optimized?

Zhou Dong: This is a lot of experience, and we mainly adjust it through a lot of testing.

We prepared some test sets, such as asking the AI "How tall is the Eiffel Tower?" "It's obviously boring to answer numbers.

The direction of our tweaks is to make the AI respond in a more humorous or interactive way, such as asking the player, "Why are you asking this?" We're not talking about this topic when we're eating. ”

Grape Jun: Who do you think Pick Me Pick Me is for?

Dong Zhou: The target audience of Pick Me Pick Me is "author players", that is, people who like to express themselves and are willing to output.

On the other hand, "reader-type gamers" are more likely to receive information passively, and are traditional story-reading players who may not know how to interact with AI.

We've also done market research, and the number of reader players is about four times that of author players, but the author player base is large enough that we prioritize them.

This product is a product that we believe can be successfully implemented under the existing technology (controllable cost of API calls, fast enough AI response, and smooth gaming experience).

In the future, we'll try to develop games for readers. However, instead of waiting for technological progress, it is necessary to understand the limits of AI technology and then design according to those limits when developing.

03

Design with the AI way of thinking 

Grape: Sounds like you're at the cutting edge?

Dong Zhou: Our core goal is to push the boundaries of games, and if we can provide a new design idea for social AI games, it would be the best. But this thing is not entirely up to us.

It's not just AI games, it's the AI industry as a whole that's interesting. In addition to improving R&D efficiency, AI can also give rise to new game paradigms, such as "persuasion" games are an important direction we are currently seeing.

I believe that in the future, AI gameplay will gradually infiltrate the game, perhaps as the core gameplay, or it may be seasoning. Maybe there will be a revolution in the end, but it will certainly not be in the form we can imagine now.

But now that many games have added AI, they haven't made them more fun, they've become cumbersome.

Grape: There's a book that says "playing a game is about voluntarily overcoming unnecessary obstacles," but AI brings more freedom.

Zhou Dong: Yes, the game has rules and restrictions, and its essence is to dance in shackles. In a sense, the AI itself is anti-game, as it will tear down those shackles and give you complete freedom. But it's not necessarily fun, and that's what game design needs to solve.

Grape: Since there is an inherent conflict between the two concepts of AI and games, how should games be made?

Zhou Dong: There are two directions.

The first is to adapt to the characteristics of AI and make gamification packaging for it, so that the freedom of AI can be packaged into interesting gameplay.

The second approach may not be so intuitive, it is not the complete freedom of using AI, but its "ability to make connections".

It's an emotional ability, and I think AI doesn't judge things by logic, but by feeling.

A lot of developers try to make AI a substitute for logic, but I don't think it's feasible. We should make it a substitute for the emotional part of the game, and that's a possible way forward.

Grape Jun: Just like many people use chatGPT, what they are pursuing is actually the emotional value provided by AI?

Zhou Dong: It's kind of similar. Emotional value is indeed a manifestation of the AI's emotional abilities, and we use that in Pick Me Pick Me.

Grape Jun: Now some games in the industry will introduce AI as intelligent NPCs.

Zhou Dong: I think this is a typical counterexample.

For players, the purpose of interacting with NPCs is simple, either to get quests or rewards. But some games turn a simple 2-to-1 process into a complex free exchange, which is not fun.

Players may want to complete tasks quickly, rather than chatting around and around. Such a cumbersome process dragged down the core game mechanics.

In this case, adding AI free dialogue reduces the fun, and the ends and means are completely reversed.

Grape Jun: Another relatively widespread use is AI-assisted UGC.

Zhou Dong: This is indeed a good direction, but it is essentially a tool application, and it brings a single-point efficiency improvement. However, the efficiency improvement of tools is limited, and the final effect tends to converge.

Grape King: What does the application of multi-point look like?

Zhou Dong: Multi-point application refers to thinking from the perspective of the entire product design.

The development model in the AI era can be said to be "cause and effect inversion" - first there is an effect, and then the cause is derived.

For example, in the early stage of the project, we need to consider the style that AI can mass-produce at this stage, and then develop and design products around this style, rather than waiting until the project starts and then use AI tools to assist.

Not only does this mindset increase efficiency, but it may also lead us to discover new opportunities. Instead of sticking to traditional linear thinking, it is better to follow the logic of AI to find a more effective path.

Grape Jun: "Causal inversion" is the overall development idea, how to do it exactly?

Zhou Dong: To put it simply, it's subtraction.

In principle, this is not an appropriate example, but it is easy to understand:

For example, why do you need a Prompt Engineer? We need to design prompts to filter out the massive amount of information generated by the large language model and leave useful content, which is a process of entropy reduction.

The process of communicating with a large language model is actually helping it to do subtraction and make it provide more accurate information.

The AI-driven process is essentially a mass generation and screening process to find the optimal solution, rather than presupposing a clear goal at the outset – the opposite of the traditional linear process from concept to implementation.

In the case of Pick Me Pick Me, for example, we also reduced entropy at the game design level.

We limited the AI to a specific setting, such as the environment of a party game, and let it play within that framework. This will avoid a lot of unnecessary situations.

For example, if the AI NPC in the game needs to go out with you, and the game itself does not have a scene or material to go out, it will appear OOC (out of character).

So, we're going to limit this part, like making NPCs reluctant to go out, or ending the game once they're out of the house.

This limitation allows for a smoother gaming experience rather than being dragged down by the freedom of the AI. It's like putting a "fig leaf" on the AI and turning its imperfections into a part of the game.

Grape: How did you adapt to this mindset when developing this game?

Zhou Dong: First of all, the current technology of AI-generated text is relatively mature, and the text can be structured, which will be more controllable when made into games. So we have to think about how to turn "chatting" into a game.

If players are just chatting with the AI, most of their fun comes from how smart the AI is. But if the AI is just a moderator or a middleman, the situation is very different. AI is not the protagonist of the game here, and whether the game is fun or not depends on whether your friends can chat or not, so we chose to do PVP.

Secondly, in order to make the characters in the game land quickly, our approach is to use AI to generate concept art, and then select and refine it into a 3D model, which is both efficient and saves art costs.

Finally, let's also consider the business model of the game. There is a cost to AI calls, and we have to make sure that buyouts cover API costs while providing value-added experiences for paying players, such as new characters and costumes.

The core of the whole process is how to package "chat" into a real game through gamification.

04

New opportunities

Grape: If you set a goal for your development in the next three years, where do you most hope that the team or product will make a breakthrough?

Dong Zhou: I've always encouraged teams to get into AI. I hope that in this era, teams can adapt to the AI way of thinking to design content, solve problems, and finally form a team or company in the AI era.

Although there are still many potential risks or impacts of AI, such as copyright issues and moral hazards, these are actually just caused by the immaturity of the current industry.

Grape Jun: The functions of the team members are very different, how do you encourage everyone to get into AI?

Zhou Dong: Different functions have different purposes for engaging in AI.

Curators can use AI to generate concept drawings to convey ideas to artists more clearly and improve communication efficiency. Even when considering game materials, he can use this to think about whether the materials are productive and whether the style can be unique.

Programmers are more likely to use AI to assist with code and problem solving. A lot of tedious little work, it will be more efficient to directly ask AI to generate preliminary results and then optimize.

Overall, we encourage teams to use AI thinking to find shortcuts when solving problems. Understand what AI can and can't do, and then get creative within that framework.

Grape: Why are you doing these things?

Dong Zhou: The current development process is no longer just about improving efficiency, but the whole way of thinking is changing. We need to plan with AI thinking from the very beginning to find unique products that can only be achieved in the AI era.

If we can adapt to this change, new opportunities will arise from design to operation.

However, how effective AI is ultimately depends on how people are creative and how they use it, not how powerful the AI itself is.

I've always believed that in the age of AI, it's still people's creativity that matters most.

This article is from Xinzhi self-media and does not represent the views and positions of Business Xinzhi.If there is any suspicion of infringement, please contact the administrator of the Business News Platform.Contact: system@shangyexinzhi.com