Why do you say that the ARC agent will break through the existing AI game experience

Teams like Parallel Colony and Virtuals are driving the development of autonomous AI agents, while ARC is carving out its niche market by focusing on human behavior cloning.

Written by: Teng Yan, Chain of Thought

Translation: Golden Finance xiaozou

In 2021, I was still an Axie Infinity player and ran a small scholarship guild. If you haven’t experienced that era, let me tell you - it was absolutely wild.

The game Axie Infinity has shown people that cryptocurrency and games can be combined. Essentially, it is a simple Pokémon-style strategy game where players need to build a team of 3 Axies (fierce warriors), each with unique abilities. You can lead your team to battle against other teams and earn SLP Token rewards by participating in the game and winning.

But what really excites non-gamers is the potential to make money through games. Axie’s rapid rise is due to two major mechanisms:

The first is Breeding Axies. Get two Axies, use SLP Token to breed them, and voilà—a new Axie with a unique combination of the original two Axies’ abilities is born. As a result, these rare and powerful Axies (referred to as OP Axies by players) have become popular commodities, and a busy breeding market has emerged.

The second mechanism is the scholarship program. Players from around the world are lending Axies to ‘scholars’. These players are usually from developing countries such as the Philippines or Argentina, where they cannot afford the upfront cost of over $1000 to purchase 3 Axie Non-fungible Tokens. Scholars play games every day to earn tokens and share profits with the scholarship guild, which typically takes a cut of 30-50%.

During its heyday, especially during the 2019 pandemic, Axie had a significant impact on the local economies of developing countries. In the Philippines (where about 40% of Axie Infinity users are located), many players earned income far above the minimum wage. Guilds made substantial profits.

This solves a key problem for game developers: player Liquidity. By incentivizing players to actively play games for a few hours each day, Axie ensures that each player will have an opponent waiting for them, making the player experience more attractive.

But this comes at a cost.

To solve the Liquidity problem for players, Axie gives out a large amount of Tokens as incentives for players to participate. The story starts here. Due to the unlimited supply of SLP, the Token inflates crazily, and the price has a big dump, causing the ecosystem to collapse. When the Token depreciates, players will leave. Axie almost overnight transformed from a profitable pet into a cautionary tale.

But what if there was a way to solve the Liquidity problem for players without the need for unsustainable Token economics?

This is exactly what ARC / AI Arena has been quietly working on for the past three years. Now, it is beginning to bear fruit.

1. Player Liquidity is vital

Player Liquidity is the lifeblood of multiplayer games and is also the key to long-term success.

Many Web3 and independent games face the problem of ‘cold start’—there are too few players to quickly match or form a thriving community. They lack the marketing budget or natural IP awareness that game giants have. This leads to long waiting times, inability to match, and higher churn rates, among other issues.

These games usually die out slowly and painfully.

Therefore, game developers must prioritize the Liquidity of players from the beginning. Games need various activities to maintain fun - chess requires two players, while large-scale battles require thousands of players. The skill matching mechanism further raises the threshold, requiring more players to maintain the fairness and attractiveness of the game.

For Web3 games, the risk is greater. According to Delphi Digital’s annual gaming report, the cost of user acquisition for Web3 games is 77% higher than that of traditional mobile games, making player retention crucial.

A strong player base ensures fair matchmaking, a vibrant in-game economy (i.e., more item trading), and more active social interactions, making the game more fun.

2, ARC- AI Game Pioneer

ARC, developed by ArenaX Labs, is leading the future of AI online gaming experience. In short, they use AI to solve the Liquidity problem that troubles new game players.

Most AI robots in the game are now too bad. Once you spend a few hours mastering the tricks, these robots will become very easy to defeat. They are designed to help new players, but they cannot bring too much challenge or stickiness to experienced players.

Imagine AI players with skills that rival top human players. Imagine being able to compete against them anytime, anywhere without waiting for a match. Imagine training your AI player to mimic your gaming style, own it, and earn rewards through its performance.

This is a win-win situation for both players and game companies.

Game companies use humanoid AI robots to make games popular, increase player liquidity, improve user experience, and increase retention rate - this is a key factor for newcomers to survive in the competitive market.

Players have gained a new way to participate in the game, building a stronger sense of belonging in the process of training AI and competing against it.

Let’s see how they do it.

3. Products and Architecture

The parent company ArenaX Labs is developing a series of products to address player liquidity issues.

• Existing product: AI Arena, an AI fighting game.
• New Product: ARC B2B, an AI-driven game SDK that can be easily integrated into any game.
• New Product: ARC Reinforcement Learning (RL)

(1) AI Arena: Game

AI Arena is a fighting game that reminds people of Nintendo’s Super Smash Bros. Various quirky cartoon characters battle in the arena.

But in the AI Arena, each character is controlled by AI - you are not playing as the warrior, but as their coach. Your task is to train your AI warriors using your strategy and expertise.

Training your warriors is like training a student for battle preparation. In training mode, you open data collection and create battle scenes to fine-tune their actions. For example, if your warrior is close to the opponent, you can teach them to block with your shield and then combo. How about long-range combat? Train them to launch ranged attacks.

You can control what kind of data to collect, ensuring that only the best actions are recorded for training. Through practice, you can fine-tune hyperparameters to gain more technical advantages, or simply use beginner-friendly default settings. Once the training is complete, your AI warrior can join the battle.

The beginning is always difficult - training an effective model takes time and experimentation. My first warrior fell off the platform several times, not because he was hit by an opponent. But after several iterations, I successfully created a well-performing model. It is deeply satisfying to see your training pay off.

AI Arena has introduced additional Depth through Non-fungible Token warriors. Each Non-fungible Token character has unique appearance characteristics and combat attributes, which will affect gameplay. This adds another layer of strategy.

Currently, AI Arena is running on the Arbitrum Mainnet, and only those who have AI Arena Non-fungible Tokens can access it, maintaining the exclusivity of the community while perfecting the gameplay. Players can join guilds, gather champion Non-fungible Tokens and NRNs for on-chain battle rankings, and receive rewards. This is done to attract loyal players and drive competition.

In the end, AI Arena is the showcase for ARC’s AI training technology. Although this is their entry point into the ecosystem, the real vision goes far beyond this game itself.

(2) ARC: Infrastructure

ARC is an AI infrastructure solution designed for game design.

The ArenaX team started from scratch, even developing their own game infrastructure, because existing solutions such as Unity and Unreal could not meet their vision.

Over the past three years, they have carefully designed a powerful technical stack that can handle data aggregation, model training, and model checking for imitation and reinforcement learning. This infrastructure is the cornerstone of AI Arena, but its potential is much greater.

As the team continues to perfect their technology, third-party studios begin to find ARC, hoping to obtain authorization or white labels for the platform. Recognizing this demand, they formalize ARC’s infrastructure into B2B products.

Today, ARC directly collaborates with game companies to provide AI game experiences. Its value proposition is:

• Perpetual player Liquidity service
• Integrate AI gameplay as a simple mechanism

Permanent player Liquidity as a service

ARC focuses on human behavior cloning - training specialized AI models to mimic human behavior. This is different from the primary use of AI in games today, which uses generative models to create game assets and uses LLM to drive dialogue.

With ARC SDK, developers can create human-like AI agents and expand them according to game requirements. The SDK simplifies heavy workloads. Game companies can introduce AI without dealing with complex machine learning.

After integration, deploying AI models only requires one line of code, and ARC is responsible for infrastructure, data processing, training, and backend deployment work.

ARC adopts a cooperative approach with gaming companies to help them:

• Capture the raw gameplay data and transform it into meaningful datasets for AI training.
• Determine key gameplay variables and decision points related to game mechanics.
• Map AI model outputs to in-game actions to ensure smooth functionality - for example, linking AI’s ‘right-click’ output to specific game controls.

How does AI work?

ARC uses four models for game interaction:

• Feedforward Neural Network: Suitable for continuous environments with numerical features such as speed or position.
• Table proxy: particularly ideal for games with limited discrete scenarios.
• Hierarchical and convolutional neural networks are under development.

There are two interactive spaces related to ARC’s AI models:

The state space defines the agent’s understanding of the game at any given moment. For a feedforward network, this is a combination of input features (such as player speed or position). For a tabular agent, this is the discrete scenarios the agent may encounter in the game.

Action Space Description Agent can do in the game, from discrete inputs (such as pressing buttons) to continuous control (such as manipulating joystick movement). This will be mapped to game input.

The state space provides input for the AI model of ARC, the AI model processes the input and generates output. Then these outputs are transformed into game actions through the action space.

ARC works closely with game developers to identify the most critical features and design the state space accordingly. They also test various model configurations and sizes to balance intelligence and speed, ensuring smooth and engaging gameplay.

According to the team, Web3 companies have a particularly high demand for their player Liquidity services. These companies pay for better player Liquidity, and ARC will use a large portion of this revenue for NRN Token buyback.

Bring AI gameplay to players: Trainer Platform

ARC SDK also allows web3 companies to access their game’s trainer platform, allowing players to train and submit agents.

Like AI Arena, players can set up simulations, get gameplay data, and train blank AI models. These models will evolve over time, incorporating new gameplay data while retaining previous knowledge, without the need to start from scratch with each update.

This opens up exciting possibilities: players can sell their custom-trained AI agents on the market, creating a new in-game economy layer. In AI Arena, skilled trainers can form guilds and offer training services to other companies.

For companies that fully integrate proxy functionality, the concept of Parallel Play becomes vivid. AI proxies are available 24/7 and can participate in multiple matches or game instances simultaneously. This solves the Liquidity problem for players and creates new opportunities for user stickiness and revenue.

But that’s not all…

(3) ARC RL: From One-to-One to Many-to-One

If AI Arena and ARC Trainer Platform feel like single-player mode (where you can train your own AI models), then ARC RL is more like multiplayer mode.

Imagine this: an entire game DAO gathering its gameplay data to train a shared AI model, which is collectively owned and beneficial to everyone. These ‘master agents’ represent the collective wisdom of all players, changing esports through the introduction of collective efforts and strategic cooperation-driven competition.

ARC RL uses reinforcement learning (i.e. “RL”) and crowdsourced human gameplay data to train these “super intelligent” agents.

The working principle of reinforcement learning is to reward the optimal behavior of the agent. It is particularly effective in games, because the reward function is explicit and objective, such as damage caused, coins obtained, or victory.

This is precedent:

DeepMind’s AlphaGo defeated professional human Go players in a Go match, improving its strategy with each iteration through millions of self-generated training matches.

I hadn’t realized this before, but OpenAI was already well known in the gaming community long before the creation of chatGPT.

OpenAI Five uses reinforcement learning to crush top human players in Dota 2, and defeated world champions in 2019. It has mastered advanced strategies such as team cooperation through accelerated simulation and massive computational resources.

OpenAI Five runs millions of games every day, equivalent to 250 years of simulated games per day, powered by 256 GPUs and 128,000 CPUs. By skipping the graphics rendering, it greatly accelerates the learning speed.

Initially, the AI exhibited unstable behavior, such as aimless wandering, but quickly improved. It mastered some basic strategies, such as creeping on the path and stealing resources, eventually evolving into more complex operations, such as ambushes.

The key concept of reinforcement learning is that AI agents learn how to succeed through experience, rather than being directly instructed on what to do.

ARC RL stands out by using offline reinforcement learning. The AI agent learns not from its own trial and error, but from the experiences of others. It’s like a student watching videos of others riding a bike, observing their successes and failures, and using that knowledge to avoid falling and improve faster.

This approach offers an additional benefit: collaborative training and joint ownership of the model. This not only makes powerful AI agents more widely available, but also aligns the incentives of players, guilds, and developers.

In the creation of ‘Super Intelligence’ game agents, there are two key roles:

• Sponsor: Similar to the leader of a guild, they stake a large amount of NRN Tokens to initiate and manage RL agents. Sponsors can be any entity, but are likely to be game guilds, DAOs, web3 communities, or even popular on-chain personalized agents like LUNA.
• Player: stake a small amount of NRN Token to contribute their gameplay data for training agents.

Sponsors coordinate and guide their player teams, ensuring high-quality training data so that their AI agents have a competitive edge in agent matches.

Rewards are distributed based on the performance of the super agent in the competition. 70% of the rewards belong to the players, 10% belong to the sponsors, and the remaining 20% belong to the NRN treasury. This structure ensures a consistent incentive mechanism for all participants.

Data Contribution

How do you make players willing to contribute their game play data? Not easy.

ARC makes providing game play data simple and beneficial. Players don’t need professional knowledge, just play the game. At the end of a session, they will be prompted to submit data to train a specific agent. The dashboard tracks their contributions and the agents they support.

ARC’s Attribution Algorithm ensures quality by evaluating contributions and rewarding high-quality, influential data.

Interestingly, even if you are a bad player (like me), your data is still useful. Poor gameplay can help agents learn what not to do, while skilled gameplay can teach the best strategies. Redundant data is filtered out to maintain quality.

In short, ARC RL is designed to be a low-friction mass market product centered around agents that collectively possess capabilities beyond human abilities.

4. Market Size

ARC’s technology platform is versatile, supporting various types of games such as shooting games, fighting games, social casinos, racing, trading card games, and RPGs. It is tailored for games that require player retention.

ARC’s products mainly target two markets:

ARC mainly follows independent developers and companies, rather than well-established big companies. Due to limited brand influence and distribution resources, these small companies often struggle to attract players in the early stages.

ARC’s AI agent solves this problem by creating a dynamic gaming environment from the start, ensuring dynamic gameplay even in the early stages of the game.

This may come as a surprise to many, but the indie game industry is indeed a major force in the gaming market:

• 99% of the games on Steam are indie games.
• In 2024, indie games on Steam accounted for 48% of total revenue.

Another target market is Web3 games. Most Web3 games are developed by emerging companies, which also face various unique challenges, such as Wallet login, encryption questioning, and high user acquisition costs. These games often have player Liquidity issues, and AI agents can fill the gap and maintain the attractiveness of the game.

Although Web3 games have been struggling recently due to a lack of engaging experiences, signs of recovery are emerging.

For example, Off the Grid, one of the earliest AAA-grade Web3 games, has recently achieved early mainstream success, with 9 million wallets conducting 100 million transactions in the first month. This has paved the way for the industry to achieve widespread success and created an opportunity for ARC to support this revival.

5, ARC team

The founding team behind ArenaX Labs has rich expertise in machine learning and investment management.

CEO and Chief Technology Officer Brandon Da Silva once led machine learning research at a Canadian investment company, focusing on reinforcement learning, Bayesian Depth learning, and model adaptability. He pioneered the development of a $1 billion quantitative trading strategy centered on risk parity and multi-asset portfolio management.

Chief Operating Officer Wei Xie manages a $7 billion Liquidity strategy investment portfolio at the same company and oversees its innovation investment projects, focusing on emerging areas such as AI, machine learning, and Web3 technology.

ArenaX Labs received a $5 million seed round financing in 2021, led by Paradigm and participated by Framework Ventures. The company received $6 million in financing in January 2024, led by SevenX Ventures, FunPlus / Xterio, and Moore Strategic Ventures.

6, NRN Token Economics - A Healthy Reform

ARC/AI Arena has a token - NRN. Let’s first take stock of the current situation.

Examining the supply side and the demand side will give us a clearer understanding of the trend.

(1) Supply Side

The total supply of NRN is 1 billion, of which approximately 409 million (40.9%) are in circulation.

At the time of writing, the Token price is $0.72, which means the Market Cap is $29 million, and the fully diluted valuation is $71 million.

NRN was released on June 24, 2024, with 40.9% of the circulating supply.

• Community Airdrop (8% of the total)
• Foundation Treasury (holding 10.9%, of which 2.9% has been unlocked, linear unlocking for 36 months)
• Community Ecosystem Rewards (30% share)

The majority of the circulating supply (30% out of 40.9%) is composed of tokens rewarded by the community ecosystem. The project manages these tokens and strategically allocates them to stake rewards, game rewards, ecosystem rise programs, and community-driven programs.

The unlocking schedule is reassuring, with no major events in the short term:

• The next unlock is the Foundation’s OTC sale (1.1%), starting in December 2024 and unlocking linearly over 12 months. This will only increase the monthly inflation rate by 0.09%, which is unlikely to cause significant concern.
• The allocation for investors and contributors (50% of total supply) will not begin unlocking until June 2025, and even then, it will be linearly unlocked over a period of 24 months.

Currently, the dumping pressure is expected to remain quite manageable, mainly stemming from ecosystem rewards. The key is to trust that the team has the ability to strategically deploy these funds to drive the rise of the protocol.

(2) Demand Side

NRN v1 - Player economy

Initially, NRN was designed as a strategic resource associated with the AI Arena game economy.

Players stake NRN on AI players, and if they win, they will receive rewards. If they lose, they will lose part of their stake. This creates a direct incentive dynamic, turning it into a competitive sport and providing economic incentives for skilled players.

Rewards are distributed using the ELO system to ensure skill-based balanced payments. Other sources of income include game item purchases, dress upgrades, and match entry fees.

The original Token model relies entirely on the success of the game and the continuous willingness of new players to purchase NRN and Non-fungible Tokens to participate in the game.

Let’s talk about why we’re so excited…

NRN v2-Player&Platform Economy

Improved v2 Tokenomics of NRN expands the utility of the Token from the AI Arena to the broader ARC platform, introducing powerful new demand-driven factors. This evolution transforms NRN from a specific game Token to a platform Token. In my opinion, this is a very positive transformation.

The three new demand drivers of NRN include:

Income from ARC integration. Game companies integrating ARC will generate revenue for the treasury through integration fees and ongoing royalties tied to game performance. Treasury funds can drive NRN repurchases, ecosystem development, and incentivize players on the Trainer platform.

Trainer market fee. NRN obtains value from the fees charged in the trainer market, and players can trade AI models and game play data in the trainer market.

Participate in the stake of ARC RL: both sponsors and players must stake NRN to join ARC RL. As more and more players enter ARC RL, the demand for NRN also increases accordingly.

Especially exciting is the revenue of the gaming company. This marks a shift from a pure B2C model to a mixed B2C and B2B model, creating a continuous inflow of external capital for the NRN economy. With ARC having a broader target market, this revenue stream will surpass what AI Arena itself can generate.

The cost of the trainer market has prospects, but it depends on whether the ecosystem can reach the critical mass - enough games, trainers, and players to sustain active trading. This is a long-term endeavor.

In the short term, ARC RL stake may be the most direct and reflexive demand-driven factor. The initial reward pool with sufficient funds and the excitement of new product releases may trigger early adoption, push up token prices, and attract participants. This forms a feedback loop of demand rise and economic rise. However, on the other hand, if ARC RL struggles to maintain user stickiness, demand may quickly disappear.

more players

7. The Mother of Game AI Models

What is the ending? The advantage of ARC is that it can promote various types of games. Over time, it allows them to collect a unique database of specific gameplay. As ARC integrates with more games, it can continuously feed this data back into its ecosystem, creating a rising and perfect virtuous cycle.

Once the cross-sectional game data set reaches critical mass, it will become a very valuable resource. Imagine using it to train a general AI model for game development - opening up new possibilities for large-scale game design, testing, and optimization.

It is still too early now, but in the era of artificial intelligence where data is the new oil, the potential in this area is limitless.

8. Our Ideas

(1) NRN evolves into platform game - Token repricing

With the issuance of ARC and ARC RL, the project is no longer just a single-product gaming company, positioning itself as a platform and AI game. This shift is expected to lead to a reevaluation of the NRN Token, which was previously limited to the success of AI Arena. The introduction of new Token sources through ARC RL, along with external demand for revenue sharing protocols with gaming companies and trainer transaction fees, creates a broader and more diverse foundation for the utility and value of NRN.

(2) Success is closely related to game partners.

ARC’s business model connects it successfully with the companies it collaborates with, as the revenue stream is based on Token distribution (in Web3 games) and game royalties payment. The closely integrated games are worth a look.

If the ARC game is a great success, the resulting value will flow back to the NRN holder. Conversely, if the cooperative game encounters difficulties, the value flow will be restricted.

(3) Looking forward to more integration with Web3 games

The ARC platform is very suitable for Web3 games. In Web3 games, competitive gameplay with incentive mechanisms perfectly combines with the existing Token economy.

By integrating ARC, Web3 games can immediately enter the “AI agent” narrative. ARC RL brings the community together, motivating them to move towards a common goal. This also opens up new opportunities for innovative mechanisms, such as making activities like “game to Airdrop” more appealing to players. By combining AI and Token incentives, ARC adds Depth and excitement that traditional games cannot replicate.

(4) The AI play has a learning curve

AI gameplay has a steep learning curve, which can create friction for new players. It took me an hour to figure out how to properly train my players in AI Arena.

However, the player experience friction is smaller in ARC RL, because AI training is done in the backend when players play the game and submit data. Another pending question is how players feel when they know their opponent is AI. Does it affect them? Will it enhance or weaken the gaming experience? Only time will tell us the answer.

9、Bright Future

AI will bring a whole new breakthrough experience in the gaming world.

Teams like Parallel Colony and Virtuals are pushing for the development of autonomous AI agents, while ARC is carving out its niche market by focusing on human behavior cloning—a innovative approach to address player Liquidity challenges without relying on unsustainable tokenomics.

The transition from a game to a mature platform is a huge leap for ARC. This not only opens up greater opportunities through cooperation with game companies, but also restructures the integration of AI and games.

With its improved tokenomics and the potential of strong network effects, the bright future of ARC seems to have just begun.