Author: SenseAI

AI Arena is an AI-driven Web3 competitive game that allows users to train their own AI characters to fight. The outcome of each battle depends on the player's skills in training. It aims to help users understand the operation and learning process of artificial intelligence. AI Arena is currently open for pre-registration, and ArenaX Labs plans to launch a beta version of the game on the Arbitrum mainnet soon.

AI Arena developer ArenaX Labs announced that it has completed a new round of financing of US$6 million, led by Framework Ventures, with participation from SevenX Ventures, FunPlus/Xterio and Moore Strategic Ventures. It plans to use the funds to build a PvP fighting platform and develop similar games.

Sense

We try to propose more divergent deductions and reflections based on the content of the article, and welcome exchanges.

- AI Arena is not only a game that combines AI, but also a platform for cultivating players' AI capabilities. With the advent of the AI ​​era, how to train an AI assistant that suits you has become an essential skill in people's work and life, and an important indicator for measuring employees' work ability in the workplace.

- The combination of AI and games allows players to improve certain soft skills while having fun and relaxing. AI Arena has made bold attempts in this regard and has found a suitable entry point. In the future, as more and more players master the ability to train AI assistants, AI Arena can also provide an AI bilateral trading market based on protecting the intellectual property rights of AI practitioners and match transactions between buyers and sellers.

This article has 2334 words in total, and it takes about 8 minutes to read carefully.

AI Native Product Analysis

AI Arena

1. Product: AI Arena

2. Founder: AI Arena was developed by its parent company ArenaX Labs, which was co-founded in 2018 by three founders (Brandon Da Silva, Dylan Pereira, and Wei Xie) and is dedicated to producing independent games.

3. Product Introduction:

AI Arena is an Ethereum native game where players around the world can buy, train, and fight NFT characters driven by artificial intelligence. It is an NFT tokenization platform powered by real AI. In the game, players design and train AI-driven NFT fighting characters in a global PVP arena competition, and let these characters fight automatically, with the ultimate goal of knocking opponents off the platform. Players help AI characters improve through a process called "imitation learning", where they learn skills by observing human behavior. In turn, players can evaluate the capabilities of AI through "AI Inspector" and point out its weaknesses as key training areas for future improvements.

4. Development Story:

- In October 2021, it completed a $5 million seed round of financing, led by Paradigm and participated by Framework Ventures;

- In January 2024, it completed a new round of financing of US$6 million, led by Framework Ventures, with participation from SevenX Ventures, FunPlus/Xterio and Moore Strategic Ventures.

01.AI Arena Product Vision

Brandon Da Silva is the CEO of ArenaX Labs, the parent company of AI Arena. Before founding AI Arena, he worked for 5 years in OPTrust, the largest pension fund in Canada, which invested and managed AI. Integrating machine learning into investment analysis is the main theme of his career. Brandon once explained on his Twitter why he decided to do AI Arena - lowering the threshold of the AI ​​industry so that all AI enthusiasts are no longer restricted by academic qualifications and have a platform to demonstrate their abilities; using NFT to carry AI models to realize a technician's dream of fully owning the proceeds of his own labor; attracting everyone to contact AI in a more interesting way, and stimulating enthusiasm for learning AI during the game. These three goals constitute the value flywheel of AI Arena. In the long run, AI Arena will create a two-sided market for AI based on the game platform, aiming to protect the intellectual property rights of AI practitioners, help them monetize, and match the needs of buyers and sellers.

02. How does AI Arena combine with AI?

Although AI Arena is a fighting game, similar to games such as Super Smash Bros. Brawl and Street Fighter, AI Arena is also a project involving multiple fields: AI/ML, encryption, games and NFT. It is similar to other fighting games. An important difference in the game is that the player does not have control over the "fighters" he or she owns.

So how does it fight?

The boxers are powered by AI, which tells it what moves to make in certain situations; each boxer has a different AI, so whether you can train your boxer to become a boxing champion is entirely up to the player.

You can think of this game as you coaching a boxer as he prepares for a fight. You can level up by configuring his training regimen or fighting him in real combat so that he learns to copy your moves.

Why do we need neural networks?

Simply put, neural network means that in theory it can learn the mapping of any user action. In order to enable boxers to use neural network to learn strategies, AI Arena will adopt simulation learning and reinforcement learning, in which the neural network architecture is stored on IPFS (InterPlanetary File System).

The connections between neurons are called "weights". When your neural network is "learning", what is happening is that it is changing the value of the weights. The weights will ultimately determine how states map to actions, which means we can interpret weights as "intelligence". Neural network weights are unique to each NFT and are stored on Ethereum.

Training a boxer is the process of changing the weights in a neural network to enable the AI ​​to function. For example, if we were in front of an opponent, we might want our boxer to proactively attack. There are a range of weights that make this happen, and the point of training is to get the AI ​​to learn to take specific actions in specific scenarios.

AI Arena embeds the following training programs into the application:

(1) Imitation Learning

The best way to understand imitation learning is to imagine that you are a master and your AI is a boxer that you are preparing for a fight. You fight with your AI and it learns to imitate your moves in a specific scenario.

With the actual demo, you can test some moves and see how the AI ​​imitates you. Please note: it will not copy your moves immediately because the neural network takes a little time to learn, so you may need to repeat your moves a few times before the AI ​​learns.

(2) Self-learning

The perfect boxing partner is the user himself. Through self-learning, your AI is always challenging itself and constantly improving. In self-learning, it doesn't make much sense for the AI ​​to learn and fight like an opponent, because the opponent is a clone of the AI ​​itself. But if there is no expert to show the AI ​​how to fight, then how does it learn what to do? - Through rewards. The AI ​​will learn to take actions that give it more positive rewards and take fewer actions that give it negative rewards.

Of course, AI Arena has repeatedly emphasized that it cares about providing equal opportunities for everyone - the team hopes that rewards will be given more to users who persist in training AI rather than rewarding users with more resources.

03. A brief analysis of the innovative path of combining games and AI

In the currently popular artificial general intelligence AGI (Artificial General Intelligence) technology, Large Language Model (LLM) is the absolute protagonist. As more and more teams invest in the development of LLM-driven artificial intelligence agent (AI-Agents) systems, it becomes possible for AI Agents to redefine the innovation path of Web3 games. For example, the game "The Sims" uses LLM technology to generate 25 virtual characters, each of which is controlled by an LLM-supported agent and lives and interacts in a sandbox environment.

Generative Agents is cleverly designed, combining LLM with memory, planning, and reflection capabilities, which allows Agent programs to make decisions based on previous experience and interact with other Agents. This game shows people the capabilities of AI Agents, such as generating new social behaviors, information dissemination, relationship memory (such as two virtual characters continuing to discuss topics), and coordination of social activities (such as hosting a party and inviting other virtual characters). In short, AI-Agent is a very interesting tool, and its application in games is also worth exploring in depth.

Although there have been many different attempts to apply AI in the field of Web3 games, it is currently recognized that the most mature application in the Web3 game track is NFT Agent, and NFT will definitely be an important part of Web3 games in the future. With the development of metadata management technology in the Ethereum ecosystem, programmable dynamic NFTs have emerged. For NFT creators, they can make NFT functions more flexible through algorithms. For users, there can be more interactions between users and NFTs, and the generated interaction data has become a source of information. AI Agent can optimize the interaction process and expand the application scenarios of interaction data, injecting more innovation and value into the NFT ecosystem.

AI Arena mentioned above is the world's first battle game that combines AI and NFT. Users can use the LLM model to continuously train their own battle elves (NFT), and then send the trained battle elves to the PvP/PvE battlefield. The battle mode is similar to Nintendo Star Smash, but AI training adds more competitive fun.

In short, the combination of games and AI can not only solve the problem that Web3 games sacrifice user experience for security and decentralization, but it is also most likely to become the first field in the application scenarios of AI to expand the user base.