Author: Darshan Gandhi, Founder of FutureX Labs; Translation: Golden Finance xiaozou
In this article, let’s explore the world of decentralized artificial intelligence (DeAI). We will learn about the following:
The AI Development Lifecycle
The need for decentralized AI
Practical applications of DeAI
Growth Catalysts in the Crypto x AI Sector
A visionary at the forefront
What are the shortcomings of DeAI?
The Future of Crypto x AI
1. Introduction to AI
I think it is safe to say that artificial intelligence (AI) will change the world. Imagine a child learning about animals by looking at pictures, memorizing their names and features. Over time, the child's ability to identify animals improves.
Artificial intelligence works in a similar way, using data to learn and improve performance over time.
Many groundbreaking applications are being built using AI, including:
ChatGPT: Capable of human-like conversations.
Perplexity AI: Improving search accuracy.
Jasper AI: Your writing assistant.
DALL-E: Generating images from textual descriptions.
Pika Art: Create HD videos from text.
There will be more and more applications like this. These tools are becoming part of our daily lives, making work easier and more efficient. AI is not just a futuristic concept, it is actively solving major problems we face today.
The development of artificial intelligence is affecting and changing many industries, such as:
Helping doctors diagnose diseases faster
Making self-driving cars safe
Provide users with a personalized online shopping experience
In principle, AI methods can be divided into three main categories:
Centralized AI: Controlled by a single entity or company.
Decentralized AI: Focus on distributed control, transparency, and incentives.
Open source AI: Emphasis on promoting collaboration and transparency.
In this article, we will specifically discuss "decentralized artificial intelligence".
2. AI Development Lifecycle
Before we delve into the details, let’s understand the components that shape the AI development lifecycle. This will make it easier for us to understand the contribution of decentralization in each link.
Innovation in AI requires years of advancement, continuous feedback, training, and engagement.
Developing an AI model involves several key phases to ensure a strong end-to-end operational flow. Here is a detailed description of the key phases of the AI development lifecycle:
Problem statement, identification and design
It all starts with identifying the business problem and defining what you want to achieve.
Data collection is one of the most critical steps to ensure that the model uses accurate and relevant data.
Data Collection and Exploration
This phase involves aggregating data from different sources and assessing the quality of this data.
Initial data analysis helps in understanding patterns and trends to develop a plan for data preprocessing and feature engineering (data improvement).
Data collation and preparation
Data preprocessing cleans and transforms raw data into rich, usable datasets.
Use feature engineering to create new features from existing data to enhance the performance of your model.
Model development
This phase selects the most suitable machine learning practice based on the problem statement and the data collected.
The next step is to train and test the model to ensure that it can make accurate predictions.
The last step is optimization, which is to improve the efficiency of the model.
Model deployment
Deploying the model to a real-world environment allows the model to start making predictions, recommendations, or whatever training task it was trained on. To put it into production requires a computing power provider.
Continuous monitoring is performed to ensure that the model remains accurate and valid.
Bias detection to ensure fairness in decision making.
Model maintenance and retraining
Maintaining models requires regular updating and retraining with new data.
The focus is on collecting as much feedback as possible and sending it back to the model for adjustments and enhancements.
Today, most of these models come from research institutions, private companies, or some open source organizations. Companies such as Google, OpenAI, IBM, AWS, and Microsoft are among the main players.
Below is a market map of GenAI models by different players in various verticals.
Let’s take a quick look at how AI technology has evolved over the years.
3. The need for decentralized AI
Centralized AI has its own problems. Think about it: a single point of failure could jeopardize everything.
On the other hand, decentralized artificial intelligence (DeAI) changes the game by distributing data across multiple nodes, making the system more secure. If one node is attacked, the others will continue to operate normally. This setup also gives users more control over their data and reduces privacy risks, especially when using technologies such as fully homomorphic encryption (FHE) and zero-knowledge machine learning (ZKML).
Censorship is another big problem with centralized systems. A single entity can take control and manipulate information. Decentralized AI, on the other hand, spreads out control, making it difficult for any single entity to dominate the narrative. This ensures that information is accessible and less susceptible to undue influence.
In my opinion, transparency is a key factor. The open source model, incentive mechanism, and collaborative workflow management mean that anyone can check and verify decisions at any time. This level of openness addresses concerns about hidden biases and opaque processes in centralized systems. In addition, it allows more people to join and contribute. For example, people with idle computing space can now rent it out through decentralized computing power providers like Akash and Render.
The decentralized model also limits the power of central entities, preventing AI from being abused for unfair purposes. By promoting collaboration and knowledge sharing, it can leverage collective intelligence and greater governance, resulting in more reliable, open, and accurate systems.
Cryptocurrency acts as this enabler, combining the best of both worlds. It provides access to top-tier services, computation, models, and data, while also providing incentive loops, security, and privacy protections for all stakeholders. This synergy ensures that DeAI is not only effective, but also fair and secure.
4. Practical application of DeAI
Here are some of the main applications in the field of DeAI:
(1) By field
medical insurance
DeAI improves healthcare by enabling secure and private data sharing among medical institutions.
AI algorithms can analyze anonymous data to identify patterns, predict disease outbreaks, and develop personalized treatment plans. For example, patients can privately share their data with hospitals and ensure that only they own the data.
finance
Decentralized Finance (DeFi) is one of the largest sub-ecosystems of web3. Artificial intelligence can help enhance risk management and trading.
These protocols use artificial intelligence to assess risk, predict asset prices, and optimize trading strategies. For example, many projects are developing tools for efficient asset management, AI-driven automated market makers (AMMs), and more.
Security and fraud detection
AI algorithms can help systems detect and prevent fraud by analyzing transaction data for patterns and anomalies.
This increases the security of the web3 protocol. For example, in the NFT ecosystem, AI can help identify fake assets and ensure integrity.
Content/emotion generation
AI can be used to create story outlines, plots, game mechanics, and more.
For example, web3 games can use AI to generate game content from text descriptions and use smart contracts to manage ownership of assets such as characters and props.
Additionally, understanding how users feel about a category, problem, or market is invaluable. Tools like Kaito and Nansen aim to provide this capability.
AI Agents and Automation
There are projects building autonomous AI agents for tasks ranging from customer service to supply chain management.
These agents can be created by anyone or collaboratively, and all stakeholders can automatically and seamlessly receive rewards.
user experience
The user experience of Web3 is not the best, but the model can help enhance the user experience through personalized recommendations and behavior prediction.
A good example is decentralized social networks that allow users to choose content recommendation algorithms or curate their feeds based on their preferences.
(2) Classification by degree of ecosystem management
Incentives
Stakeholders can be rewarded (earn tokens) by providing data, computing power or developing algorithms
There is a strong desire for people to do this, to collaborate on difficult problems, and to be reasonably rewarded for their time and effort.
Cost efficiency
DeAI platforms can help significantly reduce costs by utilizing unused resources in distributed networks. They eliminate the need for expensive data centers and ensure that resources are utilized to the maximum extent possible.
For example, projects such as Akash Network, Aethir, and Render allow users to rent out unused computing power for AI tasks, thereby improving efficiency.
Governance
DeAI can also be used to improve governance processes, especially for protocols and DAOs.
AI can automate reputation management and rewards, for example, ensuring that contributions are treated fairly in a DAO.
5. Growth catalysts in the Crypto x AI space
There are some powerful catalysts driving the convergence of Crypto and AI. Let’s look at a few of them.
First, financing in the ecosystem has been increasing. In the past year, there have been 136 rounds of financing of US$1.02 billion, with an average of US$7.5 million per round. Well-known investment companies such as Hack VC, Variant, Paradigm, and Polychain have been making large-scale investments. The influx of capital has accelerated research and innovation in the field.
Secondly, the technology aims to provide a cost-effective alternative to centralized systems. It can reduce potential operating costs by nearly 50% and efficiently handle large data volumes while also providing security and privacy protection. For example, compared to AWS, GCP, and Azure, Akash claims to offer an 85% discount on computing power provisioning.
Third, by market capitalization, leading projects in the field, such as Bittensor, Akash, Render, and Worldcoin, have performed exceptionally well in the secondary market over the past year. These projects are the best performing assets in web3. According to Coinbase's report, the Crypto x AI category also performed well among all categories.
Fourth, NVIDIA's performance in April this year was very good. Let's look at some numbers from the news reports:
Their first quarter 2024 revenues are $26 billion, up 18% from the fourth quarter of 2023 and up 262% from the same period last year.
For the first quarter of this year, GAAP diluted earnings per share were $5.98, up 21% sequentially and 629% year-over-year.
Fifth, all centralized services, including Google.com, Chatgpt, and Perplexity, have recently gone down together, while all web3 services are intact and running well. The founder of Akash Network posted the following tweets before and after the incident.
Thanks to these and many other similar initiatives, events, and innovations, the field is growing rapidly.
6. Visionaries at the forefront
The ecosystem is gathering increasing momentum thanks to the support and involvement of some key industry figures.
Erik vooroorhees
Erik Voorhees, founder of ShapeShift and a hugely influential Twitter personality, launched Venice AI to create a permissionless alternative to popular web2 LLMs like ChatGPT.
Venice focuses on user privacy and censorship resistance, using open source technology to provide uncensored and unbiased information.
Mother Mostaque
Founder and former CEO of Stability AI, now resigned to focus on the DeAI field - currently developing Schelling AI.
He believes that as artificial intelligence becomes more and more important, transparent and distributed governance will become extremely important.
Neeraj Pant
Former partner at Polychain Capital, currently developing Ritual.net.
The platform aims to build a sovereign execution layer for AI, enabling the open source, permissionless creation, distribution, and improvement of AI models.
The first phase of Ritual.net (Infernet) allows developers to access models on and off the chain through smart contracts.
7. Disadvantages
While decentralized AI has many benefits, it also encounters significant challenges that deserve attention. Here are the key issues it currently faces:
Initial setup costs and challenges
There are considerable difficulties in building a DeAI network. It takes a lot of time and resources to build the necessary infrastructure and attract participants. This cold start problem highlights the need for strong incentives to attract early adopters. However, without reaching a large enough scale, it will be difficult for the network to gain traction.
Increased coordination needs
Managing a decentralized network is complex. It takes a lot of work to synchronize multiple nodes and stakeholders, ensure data consistency, maintain network security, and run the network cost-effectively. While this coordination embodies the essence of Crypto x AI, it can sometimes become inefficient and cumbersome.
Scaling Challenges
The network faces a scaling problem. Handling the ever-increasing amount of data and transactions without degrading performance is a major challenge. Due to varying node uptime, decentralized networks can experience latency and bandwidth issues, affecting overall efficiency. Solutions like sharding are still under development and may not fully alleviate these issues.
Resource Access
Enterprises often face obstacles in acquiring cutting-edge resources. Major centralized vendors can invest heavily in the latest hardware and software, giving them a competitive advantage. DeAI projects, however, are constrained by limited funding and may fall behind, affecting their performance and capabilities. For example, NVIDIA tends to prioritize resources to hyperscale servers such as GCP, Azure, and AWS due to higher demand. However, for web3 vendors, the current supply is greater than the demand, or they may still be in the early stages of development.
Regulatory and legal challenges
Crypto largely operates in a regulatory grey area. The lack of a clear regulatory framework can create legal risks and uncertainty. In a decentralized environment, compliance with regulations such as GDPR becomes more challenging, exacerbating an ongoing global struggle.
8. The future of Crypto x AI
The convergence of crypto and AI promises to foster the development of innovative projects and applications that address real-world challenges.
In our later articles, we will delve deeper into several key subcategories in the crypto space. We will explore zero-knowledge machine learning (zkML) through projects such as Modulus Labs and Giza, which are developing products centered around model inference. In addition, we will examine decentralized cloud computing vendors such as Render, Akash Network, and Aethir, highlighting their role in providing scalable and cost-effective alternatives to traditional cloud services.
Bittensor: This project is developing a decentralized network that incentivizes participants to share AI models and datasets through a blockchain and uses “subnets” to reward contributions.
Fetch: Fetch focuses on the autonomous AI agent market and provides integration with top services such as ChatGPT and Slack, facilitating alignment through simple API integration.
Akash Network: Focused on building a decentralized marketplace for cloud computing resources, Akash Network utilizes its AKT token for governance, security, and intra-network transactions.
9. Conclusion
I firmly believe that decentralized artificial intelligence (DeAI) will be a game changer and we are only beginning to see its development in the ecosystem.
DeAI embodies the principles of transparency, collaboration, and global impact. As we have discussed, it is reshaping various key sectors.
Projects like Render, Akash, and Worldcoin, with their impressive traction and funding, not only highlight the huge potential of the space, but also indicate that it could experience substantial growth in the coming years.
Going forward, we will delve deeper into the various subcategories of Crypto x AI and continue to explore this dynamic vertical.
The future is bright, and we are just getting started.
