Harnessing the Intersection of Cryptocurrency and Artificial Intelligence
This article is based on the research article “AI <> Crypto Projects That Aren’t Complete Bullsh*t” written by a former Bankless researcher and compiled, translated, and written by Techflow.
Summary:
io.net’s true GPU count remains a mystery. Is there something fishy? The decentralized AI computing network faces four major challenges to solve.
Background:
Google’s push for OpenAI: Who will come out on top, Gemini Live or GPT-4o?
Table of Contents:
A High-Level View of AI Stacks
Open Source Empowered by Cryptocurrency
Decentralized Physical Infrastructure Networks (DePINs)
The Case of Data Networks: Grass
The Case of GPU Networks: io.net
The Use of Incentive Structures
AI Building Networks: Exploring Bittensor
Remembering Incentive Measures
Intelligent Agents: Exploring Morpheus
How to Distinguish a Completely Useless Project
Type 1 – Cryptocurrency Assisting AI
Type 2 – AI Assisting Cryptocurrency
Where Are We Headed?
When searching for new alpha information, we inevitably come across some garbage information. When a project can quickly raise five to six figures with just a semi-clear introduction and a decent brand, speculators jump on every new narrative. With the traditional financial sector joining the AI trend, the “crypto AI” narrative exacerbates this problem.
The majority of these projects have the following issues:
– Most crypto projects do not need AI.
– Most AI projects do not need cryptocurrencies.
Not all decentralized exchanges (DEX) need built-in AI assistants, and not every chatbot needs a companion token to facilitate its adoption curve. The hard coupling of AI and cryptocurrency technology almost drove me to collapse when I first delved into this narrative.
The bad news is what? Continuing down the current path, further centralizing this technology will only end in failure, and a plethora of false “AI x Crypto” projects will hinder our ability to reverse the trend.
The good news is what? There is light at the end of the tunnel. Sometimes, AI can indeed benefit from the crypto economy. Similarly, AI can solve practical problems in some use cases of cryptocurrencies.
In this article, we will explore these critical intersections. The convergence of these niche innovative ideas creates a whole that is greater than the sum of its parts.
Here is my view on different verticals in the “crypto AI” ecosystem (for a deeper understanding, refer to Tommy’s article). Note that this is a highly simplified view but aims to help lay the foundation.
At a high level, here’s how it works:
1. Massive data collection.
2. Processing the data to make machines understand how to ingest and apply it.
3. Training models on this data to create a generalized model.
4. Fine-tuning these models to handle specific use cases.
5. Deploying and hosting these models for applications to query and provide useful implementations.
All of this requires significant computational resources, which can be executed locally or obtained from the cloud.
Let’s explore each of these areas, with a special focus on how different crypto-economic designs can actually improve standard workflows.
The debate between “closed-source” and “open-source” development methods can be traced back to the Windows-Linux debate and Eric Raymond’s famous “The Cathedral and the Bazaar” theory. Although Linux is widely used among enthusiasts today, approximately 90% of users choose Windows. Why? Because of the incentives.
At least from an external perspective, open-source development has many benefits. It allows the most people to participate in the development process and contribute to it. But in this headless structure, there is no unified command. CEOs do not actively make their products available to as many people as possible to maximize their bottom line. In the open-source development process, projects can become a “hodgepodge” that diverges at every intersection of design ideas.
What is the best way to adjust the incentives? Build a system that rewards behaviors that promote goal achievement. In other words, put money in the hands of actors who can bring us closer to the goal. With cryptocurrencies, this can be hardcoded into law.
Let’s look at some projects that are doing this.
“Oh please, not this again?” Yes, I know the DePINs narrative has been talked about as much as AI itself, but please bear with me for a moment. I am willing to believe that DePINs is a genuinely transformative use case for cryptocurrencies that has the potential to change the world. Think about it.
What are cryptocurrencies really good at? Removing intermediaries and incentivizing activities.
The original vision of Bitcoin was peer-to-peer currency designed to exclude banks. Similarly, modern DePINs aim to exclude centralized powers and introduce provably fair market dynamics. As we will see, this architecture is ideal for crowdsourced AI-related networks.
DePINs use early token issuance to increase the supply side (providers), hoping to attract sustainable consumer demand. This aims to solve the cold start problem in new markets.
This means early hardware/software (“nodes”) providers earn a large number of tokens and a small amount of cash. As users leverage these nodes (in our example, machine learning builders) bring in cash flow, it begins to offset the decreasing token issuance over time until a fully self-sustaining ecosystem is established (which may take several years). Early adopters like Helium and Hivemapper have demonstrated the effectiveness of this design.
GPT-3 is claimed to have been trained on 45TB of pure text data, equivalent to about 90 million novels (but it still cannot draw a circle). GPT-4 and GPT-5 require even more data than what is currently available on the surface web, making the term “data-hungry” an understatement of the century.
If you’re not a top player (OpenAI, Microsoft, Google, Facebook), acquiring this data is extremely challenging. The common strategy for most people is web scraping, which works well until you try to scrape a large number of websites using an Amazon Web Services (AWS) instance, and you quickly run into rate limits. This is where Grass comes in.
Grass connects over two million devices and organizes them to scrape websites from users’ IP addresses, collect, structure, and sell them to AI companies in need of data. In return, users participating in the Grass network can earn a steady income from AI companies using their data.
Currently, there is no token, but future $GRASS tokens may incentivize users to download their browser extension (or mobile app). They have already attracted a large number of users through an extremely successful referral campaign.
Perhaps even more important than data is computing power. Did you know that in 2020 and 2021, China invested more money in GPUs than in oil? That’s just insane, but it’s just the beginning. Goodbye petrocoin, make way for computecoin.
There are now many GPU DePINs on the market, and their operation is roughly as follows:
– Machine learning engineers/companies in need of computation.
– On the other hand, data centers, idle mining machines, and amateur enthusiasts with idle GPUs/CPUs.
Despite the enormous global supply, coordination is lacking. Contacting 10 different data centers to get bids for your usage is not easy. A centralized solution would create a rent-seeking intermediary whose incentive is to extract maximum value from each party, but cryptocurrency technology can help.
Cryptocurrency technology is excellent at building market layers that efficiently connect buyers and sellers. A code snippet doesn’t need to be accountable to shareholders’ financial interests.
io.net stands out because it introduces some cool new technology crucial for AI training—its cluster stacks. Traditional clusters involve physically connecting a bunch of GPUs in the same data center to enable them to work together for model training. But what if your hardware is distributed across different locations? io.net, in collaboration with Ray (used for building ChatGPT), developed cluster middleware that can connect GPUs in different locations.
Moreover, the registration process with AWS can take days, while clusters on io.net can be launched without permission in 90 seconds. For these reasons, I can see io.net becoming the hub for all other GPU DePINs, where they can plug in their “IO engines” to unlock built-in clusters and provide a smooth onboarding experience. All of this is only possible with the help of cryptocurrency technology.
You will notice that most ambitious decentralized AI projects (such as Bittensor, Morpheus, Gensyn, Ritual, Sahara) have explicit “computational” requirements, which is where GPU DePINs should come in. Decentralized AI needs permissionless computation.
Returning to the insight from Bitcoin. What if miners were building AI instead of solving useless math problems? That’s what you get with Bittensor.
Bittensor aims to create AI built by Bitcoin miners instead of solving useless mathematical problems. Miners and stakers in Bitcoin and Ethereum absorb all of their native tokens because that’s what the protocols incentivize—participants become miners and stakers.
In an organization, this may come from the CEO, who defines the “vision” or “mission statement.” But humans are prone to errors and may lead the company astray. On the other hand, computer code can maintain focus better than the most dedicated wage slave. Let’s take a look at a few decentralized projects whose built-in token effects keep participants focused on noble goals.
What if we had Bitcoin miners building AI instead of solving meaningless math problems? With Bittensor, you get exactly that.
Bittensor’s goal is to create AI built by Bitcoin miners instead of solving useless mathematical problems. Miners and stakers in Bitcoin and Ethereum absorb all of their native tokens because that’s what the protocols incentivize—participants become miners and stakers.
Establish several experimental ecosystems for testing the production of “commodified intelligence” within each ecosystem. This means that one ecosystem (called a subnet, or SN for short) may focus on developing language models, while another may focus on financial models, and others may focus on speech synthesis, AI detection, or image generation (see current active projects).
For the Bittensor network, it doesn’t matter what you want to do. As long as you can prove that your project is worth funding, the incentives will flow. This is the goal of subnet owners, who register subnets and adjust the rules of the game.
The participants in this “game” are called miners. These are the ML/AI engineers and teams building models. They are locked in a continuous review “thunder dome” and compete with each other to earn the most rewards.
Validators are another aspect and are responsible for reviewing and scoring the work of miners. If collusion is detected between validators and miners, they will be expelled.
Miners earn more by outperforming other miners within the subnet, which drives the development of AI.
Validators earn more by accurately identifying high-performing and low-performing miners, which maintains the fairness of the subnet.
Subnet owners earn more when the AI models generated in their subnet are more useful than those in other subnets, which drives subnet owners to optimize their “game”.
You can think of Bittensor as a permanent reward machine for AI development. Emerging ML engineers can try to build something, pitch it to VCs, and try to raise some funds. Or they can join as miners in one of the Bittensor subnets, showcase their skills, and earn a large amount of TAO. Which is easier?
Several top teams are building on the network:
– Nous Research is the king of open source. Their subnet disrupts the traditional fine-tuning of open-source LLMs. They ensure that the leaderboard cannot be manipulated through continuous synthetic data flow testing (unlike traditional benchmark tests like HuggingFace).
– Taoshi is essentially an open-source quantitative trading company. They ask ML contributors to build trading algorithms that predict asset price trends. Their API provides quantitative-level trading signals for retail and institutional users and is moving towards significant profitability.
– Cortex.t, developed by the Corcel team, has two purposes. First, they incentivize miners to provide API access to top models like GPT-4 and Claude-3 to ensure continuous availability for developers. They also provide synthetic data generation, which is useful for model training and benchmark testing (this is also why Nous uses it). Check out their tools: Chat and Search.
Without a doubt, Bittensor reaffirms the power of incentive structures, all achieved through cryptographic economics.
Now let’s look at the two aspects of Morpheus:
1. Cryptography helps AI (cryptography empowers AI)
2. AI enables applications in cryptography (AI empowers cryptography)
“Intelligent agents” are AI models trained by smart contracts. They understand the internal workings of top DeFi protocols, know where to find profits, where to bridge, and how to identify suspicious contracts. They are the future “automated routers” and, in my opinion, they will be the way everyone interacts with blockchain in 5-10 years. In fact, once we reach that point, you may not even know you are using cryptographic technology. You will just tell the chatbot that you want to move some savings into another investment, and everything will happen in the background.
Morpheus embodies the “incentivize them, and they will come” information. Their goal is to have a platform where intelligent agents can propagate and thrive, with each agent building on the success of the previous one in a minimally externalized ecosystem.
The token inflation structure highlights the four main contributors to the protocol:
1. Code – the builders of agents.
2. Community – building front-end applications and tools to attract new users to the ecosystem.
3. Computation – providing computational power to execute agents.
4. Capital – providing their earnings to drive Morpheus’ economic machine.
Each of these categories receives an equal share of the $MOR inflation reward (with a small portion stored as an emergency fund), forcing them to:
1. Build the best agents – creators are rewarded when their agents are consistently used. Unlike offering OpenAI plugins for free, this method pays creators instantly.
2. Build the best front-end/tools – creators are rewarded when their creations are consistently used.
3. Provide stable computational power – providers are rewarded for lending their computational power.
4. Provide liquidity for the project – earn their share of MOR by maintaining liquidity for the project.
Although there are many other AI/intelligent agent projects, Morpheus’ token economic structure is particularly clear and effective in designing incentive mechanisms.
These intelligent agents are the ultimate example of how AI can eliminate barriers in cryptographic applications. The user experience of dApps is notoriously poor (despite significant progress in recent years), and the rise of LLMs ignites the passion of anyone who wants to be a founder in Web2 and Web3. While there are many profit-driven projects, outstanding projects like Morpheus and Wayfinder (see demo below) demonstrate how simple on-chain transactions can become.
(See related tweets)
Combining all of this, the interactions between these systems may look something like this, although this is an extremely simplified view.
Remember our two broad categories of “crypto x AI”:
1. Cryptography empowers AI
2. AI empowers cryptography
In this article, we mainly focused on the first category. As we can see, a well-designed token system can lay the foundation for the success of the entire ecosystem.
The DePIN framework can help kickstart the market, and creative token incentive structures can coordinate open-source projects towards previously unattainable goals. Yes, there are several other legitimate crossroads that I haven’t covered due to length constraints:
– Decentralized storage
– Trusted execution environments (TEE)
– Real-time data access (RAG)
– Zero-knowledge x machine learning for inference/source verification
When determining whether a new project is truly valuable, ask yourself:
– If it’s a derivative of another mature project, are its differences enough to make it stand out?
– Is it just a packaged version of open-source software?
– Does this project truly benefit from cryptographic technology, or is cryptographic technology being forced into it?
– Do we really need 100 crypto projects similar to HuggingFace, a popular open-source machine learning platform?
In this category, I personally see more scams, but there are indeed some cool use cases. For example, AI models can eliminate barriers in the user experience of cryptographic applications, especially intelligent agents. Here are some interesting categories worth noting in the field of AI-supported crypto applications:
– Enhanced intent systems – automating cross-chain operations
– Wallet infrastructure
– Real-time alert infrastructure for users and applications
If it’s just a “chatbot with tokens,” it’s a junk project in my opinion. Please stop hyping these projects to maintain my sanity. Also, adding the “AI” label to your project doesn’t make it interesting.
Despite the noise, some serious teams are working hard to realize the vision of “decentralized AI”. In addition to incentivizing open-source model development projects, decentralized data networks open new doors for emerging AI developers. While most of OpenAI’s competitors struggle to reach large-scale transactions like Reddit, Tumblr, or WordPress, decentralized data fetching can level the playing field.
A company’s computing power will never exceed the total computing power of all other companies in the world. But with a decentralized GPU network, anyone has the ability to compete with top companies. All you need is a crypto wallet.
We are at a crossroads today. If we focus on truly valuable “crypto x AI” projects, we have the ability to decentralize the entire AI stack.
The vision of cryptocurrencies is to create an interference-free hard currency through the power of cryptography. Just as this emerging technology begins to gain popularity, a more formidable challenger emerges.
In the best-case scenario, centralized AI will not only control your finances but also impose biases on every piece of data we encounter in our daily lives. It will enrich a very small number of tech leaders in a self-perpetuating loop of data collection, fine-tuning, and model injection.
It will know you better than you know yourself. It knows which buttons make you laugh, angry, and consume more. And despite appearances, it is not accountable to you.
Initially, cryptographic technology was seen as a force to counter the centralization of AI. Cryptographic technology can coordinate the efforts of decentralized entities to achieve a common goal. However, now this ability is facing a more powerful enemy than central banks: centralized AI. This time, the urgency is high, and we need to act quickly to resist the centralization trend of AI.
Related Reports
– What is MakerDAO’s new stablecoin PureDai? No longer pegged to the US dollar, introduces floating price, and has a new governance token…
– Vitalik proposes new Ethereum Improvement Proposal “EIP-7706” for multi-dimensional gas concept to unlock L2 scalability
– Analysis: Why is Vitalik concerned about the development of Rollups? How should Layer2 proceed?