Decentralizing AI

Crypto + AI intersection.

A high-level summary of crypto+AI intersections by Vitalik. Source: The promise and challenges of crypto + AI applications

Decentralizing AI

Last week I had the opportunity to share my ideas with an interesting group of people in the tech ecosystem divided between founders and professional investors. The purpose of the conversation was to share ideas regarding the adoption of AI, the layers where product development will take place in LatAm, and the barriers to entry for founders who want a slice of the pie. 

The things that most resonated with me during this exciting conversation were: the cost of entry for early-stage startups regarding the processing capabilities, and the security protocols required to run data at scale with enterprise stakeholders. This means most of the innovations we’ll be seeing through the upcoming founders will be in the application layer. 

There is a reason behind this argument resonated with me. I am a strong believer in decentralization. Thinking about the possibilities that blockchain can enable by enhancing data privacy, data contribution incentives, storing, tracking, etc really excites me. 

If you are into Blockchain you are aware that the AI+Crypto intersection is one of the latest trends in the current bull market,  plus the ZK Snarks narrative you have a powerful stack to show of buzzwords to your potential investors behind that,  more importantly, you have an amazing framework to incentivize high-quality data in a private way for federated learning. 

This opportunity is unique. Through decentralization, innovators get the chance to collect data and access computing power to develop competitive products against the centralized AI models. 

What is Decentralized Machine Learning, Decentralized AI, or as Google defined eight years ago what is Federated Machine Learning?

In machine learning, we have a model, and we have data. The model could be a neural network, or something else, like a classical linear regression. We use this data to perform tasks, Now, in practice, the training data we work with doesn’t originate on the machine we train the model on. It gets created somewhere else. A smartphone, biometric sensor, synthetic data, or somewhere else. When we say somewhere else, most of the time is not just one place. It could be several devices all running the same app. But it could also be several organizations, all generating data for the same task. So to use machine learning or any kind of data analysis, the approach that has been used in the past was to collect all data on a central server. Once all the data is collected in one place, we can finally use machine learning algorithms to train our model on the data. 

Federated learning simply reverses this approach. It enables machine learning on distributed data by moving the training to the data, instead of moving the data to the training.  A simple explanation: 

• Central machine learning: move the data to the computation

• Federated (machine) learning: move the computation to the data

This model can be particularly powerful in cases where the data available on one server is not enough to train a good model. When we want to use distributed computing power to run our model.  It can also be better regarding privacy issues, low latency communication, and gathering random inputs to fine-tune a model.

Here is when building with distributed capabilities comes in handy.  We are witnessing the blossom of a handful of amazing projects enabled by blockchain, divided into four main categories:

• Distributed Compute power with projects like io.net and Akash

• Decentralized Model Training like Bittensor whose mission is to make building AI applications more accessible by creating a peer-to-peer marketplace to share and tap machine learning models. 

• zkML Zero-Knowledge Machine Learning,  zkML combines sophisticated cryptographic techniques with AI to ensure the integrity of machine learning processes and the accuracy of their output. It lets us check AI's work without needing to trust anyone, which is what crypto is all about. One nice project in this area is Modulus which verifies AI outputs with ZK as a service. 

• The last category is an interface that Web 2 is more familiar with AI Agents in Blockchain, where the use case is inverted. AI enables DeFI by performing automated tasks like MEV(Maximum Extractable Value) Arbitrage Bots, Predictive Analysis Bots,  NPC Agents for gaming, Telegram, and Discord Bots. 

There are many ways to think about this intersection, yet there are many challenges to come. Probably security being one of the biggest. In “The Promise and Challenges of Crypto+AI” Vitalik Bulterin remarks on the vulnerability when we are training at scale and cost of cryptography when it comes to this intersection. Here is when decentralization and distribution become the engine of change:

"AIs participating in on-chain micro-markets" work better: each individual AI is vulnerable to the same risks, but you're intentionally creating an open ecosystem of dozens of people constantly iterating and improving them on an ongoing basis. Furthermore, each individual AI is closed: the security of the system comes from the openness of the rules of the game, not the internal workings of each player.

So after an exciting week where I got to talk to professional investors about the AI trends in the region, I got to play with Llama 3.1 and heard from Mr. Orange intentions regarding the massive Bitcoin adoption by the US government if he happens to be elected.

My personal conclusion is AI is another change of paradigm, probably one stronger than Blockchain because when you talk about AI you are talking about the evolution of a machine, the value to capture here is probably larger than within a network or networks such as blockchain where the network needs to engage with users to develop a value for itself. With a machine evolution large hardware suppliers have been able to increase massively their market cap as we´ve seen within the well-known players in the last 24 months or so. Blockchain is not our time yet, but I hope that after reading this post I made myself clear about the opportunities blockchain brings to the table as an enabler for innovators. 

With these closing lines, I want to express that this is my first post trying to engage with my crypto fellows, a community I´ve been part of for a while and I always felt welcome. My intention is to explore what else is to come in this exciting and actually happening intersection. AI is about abundance, Crypto is about scarcity the way innovators balance this to develop is the way to go.